“Can an automated lead evaluation system answer some or all of these questions accurately, in real time?”
The client and what it needed
A global consumer and SME lender:
- with a strong online and mobile presence;
- offering a range of consumer and business digital lending products;
- planning to grow its lead-generator channels across Europe, North America and Asia-Pacific;
- needing to improve portfolio profit and volume;
- focused on sub-prime and near-prime customers.
The problem
The lender struggled to acquire customers profitably through bought leads. Before buying each lead, it wanted an automated system to answer:
Interest and conversion
- How interested is the lead?
- How likely is it to complete the application on the website or app?
Risk and fraud
- How likely is the lead to be accepted?
- To default if the loan is granted?
- To commit fraud, for example by never paying?
- Is it already in the system, or on a blacklist?
Value and pricing
- What are its likely retention and lifetime value?
- Will it borrow again, and how often?
- What limit and terms can safely be offered?
- Is it prime, near-prime or sub-prime?
- Is the lead correctly priced, and will it be profitable?
Scale
- Can the system connect easily to many lead generators in many countries, and still pick out the profitable leads, in real time?
The solution
Using the components of the Zoral Finance Operating System (fOS) below, Zoral built an automated lead evaluation system. It measured and predicted lead behaviour consistently across many countries and regions, in real time. Lead volume was 3,000–5,000 a day at first, rising to 6,000–8,000.
Phase I: core lead evaluation
For each country and region, Zoral used default, fraud and limit models from the Zoral model library. Credit bureau and other data sources were connected to support them.
Input. Leads arrived through a standard interface from several lead generators. More than five were connected quickly. Zoral Decision Engine was also connected to the lender’s loan management system and blacklist.
Output. For each lead, the engine returned scores for each question above, set for each country, region, product and portfolio. The output also showed whether the lead was correctly priced before it was bought. Performance statistics were produced for each lead source, with real-time and periodic dashboards. Inputs and outputs were connected to the lender’s systems through a standard interface.
Phase I results
- 32% lower customer acquisition costs
- 21% more new customers
The lender could now grow online acquisition and cover more countries, while keeping defaults under control. The fully automated system also let it adapt quickly to later changes in markets, sectors and products.
Optimising lead buying exposed another problem. Some leads with good credit characteristics were not converting: they abandoned the online application. Phase II addressed this.
Default rate for sample lead generators before and after Zoral lead evaluation
Phase II: conversion
Three months of lead data were collected: behavioural, conversion, acceptance and default data. Conversion models were then applied for each country, region and product, with the lead’s behaviour on the website or app as a further input.
Zoral Decision Engine workflows chose the product and offer most likely to convert, within the lender’s credit and pricing rules. A/B tests tailored the products offered. Zoral Portal Studio, Zoral Behavior Matrix, Zoral Decision Engine and the model library personalised the customer journey. The conversion score was available when the lead was evaluated, and was adjusted as the customer used the website or app.
The system now:
- measured conversion accurately across countries, regions, products and lead generators;
- found good customers who were unlikely to convert, and tested ways to convert more of them.
Some lead generators charged more for good leads. Without knowing a lead’s likely retention and lifetime value, it was hard to tell whether the lead was worth the price. So retention and lifetime-value models were added, based on the lender’s portfolio, and tuned again after another three months of data from each lead generator.
Phase II results
- 27% higher conversion
- 12% more new customers on top of Phase I
Phase III: retention and lifetime value
The goal was to measure and improve retention and lifetime value further. Zoral added retention and lifetime-value models for each country, through Zoral Decision Engine. No new inputs were needed. The engine used these scores, with the default and fraud scores, to decide whether the price asked for each lead was worth paying.
The system now:
- priced leads more accurately before purchase;
- raised overall accuracy and acceptance rates;
- chose and managed lead generators better.
Phase III results
- 9% more new customers on top of Phases I and II
Results
New customer volume grew by 21% in Phase I, a further 12% in Phase II and a further 9% in Phase III. Together, that is the 48% increase in new customers reported above.
New customer acquisition volume by phase (before = 100)
Baseline
+21%Core lead evaluation
+12%Conversion models, dynamic journey
+9%Retention and LTV models
+44 20 3963 2002