Client name withheld under a non-disclosure agreement.
The challenge
The company offers working capital loans to small businesses in several Southeast Asian markets. Marketing campaigns brought in thousands of sign-ups a month, but the sales team could only call a fraction of them. Calls went out in the order the sign-ups arrived, so a sole trader with no trading history got the same attention as an established business ready to borrow.
The marketing team also had no clear picture of which segments converted. Campaigns targeted broad categories such as retail or logistics, and the best customers were discovered by accident rather than by design.
What we delivered
- Predictive Lead Intelligence & Scoring. A machine learning model that scores each new sign-up on its likelihood to complete an application and be approved, and on its expected loan value. Scores update as applicants add documents, and the sales queue is ordered by score rather than by time of arrival.
- Intelligent Audience Discovery & Profiling. Segments built from behavioral and company data on past customers, showing which business types, sizes and channels produce the most valuable borrowers. The segments feed campaign targeting and refresh monthly.
We trained the models on two years of sign-up and loan history and checked that the scores held up in each market and each customer segment. Sales ran the scored queue alongside the old one for four weeks before switching over.
Timeline
- Weeks 1 to 3: data audit, definition of a qualified lead and model design with the sales and risk teams
- Weeks 4 to 8: scoring model, segmentation and integration into the CRM
- Weeks 9 to 12: parallel run with the sales team, model tuning and handover
Outcomes
- 20 to 30% more funded loans per sales hour
- Sales now speak to high-scoring applicants within a day of sign-up
- Marketing targets segments with a known conversion profile instead of broad industry categories
