You're a PM at a neobank. Onboarding conversion has plateaued. How would you improve it?

Before I touch levers, three things change the entire approach, so I'll ask rather than assume.

Before I optimise anything, I want to agree what a good conversion actually is.

Conversion is not one number, it is a product of stage-rates: start sign-up → clear KYC → fund the account → first transaction. Every user who fails to convert is lost at exactly one of those stages, so the funnel itself is my mutually-exclusive, collectively-exhaustive backbone. I find the leaking stage, then decompose the causes at that stage.

I'll read it on three orthogonal axes, because the blended number hides the problem:

• Stage: which step bleeds the most? That is where a fix can actually move the metric.
• Segment: new-to-banking versus switchers, very different trust and familiarity.
• Failure type: did they abandon, a product and motivation problem, or get rejected, a policy and eligibility problem? Those need opposite fixes.

That gives me a grid of stage × segment × failure-type. The next move is to read the data and find the one cell carrying the loss, not to start naming features.

Now I read that grid against the data, because a single conversion number hides where the problem actually is.

Now the second MECE layer, the causes at the leaking stage. For an abandonment problem the reasons are exhaustively one of three buckets, and I'd generate hypotheses across all three rather than jumping to a feature:

Trust / motivation. A new-to-banking user has cleared KYC but has no reason to move real money into an app they don't trust yet. There's no proof it works before they're asked to commit cash, and "what if my money gets stuck" is a real fear for someone using a bank like this for the first time.

Friction. Funding might require a bank transfer with manual IFSC and account entry, a cooling-off period, or a minimum amount. Each is a reason to defer to "I'll do it later," and later never comes. For a new-to-banking user, even a two-minute transfer with unfamiliar fields is enough to stall.

Value gap. Nothing happens before funding. The product is inert until money is in, so there's no aha-moment pulling them across the line, no reason funding feels urgent today rather than next week.

Given the segmentation, friction and motivation are the live buckets: these users chose us and cleared KYC, then stalled at the money step. The value gap is real, but I'd treat it as secondary.

From those buckets, the candidate solutions, judged on impact against effort:

• Lower the funding ask (friction). Let users fund as little as ₹100 instead of a high minimum, and add instant funding via UPI or card so there's no multi-day transfer wait. Attacks the most direct cause. High impact, low effort.
• Create value before funding (value gap). Let the user see their virtual card, set a savings goal, or preview a feature before the funding gate, so funding unlocks something they already want. Medium impact, medium effort.
• Build trust mid-funnel (motivation). Social proof, deposit-protection messaging, and a "fund ₹100 to try it, withdraw anytime" framing that lowers the perceived stakes. Medium impact, low effort.
• Re-engage the abandoners. A well-timed nudge, not spam, with a concrete reason to come back and finish. Low-to-medium impact, low effort, and it recovers people we've already paid to acquire.

What I'd explicitly deprioritise: a full onboarding redesign. It's tempting and it photographs well in a deck, but it's high-effort, slow, and the data says the problem is one specific gate, not the whole flow. With only a quarter, rebuilding everything to fix one stage is how you spend the time and move nothing. I'd also leave KYC alone: it's regulatory, and it's not where the abandonment is.

I'd sequence by impact-over-effort and by learning value, inside the one quarter:

1. Weeks 1 to 3: ship instant UPI funding and the low minimum, A/B against control. Cheapest, attacks the biggest leak, fast to ship. It's the bet most likely to move the number, so it goes first and behind a clean experiment.
2. Weeks 4 to 8: add a pre-funding value moment (virtual card preview), again A/B'd. If funding-friction alone doesn't close the gap, this addresses the motivation half. The week-4 read on step 1 tells me how hard to push here.
3. Weeks 8 onward: layer trust messaging and the re-engagement nudge as fast follows, since they're cheap and additive.

Then I'd measure in three layers, so I get an early read without fooling myself:

• Leading indicators (move within days): funding-step completion rate, time-to-first-funding, and the share funding via instant UPI. These tell me fast whether the fix is landing, well before conversion can.
• Lagging headline (the outcome that matters): first-funded-account conversion and funded-at-30-days. Slower to read, but it's the number we're moving.
• Guardrails (must stay flat or better): fraud and chargeback rate, and D30 retention.

If the leading indicators move but the headline doesn't, my funnel map was wrong, and I'd re-segment rather than ship more.

Before I commit, I'd pressure-test the top fix rather than present it as free:

It could invite fraud. A ₹100 minimum on instant rails is exactly what a fraud ring likes. So I'd pair it with velocity and device checks and watch the fraud guardrail inside the A/B, not just conversion. If fraud climbs, the win is fake.

It could move funding but not retention. Converting someone with a ₹100 deposit doesn't mean they stay. That's why funded-at-30-days, not first-funding, is the headline; if these users churn by day 30, the fix is hollow and I'd rethink the offer rather than scale it.

The test could be underpowered. If funding-abandoners are a modest slice of weekly traffic, I'd size the experiment up front and predefine the minimum detectable effect, so I don't read noise as a result.

So: I'd resist optimising the average, isolate the funding-abandonment leak for new-to-banking users, ship instant low-minimum funding first as the cheapest high-impact move inside the quarter, protect fraud and retention as guardrails, and pressure-test the fix so I don't win the metric and lose the business.