Customer stories

Increasing loan forecasting accuracy through an ML-based sales forecasting system

Designing a Loan Sales Forecasting System for a Non-banking Financial Institution



Forecasting accuracy




Non-banking Finance - Automobile loans



About company

Our client is one of the leading NBFC companies in India and has a thriving consumer loans business. They provide loans for two-wheelers, new & used cars, commercial vehicles, consumer durables, and more. They provide loans for two-wheelers, new & used cars, commercial vehicles, consumer durables, and more at affordable interest rates. 

With over 10 million customers spread across Tier 2 and Tier 3 cities all over India, they have a vision to empower people from underserved markets with easily accessible vehicle and business loans. 


The client wanted an accurate fortnightly forecast of their loan disbursement. They also wanted insights into factors that drive loan disbursements so that they could make informed business decisions to increase disbursement volumes. 

Their existing forecasting model had a prediction accuracy of just 70%, which was not reliable enough to support effective decision-making. Due to this, they lost many potential opportunities and couldn’t plan their business interventions on time. This unpredictability in sales aggravated further during the post-COVID season and left them with trends and patterns that they couldn’t understand.

Following are the brief challenges they faced over a period of time, ever since then.

Inconsistent and unreliable sales predictions: They were not able to predict the loan disbursal rates accurately and on time. As one thing led to another, they couldn’t meet targets effectively. On top of this, they were unable to uncover the factors behind the missed targets as well due to dynamic market conditions due to the inconsistent spikes and falls in their sales data.

Unable to plan resource allocation: Handling a bulk number of customers, they wanted to align their operations better. This includes their marketing efforts, inventory, and workforce planning. But this wasn’t possible as they didn’t know the demand beforehand. Rapid increases in sales numbers around festival seasons made it more difficult. 

No benchmarks on sales figures: Like many financial institutions, they wanted to set realistic sales benchmarks to achieve for the upcoming months. Yet they couldn’t come up with it owing to the huge variation in the predicted numbers and real figures. 

Ineffectual risk mitigation: Not knowing the accurate figures, they couldn’t spot downtrends or possible perils before it happened. So, they weren’t able to implement any risk mitigation strategies they had planned before that.

Roadblocks in bringing strategic measures to life: A clear forecasting of the loan sales number could have helped them implement strategic measures like budgeting, goal setting, targeted marketing, and expansion plans better. Due to the lack of these reliable insights, they couldn’t visualize the future accurately and execute their strategic measures effectively.


Having goals to expand further across the country and add more useful products, they wanted a steady forecasting system that offered more than quantitative insights. They not only wanted to predict their sales accurately and meet their sales targets but also to know numerous factors that contribute to or affect their loan disbursals. 

Overall, they sought to stay unaffected despite the dynamic business landscape and cut-throat competition. This is where datakulture stepped in and changed the scenario by developing and implementing an accurate sales forecast model.

Performing an exploratory data analysis: We started with studying their data (which includes over a million transactions) which was both in structured and unstructured format. All in an extremely granular level, with both needed and irrelevant variables. We conducted an analysis and picked out ongoing trends, seasonal patterns, and important events that affected loan distribution. 

Our goal was to understand recurring patterns, seasonal changes, and significant occurrences in the loan disbursal data, providing a foundation for informed decision-making and strategic planning in loan management.

Building an accurate ML-based forecasting model: We built an ensemble model that combined different machine-learning algorithms to deliver accurate monthly forecasts. They have made sales planning based on the delivered forecast data and found the results to be 90% accurate. Model enhancement is still ongoing as we continue to integrate more business metrics they need for improved tracking and forecasting. 

Setting up an alert mechanism: We set up a checkpoint tracking report that would send alerts by the 15th of every month. This alert is about whether they can meet the targets or not, based on the performance of the first half of the respective month. If in case they are unable to meet targets, they can explore further why’s and how’s to salvage the situation and work on the possibilities. 

A random forest classifier algorithm is used with which we identify the major factors that drive or impact loan disbursals during that period. This option enabled them to be more alert to the changing customer demands and tune their strategies better depending on this bi-monthly checkpoint.

Continuous monitoring through key performance indicators: They wanted to be able to know what factors stop them from achieving their targets. These factors usually change periodically and by knowing them, they could cope better by re-altering their strategies. Additionally, they sought to implement continuous monitoring of specific Key Performance Indicators (KPIs) through the model to ensure ongoing tracking and evaluation of these crucial metrics.


With the accurate forecasting model, our client now can make the right moves required to achieve the target and set achievable benchmarks for the teams. The mid-month checkpoint report further empowers them to ace these goals as they know where their focus should be. They could make perfect interventions before it’s too late and get the numbers back on track or even surpass the predicted numbers. This crystal ball is what every financial institution requires to grow steadily in a marketplace with unpredictable twists and trends.

Let's build your data culture together

Talk to a Datakulture consultant today.

Click to

Get in touch