Building the MCA Credit Model for a Fleet Fintech

Pratham joined Indus Insights two months before the MCA project surfaced, and when it did, it was still an idea. The client, a US-based fleet card fintech with a network of over 50,000 fuel stations, had noticed that its fueling data gave it a view of merchant health that no bank possessed. The question was whether that edge was enough to build a lending-adjacent product around.

Feasibility and Cohort Design

Pratham's first task was to determine whether there was a real customer for the product. He designed a questionnaire aimed at identifying two things: which fuel station owners were structurally likely to need a cash injection, and what they would actually use the money for. He embedded the survey into the client's existing sales rep onboarding calls, sat in on a subset of interviews himself, and collected responses from roughly 5,000 potential customers.

The research surfaced a finding that changed the targeting model. The team had assumed demand would come from owners wanting to open new pumps. What they found instead was that a significant cohort wanted capital to add convenience stores or motels to existing stations, because truckers were bypassing fuel-only stops. That insight reshaped the cohort definition and became the lead finding in the feasibility presentation.

Pratham built an eight-slide deck structured as a narrative, with a supporting Excel model carrying the revenue forecasts and cohort analysis. He presented it solo to the VP of Market Expansion and the Sales President, fielded an hour and a half of questions on cohort selection and geographic patterns, and walked away with a green light to pilot.

Credit Model Design

Once the three-month pilot showed the product was gaining traction, Pratham moved into building the credit model that would govern advances at scale. He worked directly with the client three to four times a week for several months, iterating on the model in structured calls and async follow-ups.

The core decision was to anchor the model on the client's own gallon-spend data, not on external credit variables. The logic was straightforward: the client already knew it could recover advances through fuel transaction deductions, so the primary question was whether a station's volume was reliable enough to support repayment. External data from the Dun and Bradstreet portal was used as a secondary layer, with three hard elimination criteria: pending legal action, existing MCA arrangements with other providers, and prior loan defaults.

For stations that cleared those gates, D and B variables including Paydex score, financial stress score, failure score, and EBITDA margin were used to calibrate the margin rate rather than the lending decision itself. The output was a calculator built first in Excel, then formalized in Salesforce, that a salesperson could run with a customer's account number to generate an advance offer.

Segmentation and Recovery

Three months after the model went live, Pratham identified a new problem. Some stations with outstanding advances were steering cardholders toward cash payments to avoid the transaction deductions the client used to recover funds. He built a volume-tracking model that compared each station's daily gallon output against prior-year same-day and month-on-month baselines. When volumes dropped substantially, the model flagged the station as a potential recovery risk, and the team responded by routing discount incentives to cardholders at that location to drive volume back.

He also segmented the full advance portfolio into three tiers: high-value customers who repaid systematically and re-advanced repeatedly; medium-value customers with seasonal patterns but consistent historical behavior; and grade-three customers showing signs of gaming the repayment structure. The sales team was redirected to prioritize grade-one and grade-two customers and stop renewing advances to grade-three accounts.

By the third quarter of the product's operation, the portfolio had crossed 1,000 advances at an average ticket size of $110,000, and bad debt had fallen 16% quarter on quarter.