Most startup content is retrospective — a founder explaining a success after it’s already happened, with the messy parts smoothed over. This is different: a live diary of building VyapTrust, India’s first behaviour-based MSME trust score platform, documenting real decisions as they’re being made, including the ones that might turn out wrong.

The problem VyapTrust is trying to solve

Millions of India’s MSMEs (micro, small, and medium enterprises) are effectively invisible to formal credit systems — not because they’re not creditworthy, but because traditional credit scoring depends on formal financial history that many of these businesses simply don’t have in a bank-legible format. VyapTrust’s hypothesis: behavioral and transactional signals (consistency of business activity, payment patterns, digital footprint) can build a meaningful trust signal even without traditional credit history.

Where the build actually is right now

We’re past initial validation (conversations with 40+ MSME owners and 3 potential lending-side partners) and into early MVP architecture decisions. The current focus: defining exactly which behavioral signals are both meaningful and ethically/legally sourceable — this is turning out to be a harder problem than the initial technical architecture, which is a genuinely useful discovery this early.

The decision we’re currently wrestling with

Whether to build the initial trust score model on a narrow set of high-confidence signals (fewer data points, higher certainty per point) or a broader set of lower-confidence signals aggregated together. The narrow approach ships faster and is easier to explain to potential lending partners. The broad approach may produce a more accurate score eventually, but with a longer path to something demoable. As of this writing, we’re leaning narrow for the MVP, broad for the roadmap — get something real in front of lending partners fast, then expand the model based on what they actually need to trust it.

A mistake already made and corrected

Early on, we assumed traditional credit bureau-style categorical scoring (a single number, like a credit score) was the right output format. Conversations with actual lending partners revealed they wanted something closer to a structured risk profile with visible reasoning — not just a score, but why. This changed the entire output design and, subsequently, which signals mattered most to capture cleanly versus which could stay rougher. Cheap lesson to learn now; expensive lesson to learn after building the wrong output format at scale.

Why this is being documented publicly

Partly accountability — writing it down forces clearer thinking than keeping it in my head. Partly because most founders never see what a real, in-progress, imperfect build process actually looks like — every case study they read is already resolved and clean. This is intentionally not that.

What’s next

Building the narrow-signal MVP model, getting it in front of 2-3 lending partners for structured feedback, and — the part I’m least certain about — figuring out the right incentive structure for MSMEs to actually opt into sharing behavioral data in the first place. That’s arguably the harder problem than the scoring model itself, and it’s the next thing this diary will cover in depth.

Want an honest read on where your project actually stands? Send me a few lines about what you are building. Free 30-minute call, no pitch, no obligation. Available for founders in India and Australia. Book a Free Call →

Frequently asked questions