A wealth product asks a new user for PAN, Aadhaar based verification, a selfie video, bank proof, nominee details, and an FATCA declaration before showing them a single screen of value. Completion sits near 35 percent. The compliance team is blamed. The compliance team did not choose the order of those screens.
Regulated onboarding in India carries genuine obligations: customer due diligence under PMLA, the RBI Master Direction on KYC, and since the DPDP Act, an explicit consent and purpose limitation layer on top. None of those rules say collect everything up front.
Sequence By Risk, Not By Form Length
The useful question for each field is what it unlocks and what it costs to ask. Split onboarding into tiers where each tier grants real capability.
- Identity minimum. Enough to create an account and browse. Often a verified mobile number and a name. Seconds, not minutes.
- Transaction tier. Full CDD before money moves. This is where PAN validation, the Aadhaar based flow and liveness belong, gated behind the customer’s first genuine intent to transact.
- Enhanced tier. Higher limits, PEP screening depth, source of funds. Triggered by behaviour or thresholds, not by signup.
Products that gate the video KYC step behind the first deposit intent rather than at signup routinely see the funnel move 15 to 25 points, without a single control being weakened. Nothing was skipped. The ask simply arrived when the customer had a reason to say yes.
Every Step Fails, So Design The Failure
The single biggest source of abandonment is a dead end. OCR misreads a document, a name match returns 0.81 against a 0.85 threshold, the government API returns a 5xx during a maintenance window, the selfie is taken in bad light. In a naive implementation each of these produces the same red screen: verification failed.
Treat each as a distinct state with its own recovery path. A soft name mismatch goes to manual review with a promise of a decision in 24 hours, not a rejection. An upstream outage should retry with backoff and tell the customer their application is saved. A poor selfie should give one specific instruction, face the light, remove glasses, rather than a generic retry.
Manual review deserves the same product attention as the automated path. If a reviewer needs to open four systems to clear one case, the queue grows, the SLA slips, and the drop-off simply moves later in the funnel.
Consent Is Data, Not A Checkbox
Under DPDP, consent needs to be specific, informed, and withdrawable, and you should be able to show what a user agreed to and when. That is a table, not a boolean on the user record: purpose, version of the notice, timestamp, channel, and the withdrawal event if it happens. Bundling credit bureau access, marketing, and account servicing into one tick is the pattern most likely to fail review.
The engineering payoff is that a proper consent ledger also answers product questions. Which purposes get declined most, and does declining one break the rest of the journey?
The measurable targets are worth stating: verification decision under 90 seconds for the automated path, manual review cleared within one business day, and a re-entry link that restores a half finished application for at least 30 days.
At Invexa, we treat onboarding as one system where the compliance rules and the funnel are designed together, because the alternative is a product that is both slower and no safer.