The KYC drop-off audit: how to find the leak in remittance onboarding

Every remittance operator can tell you their install cost. Most can tell you their signup rate. Very few can tell you what percentage of people who start verification finish it, and almost none can tell you which specific screen loses them.

That gap is not an oversight. It is structural. Onboarding sits between three teams. Product owns the screens, compliance owns the requirements, and marketing owns the traffic arriving at them. Each measures a different thing, and the number that matters to the business, how many people who wanted to send money were actually able to, belongs to none of them.

The result is predictable. Acquisition budget grows, install cost gets optimised, and a large share of that spend evaporates somewhere between a person deciding to send money and a person being allowed to.

If verification loses half your senders, doubling ad spend doubles the waste, not the revenue.

Why nobody owns this number

Ask a compliance team about verification performance and you will get approval rate: the share of submitted applications that pass. It is a good number and it is not the one costing you money.

Approval rate measures what happens after someone submits. Completion rate measures whether they submit at all. A business can run a 94% approval rate, which looks excellent in any board pack, while 40% of people who started verification never reached the point of being approved or declined.

Approval rate and completion rate are different numbers. A 94% approval rate looks excellent and tells you nothing about the 40% who never submitted anything.

Product teams usually track their own funnel but stop at the handoff to the verification provider. Marketing teams track to signup because that is where the ad platform's reporting ends. The provider reports on what it received. Nobody assembles the sequence end to end, so nobody sees the shape of it.

There is also an uncomfortable incentive. The number is bad on almost every account we have audited. A team that measures it inherits a problem that spans three functions and has no obvious owner. Not measuring it is, in the short term, easier.

What a normal drop-off curve looks like

Below is the shape we see most often. The percentages are patterns across engagements rather than benchmarks, and yours will differ by corridor, document type and provider. What matters is not the numbers but the compounding.

Read the cumulative column rather than the stage column. Each step looks survivable in isolation. Together they mean a business can lose roughly three quarters of everyone who wanted to send before a single transfer occurs.

That is the number acquisition spend is buying into. It is also why the same media plan produces wildly different economics at two companies with identical cost per install.

The five screens where senders actually leave

In almost every audit, the loss concentrates in five places. Four of the five are mechanical rather than regulatory, which is the useful part: they can be fixed in product without reopening a compliance decision.

Document capture on mobile

This is usually the largest single loss point and it is almost always a technical problem dressed up as a user problem. Camera permissions, glare on a laminated card, cropping that cuts a corner, file size limits that reject a photo from an older handset, and passport pages photographed at an angle because the sender is holding the phone in one hand.

Split your completion rate by device tier before doing anything else. On devices released in the last three years, capture usually performs acceptably. On older Android handsets, which represent a large share of the actual sender base in most corridors, it frequently collapses.

Selfie and liveness

Lighting, older cameras and unclear instructions. The recoverable part is the failure state. A liveness check that fails hard and returns the sender to the beginning loses people who would have passed on a second attempt in better light.

Address verification

This is where a compliance decision has a direct and usually unmeasured revenue cost. The accepted document list works well for settled residents and badly for recent arrivals, students and people in shared accommodation, who are disproportionately represented among first-time senders in most corridors.

Nobody is suggesting you relax a requirement. The point is that whoever sets the list should be able to see what each item on it costs in completed transfers, so the decision is informed rather than instinctive.

Manual review limbo

The silent wait. Senders who are approved eventually but have already gone elsewhere. The counterintuitive finding here is that speed matters less than communication. A status message that sets an expectation converts materially better than silence, even when the underlying wait is identical.

Re-entry after failure

What happens when someone fails once. Most flows send them back to the start with no explanation. A specific reason plus a direct return to the failed step recovers a meaningful share, and it is one of the cheapest changes available.

Running the audit in an afternoon

With the events already in place, this takes about half a day. If the events do not exist, instrument them first: that is two to three weeks and it is a prerequisite for everything, not just for this.

1. Pull the funnel from your own events, not the platform. You need account created, verification started, documents submitted, liveness passed, approved and first transfer as distinct events.

2. Split by corridor. A blended number hides everything. Drop-off where senders hold local ID looks nothing like a corridor where most are recent arrivals.

3. Split by device tier. Older Android devices consistently fail capture and liveness at higher rates. If you have never split this, do it first.

4. Split by document type. Passport, national ID and residence permit fail at different rates for reasons that are usually mechanical.

5. Watch ten sessions. Screen recordings from real abandonments teach more in twenty minutes than a month of dashboards. Watch the failures, not the successes.

6. Time the manual review queue at the 50th and 90th percentile. The 90th is the one costing you money.

7. Read the rejection messages your senders actually see. Out loud. Most are written for auditors rather than for a person holding a phone.

If you only do one of these, do step five. Ten session recordings will tell you where the leak is faster than any analysis.

What to fix first

Prioritise by volume multiplied by fixability, not by severity. The most damaging problem is often the hardest to change, and starting there means nothing ships for a quarter.

1. Anything mechanical in document capture. Highest volume, lowest political cost, and it sits entirely within product.

2. Rejection messaging and re-entry. Cheap to fix, immediate effect, no compliance conversation required.

3. Manual review communication. Not the speed, the expectation setting.

4. The accepted document list. Highest potential impact, requires a compliance conversation, so start that conversation early even though the change lands last.

The cheapest fixes are almost never the ones anyone is arguing about in the roadmap meeting.

Measuring whether the fix worked

Hold out a control group where volume allows. Without one you are comparing periods, and remittance has seasonal cycles strong enough to manufacture an improvement that is not there.

Measure through to first transfer rather than stopping at approval. This matters more than it sounds. A change that improves approval rate but adds friction earlier, or delays the first transfer, has usually made the business worse while making the dashboard better.

Re-run the three splits after the fix. Improvements are rarely evenly distributed, and a change that lifts the average while making one corridor worse is worth knowing about before it becomes the corridor you were about to scale.

Four mistakes worth avoiding

Treating it as a compliance problem

The requirements are set by compliance. How they are presented, retried and explained is a design decision, and that is where most of the recoverable loss sits.

Comparing corridors without splitting by document type

Two corridors with identical completion rates can have entirely different underlying problems, and the fix for one will do nothing for the other.

Optimising approval rate instead of completion rate

You can raise approval rate by making verification harder to start. The dashboard improves and the business shrinks.

Scaling acquisition before the funnel is fixed

Every additional unit of spend into a leaking funnel buys a proportionally larger amount of waste. This is the argument for running the audit before a media plan rather than after one, and it is the reason we put it first in every engagement.