Supercharging Sales with Predictive Conversion Scoring
LeadSquared | Last Updated on 15th April, 2024
21%↑
Avg. Conversion Rate
18%↓
Avg. Conversion Time
4.1/5
Feature Rating
Supercharging Sales with Predictive Conversion Scoring
LeadSquared | Last Updated on 15th April, 2024
21%↑
Avg. Conversion Rate
18%↓
Avg. Conversion Time
4.1/5
Feature Rating
Supercharging Sales with Predictive Conversion Scoring
LeadSquared | Last Updated on 15th April, 2024
21%↑
Avg. Conversion Rate
18%↓
Avg. Conversion Time
4.1/5
Feature Rating
Supercharging Sales with Predictive Conversion Scoring
LeadSquared | Last Updated on 15th April, 2024
21%↑
Avg. Conversion Rate
18%↓
Avg. Conversion Time
4.1/5
Feature Rating
Supercharging Sales with Predictive Conversion Scoring
LeadSquared | Last Updated on 15th April, 2024
21%↑
Avg. Conversion Rate
18%↓
Avg. Conversion Time
4.1/5
Feature Rating
Project Details
Team
(most actively involved)
Product Manager, Lead Product Designer, Assoc. UX Designer, Sr. Software Engineer, Data Science Team, Sales Team, Professional Services Team
Duration
(from V1.0 to Dev handoff)
2 Months
My Role
(key contributions)
Qualitative Interviews
Competitor Analysis
Survey Analysis
UI Build
Design QA
Usability Testing
Overview
In high velocity sales, predictive conversion scoring is like having a superpower for your sales teams. This is fairly new in the LeadSquared CRM domain and is quickly transforming how businesses approach sales.
So, what's predictive conversion scoring all about? Well, it's like having a crystal ball that helps you figure out which leads are worth chasing and which ones might need a little more nurturing. It considers who your leads are, how they've interacted with your business before, and what they're doing now to predict who's most likely to become a customer.
By using this, you can focus your efforts on the leads that are most likely to convert, saving time and boosting efficiency. Now that you have a smart score assigned to your open leads, what's the 'next best action' to take on that lead? This case study dives into how conversion scoring and smart insights drive impact in a business and how their dynamic nature makes sales feel more human and approachable.
Jargons
Before we delve into how all of this works, let's take a look at some common terminologies:
Lead Scoring: It's a scoring system for your leads to quantify and rank active prospects in your system. The account admin sets up rules in the Settings>Lead Scoring page. For instance, if a lead opens one of your marketing emails and clicks on a link, they get 10 points. The admin decides on these rules and how much each action is worth. Then, a consolidated score is calculated for each lead, which we call the 'Lead Score.'
Lead Attributes: They are basically lead properties against which you record different aspects of the lead. For example, lead name, email id, phone number, address, etc.
Predictive Conversion Scoring: It's a data driven approach of predicting which lead is most likely to convert to a customer by analysing historical data and patterns, utilising advanced algorithms and machine learning techniques. Similar, to lead scoring, a consolidated conversion score is displayed to aid the sales user.
Data Model: It refers to a mathematical framework that captures patterns and relationships within a dataset. A data model is trained using historical data and optimised parameters to boost prediction efficiency.
Insights: Insights are generated based on the patterns the data model recognises. For example, personalised product recommendations for customers, forecasting future sales trends, and identifying areas for sales process optimisation are all insights generated by the data model we train.
Next Best Action: The data model will predict the next best possible action a sales user can take to push the lead through the sales funnel. It will be trained on historical sales activities performed on previous leads with similar attributes to recognise which activity helped nurture the lead faster.
How does Conversion Scoring work?
To kick off, you train a machine learning model on your historical lead data by defining an appropriate time range, the lead stages you consider to be 'won/converted' and 'lost/disqualified' and choosing lead attributes that you want the model to take into account.
The model analyses the data set and assigns a score to each lead based on the lead attributes that were calculated to be the most significant for conversion. You can also choose to retrain the model every 30 days to never miss out on emerging trends and dynamics in your business.
The score helps sales teams prioritise leads and reduce the time it takes to qualify a lead.
Train a Data Model
Step 1
Select the object type for which you would like to train the model.
Select duration, lead stages, lead fields , etc to be considered for training.
Configure Scoring
Step 2
Once training is completed, you can check the report to verify which lead attributes were deemed significant.
In case of any discrepancy, you can retrain the model.
Finish Setup
Step 3
Configure the score field in your lead detail pages, etc to see the it in action.
'Smart Insights' will be displayed alongside the conversion score.
What is 'Next Best Action'?
Once you have the score and insights for an open lead, as a sales user you might find yourself asking, "Alright, the score suggests this is a hot lead, but what now? What's the smartest move I can make to close this deal?"
When you enable this feature in your account, a new machine learning model is trained on historical sales activity data like past interactions, purchase history, and behavioural patterns to provide tailored recommendations to sales users.
You can choose the 'Activity type' you want the model to learn from, level of predictions required, lead stage and a few other key factors which together will define how likely is it to convert a lead and the number of steps it might take to do so.
Train a Data Model
Step 1
Select the activity type for which you would like to train the model.
Select duration, lead stage, activity fields , etc to be considered for training.
Configure NBA
Step 2
Once training is completed, you can configure this field in your lead detail page, etc.
You can choose to retrain the model in set intervals to increase accuracy.
Provide Feedback
Step 3
The next best action will be suggested for each lead in the lead details page.
You can provide feedback, objective+subjective for overall improvements.
Discovery Phase
As with any other feature request, we had to validate the ask, define scope of the build and create a mini roadmap to keep us aligned with the timelines. We employed the following methods to gather user insights like key pain points, expectations and specific user stories we can immediately address. The intent was to distill the necessary capabilities with which we can do a beta release with a handful of customers to gather insights on the accuracy and efficacy of the predictive model and calibrate it as required.
Qualitative Interview- Key Account Customers (Sales Teams)
Qualitative Interview- Engineering
Quantitative Survey- Focussed on AI/ML Awareness
Competitor Analysis- Identifying Market Gaps
Challenges
When a support ticket is raised and assigned to a support agent, they must respond within the allowed TAT(Turn around time) as outlined in the SLA(Service Level Agreement). If the issue resolution requires temporary access to the customer's account, they must contact the admin promptly to request for support access.
Challenges with Traditional Lead Scoring
Technical Constraints with Predictive Conversion Scoring
Pricing Model and Go to Market Strategy
Ideation+Testing
Utilising insights from the interviews we conducted with various stakeholders, we transformed insights into user stories that should be prioritised. We started whiteboard sessions to outline the layout, structure, UI components, and essential out of the box actions.We also studied our competitor products, to figure out the minimum viable feature list for our go-to-market strategy and to determine pricing plans.
Next, we created and tested a fully prototyped version of the feature with our in-house sales teams to identify usability issues and textual improvements to tailor the experience for them. We worked with the KAM team to identify customers with clean historical data for beta testing.
User Story 1: 'As a sales user, I want to prioritise leads and view smart insights for effective conversion.'
Convert leads faster using the new Conversion Score Card
With traditional lead scoring in place, administrators and sales users alike felt its impact was inadequate on their sales process and strategy. They wanted a dynamic scoring system that evolves with their business needs and goals. Sales managers expected the platform to provide insights on lead prioritisation to reduce hand-holding for newly onboarded users.
We created a dedicated card tailored to the sales user persona where they can quickly:
View the predicted score (range 0-100) indicating the likelihood of conversion.
View recent activity of that lead (ex- form filled, unsubscribed mail) which contributes to the score.
View smart insights that help users assess if the lead should be prioritised.
A collapsible card, that let's you focus on what's important and expands to show you further insights.
During the feedback session(unmoderated), test users immediately noticed the new conversion scoring card and understood the metrics it displayed. However, few observations were made:
Some of the users didn't realise that the card can be expanded for further insights.
The time parameter in the insights (Lead went up by 10 in the last 7 days) was found very useful by many sales users since it reflected how long ago was the lead active and were the recorded actions positive.
Higher number of newly onboarded users were convinced with the benefits of this feature compared to the veterans. However, this was expected since some veteran users have their own strategies and communication methods with the leads and tend to rely less on system suggested insights.
User Story 2: 'As a sales user, I want to sort the most likely to convert leads in the leads grid page.'
Quick access score card in the 'Manage Leads' grid page
We added a quick access score cad that displays the conversion score and smart insights just by hovering on any of the leads in the grid.
Hovering on a lead reveals the score card which then auto-expands to reveals further insights.
No clicks are required for this interaction and a well thought delay has been added to trigger the hover state. This ensures hovering over a lead without the intent to view the score card is honoured.
Additionally, users can also sort the conversion score column to view the leads with 'highest to lowest' score or vice versa for effective prioritisation.
During the feedback session, test users showed positive interest in the quick access card.
Majority of users utilised the sort column (high to low) functionality to identify high performing leads.
Once accustomed with the quick access card most users only triggered it (by deliberately hovering on a lead) for only leads that they prioritised.
User Story 3: 'As a new sales user, I am often confused on effective ways to convert the lead.'
Target leads effectively leveraging 'Next Best Action'
Once a sales user has prioritised the leads they want to work on, the next question is how to target the lead for a higher probability of conversion. So, we created an innovative solution that predicts the next best action using AI/ML models trained on historic lead activity data.
Lead data like website interaction patterns, purchase history, activity history, etc is utilised to train the data model to accurately predict what steps can lead to conversion.
One single action is suggested to the user at a given time to help the sales user make effective decisions and track the efficacy of system generated suggestions.
A simple feedback mechanism is also implemented in the same card to collect insights on the prediction accuracy and the impact it drives.
During the feedback session, we received a mixed review from the users. Considering, we deployed this feature on few test tenants who agreed to be part of the beta program, we understood:
Veteran users have trust in their self devised strategies and may require more time and smarter ways of suggesting insights that feel more like an aid to them and less like 'AI taking over'.
Some users raised concerns over the suggestion accuracy, which was found to be about 74% at the time of testing. Users were reassured, by optimising the model's parameters and fine tuning, accuracy will continue to increase over time.
A positive feedback rating of 4.1/5 was collected through the feedback feature in the card. It was noted that majority of users who rated negatively avoided filling any qualitative reason to do so and mostly relied on the quick suggestions to select the reason.
User Story 4: 'As an admin, I want to train the data model on relevant industry specific historic data.'
Train your own data model to best suit your business needs
Training your own data model empowers you to harness the full potential of lead data to drive informed decisions and build effective sales strategies.
Customisation: You can tailor the model specifically to your business's unique needs, preferences, and specific industry requirements.
Accuracy: You can ensure it learns from examples that closely reflect your specific customer base, behaviours, and lead activities. This can lead to more accurate predictions and insights.
Relevance: A customised model can take into account specific criteria that is vital to your sales process. You can also choose to exclude lead parameters that are irrelevant to conversion.
Control: Training your own model gives you control over the training process, data quality, and the evolution of the model over time.
Competitive Advantage: It can provide an edge by enabling more precise targeting, better prioritisation of leads, and ultimately, improved conversion rates and customer satisfaction.
Integration: It allows for seamless integration with other internal systems and processes, ensuring that the insights generated by the model can be effectively utilised across the platform.
Internally, positive feedback was received on this feature, however since this a brand new technology for our existing customers, the LSQ implementation team drives the data model training process for now. We plan to gather further feedback once the feature is deployed across all customers.
What impact did we create?
We launched the beta version of conversion scoring to selected key account customers who agreed to be part of the testing phase for an early preview of the capability. We trained the data models on real-time customer historical data and deployed them for respective customers to display predictive conversion scoring, smart insights and next best actions.
This helped us gather substantial feedback on prediction accuracy and efficacy which the engineering team will utilise to fine tune the models further. Even though, the feature is not yet released as 'general availability', the feature had substantial impact on the test cohort. It also aided in establishing pricing models and placement in pricing plans.
(Note: Metrics mentioned here are calculated from a small but substantial cohort of test customers and may change once the feature is released as 'General Availability'.)
Average lead conversion rate went up by
21%↑
Sales users now target the lead most likely to convert by sorting the conversion score column. This led to effective lead prioritisation and nullified the probability of missing a high performing lead.
Communication with the lead is now better informed with quickly accessible smart insights.
Sales users are now able to devise their own strategies faster and proactively contact the lead keeping in mind their interaction patterns and activity history.
Average time taken to convert a prospect to customer went down by
18%↓
With effective prioritisation, sales users feel empowered to target hot leads and reduce the churn rate.
The 'Next Best Action' feature keeps sales users proactive in their follow-ups, ensuring clarity on the next steps and minimising any potential delays or bottlenecks.
Training and deploying new sales users has become
Streamlined
Sales managers are now able to finish their sales training as well as platform training process faster and rely on the 'conversion scoring' and 'next best action' features to guide sales users.
We are expecting a positive jump in the 'Net Promoter Score' in future quarters
+ve↑
We are monitoring the NPS survey to isolate feedback addressing the conversion scoring module to better understand the efficacy of our current solution.
Next Steps
Once the Beta version is stable the feature will be released to all customers and marked 'General Availability'. In the following quarters we will focus on analysing feedback and refining our data model accuracy, along with required experience enhancements.
Robust feedback mechanism for Conversion Scoring
Priority 0
We will focus on building a more effective version of the feedback collection mechanism that can fetch us more qualitative insights.
Integration with other sister products in the LSQ suite
Priority 1
Conversion scoring will be integrated with other products and modules like Automations, Process builders, Email Campaigns, etc to create and utilise cross-selling and up-selling opportunities.
Project Details
Team
(most actively involved)
Product Manager, Lead Product Designer, Assoc. UX Designer, Sr. Software Engineer, Data Science Team, Sales Team, Professional Services Team
Duration
(from V1.0 to Dev handoff)
2 Months
My Role
(key contributions)
Qualitative Interviews
Competitor Analysis
Survey Analysis
UI Build
Design QA
Usability Testing
Overview
In high velocity sales, predictive conversion scoring is like having a superpower for your sales teams. This is fairly new in the LeadSquared CRM domain and is quickly transforming how businesses approach sales.
So, what's predictive conversion scoring all about? Well, it's like having a crystal ball that helps you figure out which leads are worth chasing and which ones might need a little more nurturing. It considers who your leads are, how they've interacted with your business before, and what they're doing now to predict who's most likely to become a customer.
By using this, you can focus your efforts on the leads that are most likely to convert, saving time and boosting efficiency. Now that you have a smart score assigned to your open leads, what's the 'next best action' to take on that lead? This case study dives into how conversion scoring and smart insights drive impact in a business and how their dynamic nature makes sales feel more human and approachable.
Jargons
Before we delve into how all of this works, let's take a look at some common terminologies:
Lead Scoring: It's a scoring system for your leads to quantify and rank active prospects in your system. The account admin sets up rules in the Settings>Lead Scoring page. For instance, if a lead opens one of your marketing emails and clicks on a link, they get 10 points. The admin decides on these rules and how much each action is worth. Then, a consolidated score is calculated for each lead, which we call the 'Lead Score.'
Lead Attributes: They are basically lead properties against which you record different aspects of the lead. For example, lead name, email id, phone number, address, etc.
Predictive Conversion Scoring: It's a data driven approach of predicting which lead is most likely to convert to a customer by analysing historical data and patterns, utilising advanced algorithms and machine learning techniques. Similar, to lead scoring, a consolidated conversion score is displayed to aid the sales user.
Data Model: It refers to a mathematical framework that captures patterns and relationships within a dataset. A data model is trained using historical data and optimised parameters to boost prediction efficiency.
Insights: Insights are generated based on the patterns the data model recognises. For example, personalised product recommendations for customers, forecasting future sales trends, and identifying areas for sales process optimisation are all insights generated by the data model we train.
Next Best Action: The data model will predict the next best possible action a sales user can take to push the lead through the sales funnel. It will be trained on historical sales activities performed on previous leads with similar attributes to recognise which activity helped nurture the lead faster.
How does Conversion Scoring work?
To kick off, you train a machine learning model on your historical lead data by defining an appropriate time range, the lead stages you consider to be 'won/converted' and 'lost/disqualified' and choosing lead attributes that you want the model to take into account.
The model analyses the data set and assigns a score to each lead based on the lead attributes that were calculated to be the most significant for conversion. You can also choose to retrain the model every 30 days to never miss out on emerging trends and dynamics in your business.
The score helps sales teams prioritise leads and reduce the time it takes to qualify a lead.
Analyse Support Ticket
Step 1
Select the object type for which you would like to train the model.
Select duration, lead stages, lead fields , etc to be considered for training.
Contact Customer Administrator
Step 2
Once training is completed, you can check the report to verify which lead attributes were deemed significant.
In case of any discrepancy, you can retrain the model.
Generate Login Credentials
Step 3
Configure the score field in your lead detail pages, etc to see the it in action.
'Smart Insights' will be displayed alongside the conversion score.
What is 'Next Best Action'?
Once you have the score and insights for an open lead, as a sales user you might find yourself asking, "Alright, the score suggests this is a hot lead, but what now? What's the smartest move I can make to close this deal?"
When you enable this feature in your account, a new machine learning model is trained on historical sales activity data like past interactions, purchase history, and behavioural patterns to provide tailored recommendations to sales users.
You can choose the 'Activity type' you want the model to learn from, level of predictions required, lead stage and a few other key factors which together will define how likely is it to convert a lead and the number of steps it might take to do so.
Analyse Support Ticket
Step 1
Select the activity type for which you would like to train the model.
Select duration, lead stage, activity fields , etc to be considered for training.
Contact Customer Administrator
Step 2
Once training is completed, you can configure this field in your lead detail page, etc.
You can choose to retrain the model in set intervals to increase accuracy.
Generate Login Credentials
Step 3
The next best action will be suggested for each lead in the lead details page.
You can provide feedback, objective+subjective for overall improvements.
Discovery Phase
As with any other feature request, we had to validate the ask, define scope of the build and create a mini roadmap to keep us aligned with the timelines. We employed the following methods to gather user insights like key pain points, expectations and specific user stories we can immediately address. The intent was to distill the necessary capabilities with which we can do a beta release with a handful of customers to gather insights on the accuracy and efficacy of the predictive model and calibrate it as required.
Qualitative Interview- Key Account Customers (Sales Teams)
Qualitative Interview- Engineering
Quantitative Survey- Focussed on AI/ML Awareness
Competitor Analysis- Identifying Market Gaps
Challenges
When a support ticket is raised and assigned to a support agent, they must respond within the allowed TAT(Turn around time) as outlined in the SLA(Service Level Agreement). If the issue resolution requires temporary access to the customer's account, they must contact the admin promptly to request for support access.
Challenges with Traditional Lead Scoring
Technical Constraints with Predictive Conversion Scoring
Pricing Model and Go to Market Strategy
Ideation+Testing
Utilising insights from the interviews we conducted with various stakeholders, we transformed insights into user stories that should be prioritised. We started whiteboard sessions to outline the layout, structure, UI components, and essential out of the box actions.We also studied our competitor products, to figure out the minimum viable feature list for our go-to-market strategy and to determine pricing plans.
Next, we created and tested a fully prototyped version of the feature with our in-house sales teams to identify usability issues and textual improvements to tailor the experience for them. We worked with the KAM team to identify customers with clean historical data for beta testing.
User Story 1: 'As a sales user, I want to prioritise leads and view smart insights for effective conversion.'
Convert leads faster using the new Conversion Score Card
With traditional lead scoring in place, administrators and sales users alike felt its impact was inadequate on their sales process and strategy. They wanted a dynamic scoring system that evolves with their business needs and goals. Sales managers expected the platform to provide insights on lead prioritisation to reduce hand-holding for newly onboarded users.
We created a dedicated card tailored to the sales user persona where they can quickly:
View the predicted score (range 0-100) indicating the likelihood of conversion.
View recent activity of that lead (ex- form filled, unsubscribed mail) which contributes to the score.
View smart insights that help users assess if the lead should be prioritised.
A collapsible card, that let's you focus on what's important and expands to show you further insights.
During the feedback session(unmoderated), test users immediately noticed the new conversion scoring card and understood the metrics it displayed. However, few observations were made:
Some of the users didn't realise that the card can be expanded for further insights.
The time parameter in the insights (Lead went up by 10 in the last 7 days) was found very useful by many sales users since it reflected how long ago was the lead active and were the recorded actions positive.
Higher number of newly onboarded users were convinced with the benefits of this feature compared to the veterans. However, this was expected since some veteran users have their own strategies and communication methods with the leads and tend to rely less on system suggested insights.
User Story 2: 'As a sales user, I want to sort the most likely to convert leads in the leads grid page.'
Quick access score card in the 'Manage Leads' grid page
We added a quick access score cad that displays the conversion score and smart insights just by hovering on any of the leads in the grid.
Hovering on a lead reveals the score card which then auto-expands to reveals further insights.
No clicks are required for this interaction and a well thought delay has been added to trigger the hover state. This ensures hovering over a lead without the intent to view the score card is honoured.
Additionally, users can also sort the conversion score column to view the leads with 'highest to lowest' score or vice versa for effective prioritisation.
During the feedback session, test users showed positive interest in the quick access card.
Majority of users utilised the sort column (high to low) functionality to identify high performing leads.
Once accustomed with the quick access card most users only triggered it (by deliberately hovering on a lead) for only leads that they prioritised.
User Story 3: 'As a new sales user, I am often confused on effective ways to convert the lead.'
Target leads effectively leveraging 'Next Best Action'
Once a sales user has prioritised the leads they want to work on, the next question is how to target the lead for a higher probability of conversion. So, we created an innovative solution that predicts the next best action using AI/ML models trained on historic lead activity data.
Lead data like website interaction patterns, purchase history, activity history, etc is utilised to train the data model to accurately predict what steps can lead to conversion.
One single action is suggested to the user at a given time to help the sales user make effective decisions and track the efficacy of system generated suggestions.
A simple feedback mechanism is also implemented in the same card to collect insights on the prediction accuracy and the impact it drives.
During the feedback session, we received a mixed review from the users. Considering, we deployed this feature on few test tenants who agreed to be part of the beta program, we understood:
Veteran users have trust in their self devised strategies and may require more time and smarter ways of suggesting insights that feel more like an aid to them and less like 'AI taking over'.
Some users raised concerns over the suggestion accuracy, which was found to be about 74% at the time of testing. Users were reassured, by optimising the model's parameters and fine tuning, accuracy will continue to increase over time.
A positive feedback rating of 4.1/5 was collected through the feedback feature in the card. It was noted that majority of users who rated negatively avoided filling any qualitative reason to do so and mostly relied on the quick suggestions to select the reason.
User Story 4: 'As an admin, I want to train the data model on relevant industry specific historic data.'
Train your own data model to best suit your business needs
Training your own data model empowers you to harness the full potential of lead data to drive informed decisions and build effective sales strategies.
Customisation: You can tailor the model specifically to your business's unique needs, preferences, and specific industry requirements.
Accuracy: You can ensure it learns from examples that closely reflect your specific customer base, behaviours, and lead activities. This can lead to more accurate predictions and insights.
Relevance: A customised model can take into account specific criteria that is vital to your sales process. You can also choose to exclude lead parameters that are irrelevant to conversion.
Control: Training your own model gives you control over the training process, data quality, and the evolution of the model over time.
Competitive Advantage: It can provide an edge by enabling more precise targeting, better prioritisation of leads, and ultimately, improved conversion rates and customer satisfaction.
Integration: It allows for seamless integration with other internal systems and processes, ensuring that the insights generated by the model can be effectively utilised across the platform.
Internally, positive feedback was received on this feature, however since this a brand new technology for our existing customers, the LSQ implementation team drives the data model training process for now. We plan to gather further feedback once the feature is deployed across all customers.
What impact did we create?
We launched the beta version of conversion scoring to selected key account customers who agreed to be part of the testing phase for an early preview of the capability. We trained the data models on real-time customer historical data and deployed them for respective customers to display predictive conversion scoring, smart insights and next best actions.
This helped us gather substantial feedback on prediction accuracy and efficacy which the engineering team will utilise to fine tune the models further. Even though, the feature is not yet released as 'general availability', the feature had substantial impact on the test cohort. It also aided in establishing pricing models and placement in pricing plans.
(Note: Metrics mentioned here are calculated from a small but substantial cohort of test customers and may change once the feature is released as 'General Availability'.)
Average lead conversion rate went up by
21%↑
Sales users now target the lead most likely to convert by sorting the conversion score column. This led to effective lead prioritisation and nullified the probability of missing a high performing lead.
Communication with the lead is now better informed with quickly accessible smart insights.
Sales users are now able to devise their own strategies faster and proactively contact the lead keeping in mind their interaction patterns and activity history.
Average time taken to convert a prospect to customer went down by
18%↓
With effective prioritisation, sales users feel empowered to target hot leads and reduce the churn rate.
The 'Next Best Action' feature keeps sales users proactive in their follow-ups, ensuring clarity on the next steps and minimising any potential delays or bottlenecks.
Training and deploying new sales users has become
Streamlined
Sales managers are now able to finish their sales training as well as platform training process faster and rely on the 'conversion scoring' and 'next best action' features to guide sales users.
We are expecting a positive jump in the 'Net Promoter Score' in future quarters
+ve↑
We are monitoring the NPS survey to isolate feedback addressing the conversion scoring module to better understand the efficacy of our current solution.
Next Steps
Once the Beta version is stable the feature will be released to all customers and marked 'General Availability'. In the following quarters we will focus on analysing feedback and refining our data model accuracy, along with required experience enhancements.
Robust feedback mechanism for Conversion Scoring
Priority 0
We will focus on building a more effective version of the feedback collection mechanism that can fetch us more qualitative insights.
Integration with other sister products in the LSQ suite
Priority 1
Conversion scoring will be integrated with other products and modules like Automations, Process builders, Email Campaigns, etc to create and utilise cross-selling and up-selling opportunities.
Project Details
Team
(most actively involved)
Product Manager, Lead Product Designer, Assoc. UX Designer, Sr. Software Engineer, Data Science Team, Sales Team, Professional Services Team
Duration
(from V1.0 to Dev handoff)
2 Months
My Role
(key contributions)
Qualitative Interviews
Competitor Analysis
Survey Analysis
UI Build
Design QA
Usability Testing
Overview
In high velocity sales, predictive conversion scoring is like having a superpower for your sales teams. This is fairly new in the LeadSquared CRM domain and is quickly transforming how businesses approach sales.
So, what's predictive conversion scoring all about? Well, it's like having a crystal ball that helps you figure out which leads are worth chasing and which ones might need a little more nurturing. It considers who your leads are, how they've interacted with your business before, and what they're doing now to predict who's most likely to become a customer.
By using this, you can focus your efforts on the leads that are most likely to convert, saving time and boosting efficiency. Now that you have a smart score assigned to your open leads, what's the 'next best action' to take on that lead? This case study dives into how conversion scoring and smart insights drive impact in a business and how their dynamic nature makes sales feel more human and approachable.
Jargons
Before we delve into how all of this works, let's take a look at some common terminologies:
Lead Scoring: It's a scoring system for your leads to quantify and rank active prospects in your system. The account admin sets up rules in the Settings>Lead Scoring page. For instance, if a lead opens one of your marketing emails and clicks on a link, they get 10 points. The admin decides on these rules and how much each action is worth. Then, a consolidated score is calculated for each lead, which we call the 'Lead Score.'
Lead Attributes: They are basically lead properties against which you record different aspects of the lead. For example, lead name, email id, phone number, address, etc.
Predictive Conversion Scoring: It's a data driven approach of predicting which lead is most likely to convert to a customer by analysing historical data and patterns, utilising advanced algorithms and machine learning techniques. Similar, to lead scoring, a consolidated conversion score is displayed to aid the sales user.
Data Model: It refers to a mathematical framework that captures patterns and relationships within a dataset. A data model is trained using historical data and optimised parameters to boost prediction efficiency.
Insights: Insights are generated based on the patterns the data model recognises. For example, personalised product recommendations for customers, forecasting future sales trends, and identifying areas for sales process optimisation are all insights generated by the data model we train.
Next Best Action: The data model will predict the next best possible action a sales user can take to push the lead through the sales funnel. It will be trained on historical sales activities performed on previous leads with similar attributes to recognise which activity helped nurture the lead faster.
How does Conversion Scoring work?
To kick off, you train a machine learning model on your historical lead data by defining an appropriate time range, the lead stages you consider to be 'won/converted' and 'lost/disqualified' and choosing lead attributes that you want the model to take into account.
The model analyses the data set and assigns a score to each lead based on the lead attributes that were calculated to be the most significant for conversion. You can also choose to retrain the model every 30 days to never miss out on emerging trends and dynamics in your business.
The score helps sales teams prioritise leads and reduce the time it takes to qualify a lead.
Train a Data Model
Step 1
Select the object type for which you would like to train the model.
Select duration, lead stages, lead fields , etc to be considered for training.
Configure Scoring
Step 2
Once training is completed, you can check the report to verify which lead attributes were deemed significant.
In case of any discrepancy, you can retrain the model.
Finish Setup
Step 3
Configure the score field in your lead detail pages, etc to see the it in action.
'Smart Insights' will be displayed alongside the conversion score.
What is 'Next Best Action'?
Once you have the score and insights for an open lead, as a sales user you might find yourself asking, "Alright, the score suggests this is a hot lead, but what now? What's the smartest move I can make to close this deal?"
When you enable this feature in your account, a new machine learning model is trained on historical sales activity data like past interactions, purchase history, and behavioural patterns to provide tailored recommendations to sales users.
You can choose the 'Activity type' you want the model to learn from, level of predictions required, lead stage and a few other key factors which together will define how likely is it to convert a lead and the number of steps it might take to do so.
Train a Data Model
Step 1
Select the activity type for which you would like to train the model.
Select duration, lead stage, activity fields , etc to be considered for training.
Configure NBA
Step 2
Once training is completed, you can configure this field in your lead detail page, etc.
You can choose to retrain the model in set intervals to increase accuracy.
Provide Feedback
Step 3
The next best action will be suggested for each lead in the lead details page.
You can provide feedback, objective+subjective for overall improvements.
Discovery Phase
As with any other feature request, we had to validate the ask, define scope of the build and create a mini roadmap to keep us aligned with the timelines. We employed the following methods to gather user insights like key pain points, expectations and specific user stories we can immediately address. The intent was to distill the necessary capabilities with which we can do a beta release with a handful of customers to gather insights on the accuracy and efficacy of the predictive model and calibrate it as required.
Qualitative Interview- Key Account Customers (Sales Teams)
Qualitative Interview- Engineering
Quantitative Survey- Focussed on AI/ML Awareness
Competitor Analysis- Identifying Market Gaps
Challenges
When a support ticket is raised and assigned to a support agent, they must respond within the allowed TAT(Turn around time) as outlined in the SLA(Service Level Agreement). If the issue resolution requires temporary access to the customer's account, they must contact the admin promptly to request for support access.
Challenges with Traditional Lead Scoring
Technical Constraints with Predictive Conversion Scoring
Pricing Model and Go to Market Strategy
Ideation+Testing
Utilising insights from the interviews we conducted with various stakeholders, we transformed insights into user stories that should be prioritised. We started whiteboard sessions to outline the layout, structure, UI components, and essential out of the box actions.We also studied our competitor products, to figure out the minimum viable feature list for our go-to-market strategy and to determine pricing plans.
Next, we created and tested a fully prototyped version of the feature with our in-house sales teams to identify usability issues and textual improvements to tailor the experience for them. We worked with the KAM team to identify customers with clean historical data for beta testing.
User Story 1: 'As a sales user, I want to prioritise leads and view smart insights for effective conversion.'
Convert leads faster using the new Conversion Score Card
With traditional lead scoring in place, administrators and sales users alike felt its impact was inadequate on their sales process and strategy. They wanted a dynamic scoring system that evolves with their business needs and goals. Sales managers expected the platform to provide insights on lead prioritisation to reduce hand-holding for newly onboarded users.
We created a dedicated card tailored to the sales user persona where they can quickly:
View the predicted score (range 0-100) indicating the likelihood of conversion.
View recent activity of that lead (ex- form filled, unsubscribed mail) which contributes to the score.
View smart insights that help users assess if the lead should be prioritised.
A collapsible card, that let's you focus on what's important and expands to show you further insights.
During the feedback session(unmoderated), test users immediately noticed the new conversion scoring card and understood the metrics it displayed. However, few observations were made:
Some of the users didn't realise that the card can be expanded for further insights.
The time parameter in the insights (Lead went up by 10 in the last 7 days) was found very useful by many sales users since it reflected how long ago was the lead active and were the recorded actions positive.
Higher number of newly onboarded users were convinced with the benefits of this feature compared to the veterans. However, this was expected since some veteran users have their own strategies and communication methods with the leads and tend to rely less on system suggested insights.
User Story 2: 'As a sales user, I want to sort the most likely to convert leads in the leads grid page.'
Quick access score card in the 'Manage Leads' grid page
We added a quick access score cad that displays the conversion score and smart insights just by hovering on any of the leads in the grid.
Hovering on a lead reveals the score card which then auto-expands to reveals further insights.
No clicks are required for this interaction and a well thought delay has been added to trigger the hover state. This ensures hovering over a lead without the intent to view the score card is honoured.
Additionally, users can also sort the conversion score column to view the leads with 'highest to lowest' score or vice versa for effective prioritisation.
During the feedback session, test users showed positive interest in the quick access card.
Majority of users utilised the sort column (high to low) functionality to identify high performing leads.
Once accustomed with the quick access card most users only triggered it (by deliberately hovering on a lead) for only leads that they prioritised.
User Story 3: 'As a new sales user, I am often confused on effective ways to convert the lead.'
Target leads effectively leveraging 'Next Best Action'
Once a sales user has prioritised the leads they want to work on, the next question is how to target the lead for a higher probability of conversion. So, we created an innovative solution that predicts the next best action using AI/ML models trained on historic lead activity data.
Lead data like website interaction patterns, purchase history, activity history, etc is utilised to train the data model to accurately predict what steps can lead to conversion.
One single action is suggested to the user at a given time to help the sales user make effective decisions and track the efficacy of system generated suggestions.
A simple feedback mechanism is also implemented in the same card to collect insights on the prediction accuracy and the impact it drives.
During the feedback session, we received a mixed review from the users. Considering, we deployed this feature on few test tenants who agreed to be part of the beta program, we understood:
Veteran users have trust in their self devised strategies and may require more time and smarter ways of suggesting insights that feel more like an aid to them and less like 'AI taking over'.
Some users raised concerns over the suggestion accuracy, which was found to be about 74% at the time of testing. Users were reassured, by optimising the model's parameters and fine tuning, accuracy will continue to increase over time.
A positive feedback rating of 4.1/5 was collected through the feedback feature in the card. It was noted that majority of users who rated negatively avoided filling any qualitative reason to do so and mostly relied on the quick suggestions to select the reason.
User Story 4: 'As an admin, I want to train the data model on relevant industry specific historic data.'
Train your own data model to best suit your business needs
Training your own data model empowers you to harness the full potential of lead data to drive informed decisions and build effective sales strategies.
Customisation: You can tailor the model specifically to your business's unique needs, preferences, and specific industry requirements.
Accuracy: You can ensure it learns from examples that closely reflect your specific customer base, behaviours, and lead activities. This can lead to more accurate predictions and insights.
Relevance: A customised model can take into account specific criteria that is vital to your sales process. You can also choose to exclude lead parameters that are irrelevant to conversion.
Control: Training your own model gives you control over the training process, data quality, and the evolution of the model over time.
Competitive Advantage: It can provide an edge by enabling more precise targeting, better prioritisation of leads, and ultimately, improved conversion rates and customer satisfaction.
Integration: It allows for seamless integration with other internal systems and processes, ensuring that the insights generated by the model can be effectively utilised across the platform.
Internally, positive feedback was received on this feature, however since this a brand new technology for our existing customers, the LSQ implementation team drives the data model training process for now. We plan to gather further feedback once the feature is deployed across all customers.
What impact did we create?
We launched the beta version of conversion scoring to selected key account customers who agreed to be part of the testing phase for an early preview of the capability. We trained the data models on real-time customer historical data and deployed them for respective customers to display predictive conversion scoring, smart insights and next best actions.
This helped us gather substantial feedback on prediction accuracy and efficacy which the engineering team will utilise to fine tune the models further. Even though, the feature is not yet released as 'general availability', the feature had substantial impact on the test cohort. It also aided in establishing pricing models and placement in pricing plans.
(Note: Metrics mentioned here are calculated from a small but substantial cohort of test customers and may change once the feature is released as 'General Availability'.)
Average lead conversion rate went up by
21%↑
Sales users now target the lead most likely to convert by sorting the conversion score column. This led to effective lead prioritisation and nullified the probability of missing a high performing lead.
Communication with the lead is now better informed with quickly accessible smart insights.
Sales users are now able to devise their own strategies faster and proactively contact the lead keeping in mind their interaction patterns and activity history.
Average time taken to convert a prospect to customer went down by
18%↓
With effective prioritisation, sales users feel empowered to target hot leads and reduce the churn rate.
The 'Next Best Action' feature keeps sales users proactive in their follow-ups, ensuring clarity on the next steps and minimising any potential delays or bottlenecks.
Training and deploying new sales users has become
Streamlined
Sales managers are now able to finish their sales training as well as platform training process faster and rely on the 'conversion scoring' and 'next best action' features to guide sales users.
We are expecting a positive jump in the 'Net Promoter Score' in future quarters
+ve↑
We are monitoring the NPS survey to isolate feedback addressing the conversion scoring module to better understand the efficacy of our current solution.
Next Steps
Once the Beta version is stable the feature will be released to all customers and marked 'General Availability'. In the following quarters we will focus on analysing feedback and refining our data model accuracy, along with required experience enhancements.
Robust feedback mechanism for Conversion Scoring
Priority 0
We will focus on building a more effective version of the feedback collection mechanism that can fetch us more qualitative insights.
Integration with other sister products in the LSQ suite
Priority 1
Conversion scoring will be integrated with other products and modules like Automations, Process builders, Email Campaigns, etc to create and utilise cross-selling and up-selling opportunities.
Project Details
Team
(most actively involved)
Product Manager, Lead Product Designer, Assoc. UX Designer, Sr. Software Engineer, Data Science Team, Sales Team, Professional Services Team
Duration
(from V1.0 to Dev handoff)
2 Months
My Role
(key contributions)
Qualitative Interviews
Competitor Analysis
Survey Analysis
UI Build
Design QA
Usability Testing
Overview
In high velocity sales, predictive conversion scoring is like having a superpower for your sales teams. This is fairly new in the LeadSquared CRM domain and is quickly transforming how businesses approach sales.
So, what's predictive conversion scoring all about? Well, it's like having a crystal ball that helps you figure out which leads are worth chasing and which ones might need a little more nurturing. It considers who your leads are, how they've interacted with your business before, and what they're doing now to predict who's most likely to become a customer.
By using this, you can focus your efforts on the leads that are most likely to convert, saving time and boosting efficiency. Now that you have a smart score assigned to your open leads, what's the 'next best action' to take on that lead? This case study dives into how conversion scoring and smart insights drive impact in a business and how their dynamic nature makes sales feel more human and approachable.
Jargons
Before we delve into how all of this works, let's take a look at some common terminologies:
Lead Scoring: It's a scoring system for your leads to quantify and rank active prospects in your system. The account admin sets up rules in the Settings>Lead Scoring page. For instance, if a lead opens one of your marketing emails and clicks on a link, they get 10 points. The admin decides on these rules and how much each action is worth. Then, a consolidated score is calculated for each lead, which we call the 'Lead Score.'
Lead Attributes: They are basically lead properties against which you record different aspects of the lead. For example, lead name, email id, phone number, address, etc.
Predictive Conversion Scoring: It's a data driven approach of predicting which lead is most likely to convert to a customer by analysing historical data and patterns, utilising advanced algorithms and machine learning techniques. Similar, to lead scoring, a consolidated conversion score is displayed to aid the sales user.
Data Model: It refers to a mathematical framework that captures patterns and relationships within a dataset. A data model is trained using historical data and optimised parameters to boost prediction efficiency.
Insights: Insights are generated based on the patterns the data model recognises. For example, personalised product recommendations for customers, forecasting future sales trends, and identifying areas for sales process optimisation are all insights generated by the data model we train.
Next Best Action: The data model will predict the next best possible action a sales user can take to push the lead through the sales funnel. It will be trained on historical sales activities performed on previous leads with similar attributes to recognise which activity helped nurture the lead faster.
How does Conversion Scoring work?
To kick off, you train a machine learning model on your historical lead data by defining an appropriate time range, the lead stages you consider to be 'won/converted' and 'lost/disqualified' and choosing lead attributes that you want the model to take into account.
The model analyses the data set and assigns a score to each lead based on the lead attributes that were calculated to be the most significant for conversion. You can also choose to retrain the model every 30 days to never miss out on emerging trends and dynamics in your business.
The score helps sales teams prioritise leads and reduce the time it takes to qualify a lead.
Train a Data Model
Step 1
Select the object type for which you would like to train the model.
Select duration, lead stages, lead fields , etc to be considered for training.
Configure Scoring
Step 2
Once training is completed, you can check the report to verify which lead attributes were deemed significant.
In case of any discrepancy, you can retrain the model.
Finish Setup
Step 3
Configure the score field in your lead detail pages, etc to see the it in action.
'Smart Insights' will be displayed alongside the conversion score.
What is 'Next Best Action'?
Once you have the score and insights for an open lead, as a sales user you might find yourself asking, "Alright, the score suggests this is a hot lead, but what now? What's the smartest move I can make to close this deal?"
When you enable this feature in your account, a new machine learning model is trained on historical sales activity data like past interactions, purchase history, and behavioural patterns to provide tailored recommendations to sales users.
You can choose the 'Activity type' you want the model to learn from, level of predictions required, lead stage and a few other key factors which together will define how likely is it to convert a lead and the number of steps it might take to do so.
Train a Data Model
Step 1
Select the activity type for which you would like to train the model.
Select duration, lead stage, activity fields , etc to be considered for training.
Configure NBA
Step 2
Once training is completed, you can configure this field in your lead detail page, etc.
You can choose to retrain the model in set intervals to increase accuracy.
Provide Feedback
Step 3
The next best action will be suggested for each lead in the lead details page.
You can provide feedback, objective+subjective for overall improvements.
Discovery Phase
As with any other feature request, we had to validate the ask, define scope of the build and create a mini roadmap to keep us aligned with the timelines. We employed the following methods to gather user insights like key pain points, expectations and specific user stories we can immediately address. The intent was to distill the necessary capabilities with which we can do a beta release with a handful of customers to gather insights on the accuracy and efficacy of the predictive model and calibrate it as required.
Qualitative Interview- Key Account Customers (Sales Teams)
Qualitative Interview- Engineering
Quantitative Survey- Focussed on AI/ML Awareness
Competitor Analysis- Identifying Market Gaps
Challenges
When a support ticket is raised and assigned to a support agent, they must respond within the allowed TAT(Turn around time) as outlined in the SLA(Service Level Agreement). If the issue resolution requires temporary access to the customer's account, they must contact the admin promptly to request for support access.
Challenges with Traditional Lead Scoring
Technical Constraints with Predictive Conversion Scoring
Pricing Model and Go to Market Strategy
Ideation+Testing
Utilising insights from the interviews we conducted with various stakeholders, we transformed insights into user stories that should be prioritised. We started whiteboard sessions to outline the layout, structure, UI components, and essential out of the box actions.We also studied our competitor products, to figure out the minimum viable feature list for our go-to-market strategy and to determine pricing plans.
Next, we created and tested a fully prototyped version of the feature with our in-house sales teams to identify usability issues and textual improvements to tailor the experience for them. We worked with the KAM team to identify customers with clean historical data for beta testing.
User Story 1: 'As a sales user, I want to prioritise leads and view smart insights for effective conversion.'
Convert leads faster using the new Conversion Score Card
With traditional lead scoring in place, administrators and sales users alike felt its impact was inadequate on their sales process and strategy. They wanted a dynamic scoring system that evolves with their business needs and goals. Sales managers expected the platform to provide insights on lead prioritisation to reduce hand-holding for newly onboarded users.
We created a dedicated card tailored to the sales user persona where they can quickly:
View the predicted score (range 0-100) indicating the likelihood of conversion.
View recent activity of that lead (ex- form filled, unsubscribed mail) which contributes to the score.
View smart insights that help users assess if the lead should be prioritised.
A collapsible card, that let's you focus on what's important and expands to show you further insights.
During the feedback session(unmoderated), test users immediately noticed the new conversion scoring card and understood the metrics it displayed. However, few observations were made:
Some of the users didn't realise that the card can be expanded for further insights.
The time parameter in the insights (Lead went up by 10 in the last 7 days) was found very useful by many sales users since it reflected how long ago was the lead active and were the recorded actions positive.
Higher number of newly onboarded users were convinced with the benefits of this feature compared to the veterans. However, this was expected since some veteran users have their own strategies and communication methods with the leads and tend to rely less on system suggested insights.
User Story 2: 'As a sales user, I want to sort the most likely to convert leads in the leads grid page.'
Quick access score card in the 'Manage Leads' grid page
We added a quick access score cad that displays the conversion score and smart insights just by hovering on any of the leads in the grid.
Hovering on a lead reveals the score card which then auto-expands to reveals further insights.
No clicks are required for this interaction and a well thought delay has been added to trigger the hover state. This ensures hovering over a lead without the intent to view the score card is honoured.
Additionally, users can also sort the conversion score column to view the leads with 'highest to lowest' score or vice versa for effective prioritisation.
During the feedback session, test users showed positive interest in the quick access card.
Majority of users utilised the sort column (high to low) functionality to identify high performing leads.
Once accustomed with the quick access card most users only triggered it (by deliberately hovering on a lead) for only leads that they prioritised.
User Story 3: 'As a new sales user, I am often confused on effective ways to convert the lead.'
Target leads effectively leveraging 'Next Best Action'
Once a sales user has prioritised the leads they want to work on, the next question is how to target the lead for a higher probability of conversion. So, we created an innovative solution that predicts the next best action using AI/ML models trained on historic lead activity data.
Lead data like website interaction patterns, purchase history, activity history, etc is utilised to train the data model to accurately predict what steps can lead to conversion.
One single action is suggested to the user at a given time to help the sales user make effective decisions and track the efficacy of system generated suggestions.
A simple feedback mechanism is also implemented in the same card to collect insights on the prediction accuracy and the impact it drives.
During the feedback session, we received a mixed review from the users. Considering, we deployed this feature on few test tenants who agreed to be part of the beta program, we understood:
Veteran users have trust in their self devised strategies and may require more time and smarter ways of suggesting insights that feel more like an aid to them and less like 'AI taking over'.
Some users raised concerns over the suggestion accuracy, which was found to be about 74% at the time of testing. Users were reassured, by optimising the model's parameters and fine tuning, accuracy will continue to increase over time.
A positive feedback rating of 4.1/5 was collected through the feedback feature in the card. It was noted that majority of users who rated negatively avoided filling any qualitative reason to do so and mostly relied on the quick suggestions to select the reason.
User Story 4: 'As an admin, I want to train the data model on relevant industry specific historic data.'
Train your own data model to best suit your business needs
Training your own data model empowers you to harness the full potential of lead data to drive informed decisions and build effective sales strategies.
Customisation: You can tailor the model specifically to your business's unique needs, preferences, and specific industry requirements.
Accuracy: You can ensure it learns from examples that closely reflect your specific customer base, behaviours, and lead activities. This can lead to more accurate predictions and insights.
Relevance: A customised model can take into account specific criteria that is vital to your sales process. You can also choose to exclude lead parameters that are irrelevant to conversion.
Control: Training your own model gives you control over the training process, data quality, and the evolution of the model over time.
Competitive Advantage: It can provide an edge by enabling more precise targeting, better prioritisation of leads, and ultimately, improved conversion rates and customer satisfaction.
Integration: It allows for seamless integration with other internal systems and processes, ensuring that the insights generated by the model can be effectively utilised across the platform.
Internally, positive feedback was received on this feature, however since this a brand new technology for our existing customers, the LSQ implementation team drives the data model training process for now. We plan to gather further feedback once the feature is deployed across all customers.
What impact did we create?
We launched the beta version of conversion scoring to selected key account customers who agreed to be part of the testing phase for an early preview of the capability. We trained the data models on real-time customer historical data and deployed them for respective customers to display predictive conversion scoring, smart insights and next best actions.
This helped us gather substantial feedback on prediction accuracy and efficacy which the engineering team will utilise to fine tune the models further. Even though, the feature is not yet released as 'general availability', the feature had substantial impact on the test cohort. It also aided in establishing pricing models and placement in pricing plans.
(Note: Metrics mentioned here are calculated from a small but substantial cohort of test customers and may change once the feature is released as 'General Availability'.)
Average lead conversion rate went up by
21%↑
Sales users now target the lead most likely to convert by sorting the conversion score column. This led to effective lead prioritisation and nullified the probability of missing a high performing lead.
Communication with the lead is now better informed with quickly accessible smart insights.
Sales users are now able to devise their own strategies faster and proactively contact the lead keeping in mind their interaction patterns and activity history.
Average time taken to convert a prospect to customer went down by
18%↓
With effective prioritisation, sales users feel empowered to target hot leads and reduce the churn rate.
The 'Next Best Action' feature keeps sales users proactive in their follow-ups, ensuring clarity on the next steps and minimising any potential delays or bottlenecks.
Training and deploying new sales users has become
Streamlined
Sales managers are now able to finish their sales training as well as platform training process faster and rely on the 'conversion scoring' and 'next best action' features to guide sales users.
We are expecting a positive jump in the 'Net Promoter Score' in future quarters
+ve↑
We are monitoring the NPS survey to isolate feedback addressing the conversion scoring module to better understand the efficacy of our current solution.
Next Steps
Once the Beta version is stable the feature will be released to all customers and marked 'General Availability'. In the following quarters we will focus on analysing feedback and refining our data model accuracy, along with required experience enhancements.
Robust feedback mechanism for Conversion Scoring
Priority 0
We will focus on building a more effective version of the feedback collection mechanism that can fetch us more qualitative insights.
Integration with other sister products in the LSQ suite
Priority 1
Conversion scoring will be integrated with other products and modules like Automations, Process builders, Email Campaigns, etc to create and utilise cross-selling and up-selling opportunities.
Project Details
Team
(most actively involved)
Product Manager, Lead Product Designer, Assoc. UX Designer, Sr. Software Engineer, Data Science Team, Sales Team, Professional Services Team
Duration
(from V1.0 to Dev handoff)
2 Months
My Role
(key contributions)
Qualitative Interviews
Competitor Analysis
Survey Analysis
UI Build
Design QA
Usability Testing
Overview
In high velocity sales, predictive conversion scoring is like having a superpower for your sales teams. This is fairly new in the LeadSquared CRM domain and is quickly transforming how businesses approach sales.
So, what's predictive conversion scoring all about? Well, it's like having a crystal ball that helps you figure out which leads are worth chasing and which ones might need a little more nurturing. It considers who your leads are, how they've interacted with your business before, and what they're doing now to predict who's most likely to become a customer.
By using this, you can focus your efforts on the leads that are most likely to convert, saving time and boosting efficiency. Now that you have a smart score assigned to your open leads, what's the 'next best action' to take on that lead? This case study dives into how conversion scoring and smart insights drive impact in a business and how their dynamic nature makes sales feel more human and approachable.
Jargons
Before we delve into how all of this works, let's take a look at some common terminologies:
Lead Scoring: It's a scoring system for your leads to quantify and rank active prospects in your system. The account admin sets up rules in the Settings>Lead Scoring page. For instance, if a lead opens one of your marketing emails and clicks on a link, they get 10 points. The admin decides on these rules and how much each action is worth. Then, a consolidated score is calculated for each lead, which we call the 'Lead Score.'
Lead Attributes: They are basically lead properties against which you record different aspects of the lead. For example, lead name, email id, phone number, address, etc.
Predictive Conversion Scoring: It's a data driven approach of predicting which lead is most likely to convert to a customer by analysing historical data and patterns, utilising advanced algorithms and machine learning techniques. Similar, to lead scoring, a consolidated conversion score is displayed to aid the sales user.
Data Model: It refers to a mathematical framework that captures patterns and relationships within a dataset. A data model is trained using historical data and optimised parameters to boost prediction efficiency.
Insights: Insights are generated based on the patterns the data model recognises. For example, personalised product recommendations for customers, forecasting future sales trends, and identifying areas for sales process optimisation are all insights generated by the data model we train.
Next Best Action: The data model will predict the next best possible action a sales user can take to push the lead through the sales funnel. It will be trained on historical sales activities performed on previous leads with similar attributes to recognise which activity helped nurture the lead faster.
How does Conversion Scoring work?
To kick off, you train a machine learning model on your historical lead data by defining an appropriate time range, the lead stages you consider to be 'won/converted' and 'lost/disqualified' and choosing lead attributes that you want the model to take into account.
The model analyses the data set and assigns a score to each lead based on the lead attributes that were calculated to be the most significant for conversion. You can also choose to retrain the model every 30 days to never miss out on emerging trends and dynamics in your business.
The score helps sales teams prioritise leads and reduce the time it takes to qualify a lead.
Train a Data Model
Step 1
Select the object type for which you would like to train the model.
Select duration, lead stages, lead fields , etc to be considered for training.
Configure Scoring
Step 2
Once training is completed, you can check the report to verify which lead attributes were deemed significant.
In case of any discrepancy, you can retrain the model.
Finish Setup
Step 3
Configure the score field in your lead detail pages, etc to see the it in action.
'Smart Insights' will be displayed alongside the conversion score.
What is 'Next Best Action'?
Once you have the score and insights for an open lead, as a sales user you might find yourself asking, "Alright, the score suggests this is a hot lead, but what now? What's the smartest move I can make to close this deal?"
When you enable this feature in your account, a new machine learning model is trained on historical sales activity data like past interactions, purchase history, and behavioural patterns to provide tailored recommendations to sales users.
You can choose the 'Activity type' you want the model to learn from, level of predictions required, lead stage and a few other key factors which together will define how likely is it to convert a lead and the number of steps it might take to do so.
Train a Data Model
Step 1
Select the activity type for which you would like to train the model.
Select duration, lead stage, activity fields , etc to be considered for training.
Configure NBA
Step 2
Once training is completed, you can configure this field in your lead detail page, etc.
You can choose to retrain the model in set intervals to increase accuracy.
Provide Feedback
Step 3
The next best action will be suggested for each lead in the lead details page.
You can provide feedback, objective+subjective for overall improvements.
Discovery Phase
As with any other feature request, we had to validate the ask, define scope of the build and create a mini roadmap to keep us aligned with the timelines. We employed the following methods to gather user insights like key pain points, expectations and specific user stories we can immediately address. The intent was to distill the necessary capabilities with which we can do a beta release with a handful of customers to gather insights on the accuracy and efficacy of the predictive model and calibrate it as required.
Qualitative Interview- Key Customers
Qualitative Interview- Engineering
Survey- AI/ML Awareness
Competitor Analysis- Market Gaps
Challenges
When a support ticket is raised and assigned to a support agent, they must respond within the allowed TAT(Turn around time) as outlined in the SLA(Service Level Agreement). If the issue resolution requires temporary access to the customer's account, they must contact the admin promptly to request for support access.
Challenges with Traditional Lead Scoring
Technical Constraints with Predictive Conversion Scoring
Pricing Model and Go to Market Strategy
Ideation+Testing
Utilising insights from the interviews we conducted with various stakeholders, we transformed insights into user stories that should be prioritised. We started whiteboard sessions to outline the layout, structure, UI components, and essential out of the box actions.We also studied our competitor products, to figure out the minimum viable feature list for our go-to-market strategy and to determine pricing plans.
Next, we created and tested a fully prototyped version of the feature with our in-house sales teams to identify usability issues and textual improvements to tailor the experience for them. We worked with the KAM team to identify customers with clean historical data for beta testing.
User Story 1: 'As a sales user, I want to prioritise leads and view smart insights for effective conversion.'
Convert leads faster using the new Conversion Score Card
With traditional lead scoring in place, administrators and sales users alike felt its impact was inadequate on their sales process and strategy. They wanted a dynamic scoring system that evolves with their business needs and goals. Sales managers expected the platform to provide insights on lead prioritisation to reduce hand-holding for newly onboarded users.
We created a dedicated card tailored to the sales user persona where they can quickly:
View the predicted score (range 0-100) indicating the likelihood of conversion.
View recent activity of that lead (ex- form filled, unsubscribed mail) which contributes to the score.
View smart insights that help users assess if the lead should be prioritised.
A collapsible card, that let's you focus on what's important and expands to show you further insights.
During the feedback session(unmoderated), test users immediately noticed the new conversion scoring card and understood the metrics it displayed. However, few observations were made:
Some of the users didn't realise that the card can be expanded for further insights.
The time parameter in the insights (Lead went up by 10 in the last 7 days) was found very useful by many sales users since it reflected how long ago was the lead active and were the recorded actions positive.
Higher number of newly onboarded users were convinced with the benefits of this feature compared to the veterans. However, this was expected since some veteran users have their own strategies and communication methods with the leads and tend to rely less on system suggested insights.
User Story 2: 'As a sales user, I want to sort the most likely to convert leads in the leads grid page.'
Quick access score card in the 'Manage Leads' grid page
We added a quick access score cad that displays the conversion score and smart insights just by hovering on any of the leads in the grid.
Hovering on a lead reveals the score card which then auto-expands to reveals further insights.
No clicks are required for this interaction and a well thought delay has been added to trigger the hover state. This ensures hovering over a lead without the intent to view the score card is honoured.
Additionally, users can also sort the conversion score column to view the leads with 'highest to lowest' score or vice versa for effective prioritisation.
During the feedback session, test users showed positive interest in the quick access card.
Majority of users utilised the sort column (high to low) functionality to identify high performing leads.
Once accustomed with the quick access card most users only triggered it (by deliberately hovering on a lead) for only leads that they prioritised.
User Story 3: 'As a new sales user, I am often confused on effective ways to convert the lead.'
Target leads effectively leveraging 'Next Best Action'
Once a sales user has prioritised the leads they want to work on, the next question is how to target the lead for a higher probability of conversion. So, we created an innovative solution that predicts the next best action using AI/ML models trained on historic lead activity data.
Lead data like website interaction patterns, purchase history, activity history, etc is utilised to train the data model to accurately predict what steps can lead to conversion.
One single action is suggested to the user at a given time to help the sales user make effective decisions and track the efficacy of system generated suggestions.
A simple feedback mechanism is also implemented in the same card to collect insights on the prediction accuracy and the impact it drives.
During the feedback session, we received a mixed review from the users. Considering, we deployed this feature on few test tenants who agreed to be part of the beta program, we understood:
Veteran users have trust in their self devised strategies and may require more time and smarter ways of suggesting insights that feel more like an aid to them and less like 'AI taking over'.
Some users raised concerns over the suggestion accuracy, which was found to be about 74% at the time of testing. Users were reassured, by optimising the model's parameters and fine tuning, accuracy will continue to increase over time.
A positive feedback rating of 4.1/5 was collected through the feedback feature in the card. It was noted that majority of users who rated negatively avoided filling any qualitative reason to do so and mostly relied on the quick suggestions to select the reason.
User Story 4: 'As an admin, I want to train the data model on relevant industry specific historic data.'
Train your own data model to best suit your business needs
Training your own data model empowers you to harness the full potential of lead data to drive informed decisions and build effective sales strategies.
Customisation: You can tailor the model specifically to your business's unique needs, preferences, and specific industry requirements.
Accuracy: You can ensure it learns from examples that closely reflect your specific customer base, behaviours, and lead activities. This can lead to more accurate predictions and insights.
Relevance: A customised model can take into account specific criteria that is vital to your sales process. You can also choose to exclude lead parameters that are irrelevant to conversion.
Control: Training your own model gives you control over the training process, data quality, and the evolution of the model over time.
Competitive Advantage: It can provide an edge by enabling more precise targeting, better prioritisation of leads, and ultimately, improved conversion rates and customer satisfaction.
Integration: It allows for seamless integration with other internal systems and processes, ensuring that the insights generated by the model can be effectively utilised across the platform.
Internally, positive feedback was received on this feature, however since this a brand new technology for our existing customers, the LSQ implementation team drives the data model training process for now. We plan to gather further feedback once the feature is deployed across all customers.
What impact did we create?
We launched the beta version of conversion scoring to selected key account customers who agreed to be part of the testing phase for an early preview of the capability. We trained the data models on real-time customer historical data and deployed them for respective customers to display predictive conversion scoring, smart insights and next best actions.
This helped us gather substantial feedback on prediction accuracy and efficacy which the engineering team will utilise to fine tune the models further. Even though, the feature is not yet released as 'general availability', the feature had substantial impact on the test cohort. It also aided in establishing pricing models and placement in pricing plans.
(Note: Metrics mentioned here are calculated from a small but substantial cohort of test customers and may change once the feature is released as 'General Availability'.)
Average lead conversion rate went up by
21%↑
Sales users now target the lead most likely to convert by sorting the conversion score column. This led to effective lead prioritisation and nullified the probability of missing a high performing lead.
Communication with the lead is now better informed with quickly accessible smart insights.
Sales users are now able to devise their own strategies faster and proactively contact the lead keeping in mind their interaction patterns and activity history.
Average time taken to convert a prospect to customer went down by
18%↓
With effective prioritisation, sales users feel empowered to target hot leads and reduce the churn rate.
The 'Next Best Action' feature keeps sales users proactive in their follow-ups, ensuring clarity on the next steps and minimising any potential delays or bottlenecks.
Training and deploying new sales users has become
Streamlined
Sales managers are now able to finish their sales training as well as platform training process faster and rely on the 'conversion scoring' and 'next best action' features to guide sales users.
We are expecting a positive jump in the 'Net Promoter Score' in future quarters
+ve↑
We are monitoring the NPS survey to isolate feedback addressing the conversion scoring module to better understand the efficacy of our current solution.
Next Steps
Once the Beta version is stable the feature will be released to all customers and marked 'General Availability'. In the following quarters we will focus on analysing feedback and refining our data model accuracy, along with required experience enhancements.
Robust feedback mechanism for Conversion Scoring
Priority 0
We will focus on building a more effective version of the feedback collection mechanism that can fetch us more qualitative insights.
Integration with other sister products in the LSQ suite
Priority 1
Conversion scoring will be integrated with other products and modules like Automations, Process builders, Email Campaigns, etc to create and utilise cross-selling and up-selling opportunities.
Would you like to know more about this project and the impact we created?