A bank may have thousands of customers eligible for a personal loan, credit card, or fixed deposit. Yet only a small proportion may be interested in purchasing that product at a particular time. Contacting everyone creates unnecessary campaign costs, increases customer fatigue, and consumes valuable sales capacity.
A propensity model for cross-sell helps banks identify which existing customers are most likely to purchase a particular financial product within a defined period. By analysing transaction behaviour, product holdings, customer engagement, and historical purchase patterns, banks can prioritise relevant offers and allocate their resources more effectively.
For Indian banks and NBFCs, this approach creates an opportunity to improve cross-sell conversion, strengthen existing customer relationships, and make better use of first-party customer data.
However, a propensity score alone cannot determine which product a bank should offer. Successful cross-selling also requires eligibility checks, appropriate timing, customer preferences, controlled outreach, and continuous measurement.
What Is a Propensity Model for Cross-Sell in Banking?
A propensity model for cross-sell is a predictive analytics model that estimates the probability of an existing customer purchasing an additional banking or financial product within a specified time period.
For example, a bank might want to predict which savings account customers are likely to apply for a credit card over the next 30 days.
The model examines relevant historical data and assigns a score to each customer. Higher scores indicate a greater predicted likelihood of the target outcome.
Consider three hypothetical customers:
Customer | Predicted credit card purchase probability | Cross-sell priority |
|---|---|---|
Customer A | 78% | High |
Customer B | 42% | Medium |
Customer C | 11% | Low |
These are illustrative model outputs, not actual banking conversion benchmarks. The percentages should only be interpreted as probabilities if the model has been appropriately calibrated.
Instead of contacting all three customers with the same offer, the bank can prioritise Customer A, evaluate whether the product is suitable, and select an appropriate engagement channel.
A reliable customer propensity scoring system should answer three questions:
- Which customer is likely to purchase?
- Which financial product is the customer likely to purchase?
- When and how should the bank approach that customer?
The first question is a prediction problem. The other two require additional business rules, decisioning, and customer engagement intelligence.
Why Do Banks Need Cross-Sell Propensity Models?
Banks already hold extensive information about customer relationships, including account activity, existing products, and previous interactions.
The challenge is converting that information into relevant, actionable opportunities.
Improve Customer Targeting
Traditional cross-selling often relies on broad customer segments.
For example, a bank might promote a premium credit card to customers above a particular income threshold. While this identifies a potentially relevant audience, it does not establish whether those customers are interested in the card.
A likelihood to buy model in banking considers additional behavioural signals, such as existing card usage, recent enquiries, and engagement with previous campaigns.
This allows the bank to distinguish between customers who meet basic targeting criteria and those who demonstrate stronger purchase intent.
Reduce Unnecessary Campaign Spending
Every marketing campaign involves communication costs, operational effort, and potentially relationship manager time.
Propensity-based targeting allows banks to prioritise customers with stronger predicted purchase likelihood before allocating these resources.
The commercial value depends on whether the targeted campaign produces sufficient incremental conversions to justify its costs.
Strengthen Existing Customer Relationships
A customer receiving repeated, irrelevant offers may become less responsive to future communication.
Cross-sell analytics for lenders can help identify products that better match a customer's current circumstances.
For example, an existing borrower approaching loan maturity may be interested in a suitable refinancing or top-up option. Another customer building regular savings may be more receptive to a fixed deposit.
Relevant offers can support deeper customer relationships when they reflect genuine needs and customer choice.
Support Better Sales Productivity
Banks can use propensity scores to prioritise leads for relationship managers, contact centres, and digital sales teams.
Instead of working through an undifferentiated customer list, teams can focus on customers with stronger predicted interest and appropriate product eligibility.
This creates a clearer connection between analytics, sales activity, and measurable outcomes.
How Does a Propensity Model for Cross-Sell Work?
A cross-sell propensity model uses historical customer data to learn which characteristics and behaviours are associated with subsequent product purchases.
The process typically includes five stages.
1. Define the Product and Prediction Window
A bank must first specify the outcome it wants to predict.
Examples include:
- Purchase of a credit card within 30 days.
- Opening a fixed deposit within 60 days.
- Acceptance of a personal loan offer within 45 days.
- Adoption of an additional banking product within 90 days.
Each prediction requires a clearly defined observation period and outcome.
A generic objective such as "identify customers interested in loans" is insufficient for developing a reliable model.
2. Collect Relevant Customer Data
The bank identifies permitted data sources that may explain purchasing behaviour.
These can include transaction activity, existing financial products, historical enquiries, campaign responses, and customer relationship information.
Data quality is essential. Incomplete records, inconsistent customer identifiers, and outdated information can reduce predictive reliability.
3. Engineer Predictive Features
Feature engineering converts raw customer data into variables that a machine learning model can analyse.
For example, instead of using every individual account transaction, a bank could calculate:
- Average monthly account balance over the previous 90 days.
- Number of salary credits over the previous six months.
- Change in monthly spending.
- Number of active banking products.
- Days since the last product enquiry.
- Response to previous product campaigns.
These features can help identify patterns associated with future purchases.
4. Train and Validate the Model
Historical examples are used to train a classification model.
For each customer and observation date, the target variable might be:
- 1: The customer purchased the product within the next 30 days.
- 0: The customer did not purchase the product within that period.
The model learns relationships between the available features and the target outcome.
It is then evaluated using data that was not used for training, preferably from a later time period.
5. Generate Scores and Apply Business Rules
Once validated, the model scores eligible customer-product combinations.
The bank can then apply additional rules covering product suitability, credit eligibility, communication permissions, customer exclusions, and campaign priorities.
The result is a prioritised audience that can be passed into an approved engagement workflow.
Importantly, a high propensity score predicts potential demand; it does not establish creditworthiness or constitute loan approval.
What Customer Data Is Used in Banking Cross-Sell Models?
Different products require different predictive signals.
A bank promoting a credit card may use different variables from a bank targeting customers for term deposits.
Data category | Example signals | Potential use |
|---|---|---|
Customer relationship | Account tenure, existing products, relationship segment | Understand current product ownership |
Transaction behaviour | Salary credits, recurring transactions, expenditure trends | Identify financial activity patterns |
Savings behaviour | Average balance, surplus funds, deposit frequency | Identify potential deposit opportunities |
Lending history | Existing loan relationships, repayment behaviour, current obligations | Support relevant lending segmentation |
Digital engagement | App activity, product-page interactions, permitted enquiry events | Understand demonstrated interest |
Campaign history | Previous offer exposure, responses, opt-outs | Improve targeting and reduce repeated outreach |
Customer preferences | Preferred channel, communication permissions | Support appropriate engagement |
Data availability does not automatically mean a bank is permitted to use that information for marketing.
Each data source should be assessed for lawful processing, permitted purpose, quality, access restrictions, and applicable consent requirements.
Where third-party or alternative financial information is involved, the bank must also evaluate the conditions under which it was collected and can be reused.
Which Machine Learning Models Are Used for Cross-Sell Propensity Scoring?
Banks can use several statistical and machine learning methods to predict product purchase likelihood.
The appropriate model depends on data availability, interpretability requirements, operational complexity, and predictive performance.
Model | Key advantage | Important consideration |
|---|---|---|
Logistic regression | Interpretable baseline for purchase probability | May require additional features to capture complex relationships |
Decision tree | Easy-to-understand decision rules | Individual trees can be unstable or overfit |
Random forest | Captures nonlinear relationships | Less straightforward to explain than a simple regression |
Gradient boosting | Handles complex tabular data relationships | Requires careful tuning, validation, and monitoring |
Neural networks | Can model complex patterns across large datasets | May require more data, infrastructure, and explainability work |
For many banking applications, logistic regression provides a useful starting point. Gradient-boosted models can then be evaluated to determine whether additional complexity delivers meaningful improvement.
The goal is to select a model that performs reliably in production and supports the bank's governance requirements.
How Is a Propensity Score Calculated?
A logistic regression model estimates purchase probability using the logistic function:
P(purchase) = 1 ÷ (1 + e^(−z))
Where:
z = β₀ + β₁X₁ + β₂X₂ + … + βₙXₙ
Here:
- P(purchase) represents the estimated purchase probability within the defined prediction window.
- X₁, X₂, …, Xₙ represent customer features.
- β₀, β₁, …, βₙ are coefficients learned during training.
The output ranges from 0 to 1.
For example, a score of 0.72 represents a model-estimated 72% purchase probability, provided the output has been appropriately calibrated.
The score is a statistical estimate. It cannot guarantee that an individual customer will purchase the product.
Propensity Scoring vs Next Best Product in Banking
Propensity scoring and next best product recommendations are closely related, but they perform different functions.
Propensity scoring predicts the likelihood of purchasing a specific product. Next best product decisioning uses those predictions alongside eligibility, customer suitability, business value, and engagement rules to select an appropriate offer.
Consider a customer with the following hypothetical scores:
Product | Purchase propensity | Eligibility | Decision |
|---|---|---|---|
Personal loan 82% |
| Not eligible | Exclude |
Credit card | 69% | Eligible | Consider for recommendation |
Fixed deposit | 55% | Eligible | Consider as an alternative |
The personal loan has the highest predicted purchase probability. However, the customer does not satisfy the bank's lending eligibility requirements.
The bank should therefore exclude that offer.
Among the remaining products, it can evaluate customer suitability, expected business value, and engagement context to select the most appropriate recommendation.
This illustrates why next best product banking systems require more than a predictive model.
A Practical Next Best Product Framework
A bank can structure offer selection into four layers:
Layer 1: Propensity
Estimate the probability that the customer will purchase each product.
Layer 2: Eligibility and Suitability
Apply product eligibility, risk policy, regulatory restrictions, and relevant customer suitability requirements.
Layer 3: Business Value
Estimate the expected incremental contribution from each permitted offer after accounting for costs and relevant risks.
Layer 4: Engagement Decision
Determine whether to contact the customer, which channel to use, and when the communication should occur.
A simplified ranking approach might be:
Offer priority = P(purchase) × Expected contribution
This can be a useful initial ranking formula after mandatory exclusions.
However, the bank should use incremental response estimates where available, because customers who are highly likely to purchase may do so without any additional marketing intervention.
The final framework should also account for contact frequency, customer experience, competing offers, and the option to make no offer.
Practical Cross-Sell Propensity Use Cases for Indian Banks
1. Savings Account to Credit Card Cross-Sell
A bank wants to identify savings account customers who may be interested in a credit card.
Potential signals include regular account activity, recurring salary deposits, previous card enquiries, and permitted digital engagement data.
The model identifies customers with stronger predicted purchase intent.
The bank then performs the required eligibility and suitability checks before presenting the offer.
Customers who express interest can be directed towards the appropriate application process.
2. Existing Borrower to Personal Loan Top-Up
A lender may identify borrowers who could be interested in an additional credit facility.
Relevant signals could include existing loan tenure, repayment history, financial behaviour, and previous enquiries.
Propensity modelling helps determine likely interest.
A separate underwriting and affordability assessment determines whether any additional credit can be offered.
The distinction is important because a customer's willingness to borrow does not establish their ability to repay.
3. Savings Account to Fixed Deposit
Customers maintaining regular surplus balances may have greater interest in deposit products.
A bank could use account balance patterns, existing deposit relationships, and historical deposit activity to estimate the likelihood of opening a fixed deposit.
The engagement process should consider the customer's liquidity needs, product terms, and preferences.
4. Current Account to Business Banking Products
For SME banking, cross-sell opportunities might include working capital facilities, business credit cards, or other relevant banking services.
A bank may examine permitted business relationship data and transaction patterns to identify potential demand.
For lending products, these signals can support lead prioritisation, while credit assessment remains a separate controlled process.
5. Credit Card Upgrade and Additional Products
Existing credit card customers may be interested in an upgrade or another eligible banking product.
Banks can use permitted customer engagement information, product holdings, and previous responses to prioritise suitable offers.
Credit limits, fees, product disclosures, and customer approvals must continue to follow the applicable product and regulatory requirements.
How Should Banks Target Pre-Approved Offers?
Pre-approved offer targeting combines customer segmentation, eligibility assessment, and propensity-based prioritisation.
The critical distinction is between a customer who is likely to accept an offer and one who is actually approved under the bank's credit policy.
A bank should establish a controlled process before using pre-approved terminology.
Step 1: Apply Lending Eligibility Rules
Determine which customers satisfy the approved credit policy and product-specific conditions.
Relevant factors may include repayment capacity, existing liabilities, internal risk criteria, and applicable credit information.
Step 2: Score Purchase Propensity
Use the propensity model to rank customers within the eligible population.
This helps the bank determine where marketing resources may be most useful.
Step 3: Define the Offer
Determine the product terms and whether the offer is genuinely approved or remains subject to further checks.
A high purchase propensity score cannot justify representing a conditional offer as an unconditional approval.
Step 4: Check Communication Permissions
Apply applicable consent, channel, preference, and suppression controls.
Step 5: Present Accurate Product Information
The customer should receive clear, appropriate information about the product and any applicable conditions.
For digital lending, banks and other regulated lenders must also consider the Reserve Bank of India's Digital Lending Directions, 2025, including the applicable requirements for creditworthiness assessment, borrower disclosures, and data handling.
Step 6: Monitor Customer Outcomes
Track how customers respond to the offer, whether applications are completed, and whether the campaign creates incremental value.
Where relevant, banks should monitor subsequent credit quality and customer complaints.
How to Build and Deploy a Cross-Sell Propensity Model
A banking propensity model becomes valuable when it can be used consistently within an operational workflow.
The following implementation approach connects modelling with campaign execution.
1. Select a Specific Cross-Sell Opportunity
Start with one product and a clearly defined customer population.
For example, a bank could build a credit card propensity model for existing savings account customers.
Define the purchase event, prediction window, business objective, and baseline performance before development begins.
2. Prepare Historical Data Without Leakage
Create customer-level records using only information available at the relevant prediction date.
Suppose a model uses customer activity recorded after a credit card application to predict that same application.
The model may appear accurate during testing, but it has learned from information that would not have been available when the campaign decision was made.
This is known as data leakage.
To prevent leakage:
- Use clearly defined observation dates.
- Exclude information created after the prediction date.
- Separate training and testing periods chronologically.
- Remove duplicate customer-event records.
- Verify the accuracy of product purchase dates.
- Account for customers whose purchase outcome cannot yet be fully observed.
Historical campaign targeting can also create selection bias. A model trained only on customers who received past offers may learn previous marketing policies rather than general purchase demand.
3. Train an Interpretable Baseline
Begin with a relatively simple model and establish baseline performance.
Compare more complex algorithms only when they produce meaningful improvements.
This helps the bank evaluate whether additional technical complexity is justified.
4. Calibrate and Validate Predictions
A model that ranks customers correctly may still produce poorly calibrated probabilities.
For example, if a model assigns approximately 20% probability to a large group of customers, calibration assessment examines whether roughly 20% of that group actually purchase within the specified period.
Banks should evaluate both ranking performance and probability calibration before using scores in business decisions.
5. Connect Scoring With Eligibility and Campaign Systems
The propensity output must be combined with current eligibility rules, customer preferences, and relevant business exclusions.
The bank can then transfer the approved target audience into its campaign management process using an appropriately governed integration.
Any production workflow should define data access permissions, operational ownership, failure handling, and audit requirements.
6. Launch a Controlled Pilot
A pilot should compare model-driven targeting against a relevant existing targeting approach.
A randomised holdout group helps estimate the additional conversions generated by the campaign.
Banks should avoid judging success only by response rates among high-scoring customers, because some of those customers may have purchased without being contacted.
7. Monitor and Improve
Track changing customer behaviour, model performance, campaign results, and relevant risk indicators.
Retraining should be based on observed performance deterioration, changing data patterns, or an appropriate governance schedule.
Frequent retraining is not automatically better if the available outcome data is incomplete.
How to Measure Cross-Sell Propensity Model Performance
Banks should measure both predictive performance and actual business impact.
A model can rank customers effectively while delivering little incremental campaign value.
Metric | What it measures | Why it matters |
|---|---|---|
Precision at K | Percentage of actual buyers among the top K ranked customers | Measures lead quality within a fixed outreach capacity |
Lift at K | Purchase rate in the top-ranked group relative to the overall baseline | Measures concentration of likely buyers |
ROC-AUC | Ability to rank buyers above non-buyers | Evaluates overall discrimination |
PR-AUC | Precision-recall performance across thresholds | Useful when purchases are relatively rare |
Calibration | Agreement between predicted probabilities and actual outcomes | Supports probability-based decisions |
Incremental conversion lift | Additional conversions attributable to the campaign | Measures true campaign impact |
Cost per incremental acquisition | Campaign cost divided by additional acquisitions | Measures economic efficiency |
Incremental contribution | Additional contribution after relevant costs and risks | Measures business value |
Example: Measuring Lift at K
Suppose a bank has 100,000 customers and an overall historical product purchase rate of 4%.
A propensity model ranks customers by predicted purchase likelihood.
Among the top 10,000 customers, the observed purchase rate is 10%.
The lift is:
Lift at 10% = 10% ÷ 4% = 2.5
The model's top-ranked group has a purchase rate 2.5 times the overall baseline.
This is a hypothetical example used to explain the calculation, not a claimed industry benchmark.
Crucially, lift measures how well the model identifies likely buyers. It does not establish that marketing caused those purchases.
Measuring Incremental Campaign Value
Consider a randomised campaign experiment in which comparable customer groups receive different treatment:
Metric | Targeted group | Holdout group |
|---|---|---|
Customers | 10,000 | 10,000 |
Purchases | 500 | 300 |
Purchase rate | 5% | 3% |
The absolute incremental conversion lift is:
5% − 3% = 2 percentage points
Under valid randomisation and measurement conditions, the campaign generated an estimated two additional purchases per 100 targeted customers.
This distinction matters because banks should optimise for additional business outcomes, rather than simply identifying customers who were already likely to buy.
Common Mistakes Banks Make With Propensity Models
Even technically accurate models can underperform when the surrounding operational process is weak.
Treating High Propensity as Credit Eligibility
A customer with a high likelihood of accepting a personal loan may still be unsuitable or ineligible for additional borrowing.
Propensity modelling must remain separate from credit risk assessment and underwriting decisions.
Optimising Only for Model Accuracy
Overall classification accuracy can be misleading when relatively few customers purchase a product.
For example, a model that predicts every customer will decline could show high accuracy in a dataset with a very low purchase rate.
Banks should focus on ranking quality, calibration, and business outcomes.
Ignoring Customer Contact Fatigue
Repeated outreach can reduce customer responsiveness and create a poor experience.
Campaign frequency, previous refusals, opt-outs, and recent interactions should influence whether another offer is appropriate.
Using Outdated Customer Signals
A customer's financial behaviour and product requirements can change.
Scores should be refreshed at intervals appropriate to the product, prediction window, and available data.
Failing to Measure Incremental Impact
A bank may observe strong conversions among high-propensity customers and conclude that its campaign was successful.
Without a suitable comparison group, it cannot reliably determine how many purchases resulted from the campaign.
Overlooking Model Governance
Banks should document the model's purpose, data inputs, assumptions, validation results, limitations, and ownership.
Monitoring should also cover material shifts in performance, customer complaints, and potential unfair outcomes across relevant customer groups.
Data Privacy and Regulatory Considerations for Indian Banks
Cross-sell propensity modelling involves customer information, making governance an essential part of implementation.
Banks must assess the regulatory requirements applicable to their specific products, data sources, processing activities, and communication channels.
Customer Data and Consent
India's Digital Personal Data Protection Act, 2023 and Digital Personal Data Protection Rules, 2025 establish a framework for personal data processing, with provisions taking effect in phases.
Banks should assess which provisions are currently effective and prepare for subsequent commencement dates rather than assuming every obligation became applicable simultaneously.
Consent and other applicable lawful processing grounds must be considered according to the nature and purpose of the activity.
Customer information collected for one banking purpose should not automatically be assumed available for unrestricted marketing use.
RBI Digital Lending Requirements
Where cross-selling involves digital lending, the applicable RBI framework addresses areas including borrower creditworthiness, disclosures, data collection, usage, sharing, and customer protection.
The RBI Digital Lending Directions, 2025 include specific requirements for digital lending activities within their scope.
These requirements should not be incorrectly described as universal rules for every type of banking cross-sell campaign.
Responsible Customer Communication
Banks should implement applicable communication permissions, opt-out mechanisms, and channel-specific requirements.
For example, promotional voice and messaging campaigns may involve telecom and commercial communication requirements in addition to banking-specific rules.
Model Explainability and Oversight
Internal teams should be able to understand the model's intended purpose, relevant input signals, limitations, and decision boundaries.
Banks should also establish clear accountability for how scores influence customer engagement.
Where a model contributes to regulated lending decisions, the relevant credit risk and model governance requirements must be assessed separately.
A responsible framework combines predictive analytics with controlled decisioning, accurate product communication, and appropriate human oversight.
Turning Propensity Scores Into Customer Engagement With YuVerse
Building a propensity model is one part of the cross-sell process. The bank must also translate its predictions into relevant customer interactions.
This is where YuVerse's Last-Mile AI approach becomes relevant.
YuVerse provides an enterprise AI platform designed to connect intelligence with operational outcomes across customer conversations, documents, decisions, and engagement workflows.
For banking cross-sell programmes, several capabilities can support the stages following customer identification and offer selection.
YuCamp: Orchestrating Cross-Sell Campaigns
After a bank has scored customers and applied its eligibility and marketing rules, the approved audience needs an engagement journey.
YuCamp's AI-powered campaign management platform supports multichannel campaigns across WhatsApp, SMS, email, and voice.
Teams can use its campaign-building capabilities to create engagement journeys and define how follow-up actions should respond to customer outcomes.
For example, an approved credit card campaign could use an initial message, route unanswered interactions towards a subsequent channel, and stop or redirect the journey based on the customer's response.
The bank remains responsible for supplying appropriately selected audiences and enforcing applicable eligibility, consent, and communication controls.
YuCamp's documented campaign orchestration functionality should not be confused with a verified, dedicated cross-sell propensity modelling capability.
YuVin: Personalising Product Communication
A propensity score can identify a relevant customer. The communication must then explain the offer clearly.
YuVin's personalised video messaging supports customer-specific video communications at scale.
Banks can use personalised video workflows for relevant product pitches, customer education, and engagement, using approved customer and offer information.
For example, a customer eligible for a credit card could receive a personalised explanation of the product's relevant benefits and application process.
The objective is to improve communication relevance while ensuring that every statement reflects the customer's actual offer and applicable terms.
YuChat: Supporting Customer Conversations
Customers may have questions after receiving an offer.
YuChat's omnichannel messaging platform supports customer interactions across channels such as WhatsApp, RCS, and SMS.
In an approved cross-sell journey, conversational messaging can help customers ask product questions, express interest, and move towards the appropriate next step.
This provides an engagement layer around the bank's existing product selection and decisioning processes.
Connecting the Broader Banking Journey
Cross-sell propensity models can form one component of a wider customer relationship strategy.
YuVerse's banking and financial services solutions illustrate how customer acquisition, engagement, credit workflows, servicing, and cross-selling can be addressed across the banking lifecycle.
For lending-related cross-sell programmes, credit assessment and risk controls remain distinct from marketing propensity. The bank must validate any proposed connection between its own scoring systems, eligibility engines, and engagement platforms before deployment.
The operational objective is straightforward: turn a relevant customer insight into an appropriate interaction, supported by clear controls and measurable outcomes.
What Should Banks Look for in a Cross-Sell Propensity Solution?
Banks evaluating a cross-sell analytics or next best product solution should assess the complete operational workflow.
A useful evaluation framework includes:
Evaluation area | Key question |
|---|---|
Data readiness | Can the solution use the bank's permitted, relevant customer data? |
Predictive performance | How are ranking quality, calibration, and stability validated? |
Model explainability | Can teams understand the factors influencing scores? |
Eligibility controls | Can product and risk restrictions be applied before outreach? |
Offer decisioning | Can the bank rank suitable offers using business value and customer context? |
Integration | Can approved scores and audiences reach existing banking and campaign systems? |
Engagement | Can the bank execute and monitor customer journeys across relevant channels? |
Measurement | Can incremental conversions and economic contribution be measured? |
Governance | Are data access, monitoring, auditability, and operational responsibilities clearly defined? |
Banks should distinguish between vendors providing predictive model development, next best product decisioning, and campaign execution.
These capabilities may be offered through separate systems, and their integration should be verified rather than assumed.
Frequently Asked Questions
What is a propensity model for cross-sell?
A propensity model for cross-sell predicts the probability that an existing customer will purchase an additional product within a specified period. Banks use these predictions to prioritise customers for relevant offers, subject to eligibility, suitability, and communication requirements.
How is a likelihood to buy model used in banking?
A likelihood to buy model in banking analyses historical customer behaviour, product holdings, transactions, and engagement information to estimate purchase probability. Banks can use the resulting scores to prioritise marketing audiences, support relationship managers, and improve product targeting.
What is the difference between a propensity model and a credit scoring model?
A propensity model estimates the likelihood of a customer purchasing or accepting a product. A credit scoring model estimates credit-related risk, such as the probability of default over a defined period. Both can support lending operations, but they serve different purposes and should not be used interchangeably.
Can banks use propensity scores for pre-approved loan offers?
Yes, propensity scores can help banks prioritise customers for appropriately approved lending offers. However, the score itself does not establish approval. The bank must separately apply its credit policy, assess repayment capacity, and meet the relevant regulatory and disclosure requirements before describing an offer as pre-approved.
How do banks measure the success of cross-sell propensity models?
Banks can measure predictive quality using precision at K, lift, ROC-AUC, PR-AUC, and calibration. They should also assess incremental conversions, customer acquisition costs, campaign contribution, and relevant risk outcomes. Randomised holdout testing can help determine whether campaigns generate additional purchases.
What is the difference between cross-sell propensity and next best product banking?
Cross-sell propensity predicts how likely a customer is to purchase a particular product. Next best product banking uses propensity estimates alongside eligibility, customer suitability, business value, and engagement context to recommend an appropriate product or decide that no offer should be made.
Conclusion
Cross-sell propensity models help banks make more informed decisions about which customers to approach, which products to consider, and where to focus their sales and marketing resources.
Their value depends on more than predictive accuracy.
Banks need reliable customer data, appropriately calibrated models, clear eligibility controls, responsible communication practices, and a measurement framework that distinguishes likely purchases from genuinely incremental conversions.
For Indian banks and NBFCs, the opportunity is to build a more relevant and measurable approach to customer relationship growth.
YuVerse supports the engagement side of this journey through its Last-Mile AI platform, including campaign orchestration, conversational messaging, and personalised customer communication.
Explore how YuVerse can help operationalise your banking customer engagement workflows. Connect with the YuVerse team to discuss how approved customer insights and offers can be translated into relevant, measurable cross-sell journeys.



