Where KYC onboarding actually loses customers — and what fixes it

Most onboarding teams already know the uncomfortable truth. Every extra step in a KYC flow loses some share of applicants. The only real questions are how many, and where.

The scale is hard to ignore. In 2022, Sapio Research surveyed 7,600 consumers across 14 European countries about financial onboarding. Some 68% had abandoned a financial application in the previous year.

Institutions feel it too. A 2025 global survey polled 600 senior decision-makers at banks, asset managers and fund administrators. Fully 70% said they had lost clients to slow or inefficient onboarding in the past year. These studies use different markets and methods, so treat exact figures as directional.

The pattern behind the numbers is consistent: friction compounds, and applicants walk. For product and growth teams, this is the funnel nobody owns end to end. Compliance owns the checks, engineering owns the forms, and the drop-off falls in between.

Where the drop-off tends to cluster

Abandonment rarely has a single cause. In the funnels we see, it clusters around five familiar friction points.

• Manual forms that ask applicants to retype what is already printed on their ID
• Blurry document photos that trigger retake after retake
• Manual review queues that leave applicants waiting hours or days
• Requests for more personal data than the product plainly needs
• False rejections that quietly turn genuine customers away

Each one has a practical fix. None of the fixes requires weakening compliance.

Retyping the document, field by field

Typing a name, address, birth date and licence number is pure duplication. All of it already sits on the document the applicant is about to photograph. On a phone keyboard, every field is another chance to mistype or quit.

OCR autofill removes the step almost entirely. The applicant photographs the ID, and trained document models extract the fields in seconds. People confirm instead of typing, which tends to be faster and produces cleaner data. Our features page covers how this works across licence and passport formats.

Blurry photos and the retake loop

Document capture usually fails for mundane reasons. Glare, shadow, camera shake and cropped edges are the usual suspects. Each failed attempt asks the user to try again, and patience runs out fast.

Two things help here. Real-time capture guidance rejects a bad frame before the photo is taken, not after. Automatic image correction then rescues borderline shots through de-skewing, glare removal and sharpening. Fewer retakes means fewer moments where leaving feels easier than persisting.

The manual review queue

When every application lands in a human queue, speed depends on staffing. An applicant who applied in two minutes may wait a day for an answer. Some will have opened an account elsewhere by then.

Queues create a second problem too: inconsistent decisions. Two reviewers can reach different outcomes on the same borderline document.

Automated face matching with liveness detection can clear the obvious cases on the spot. The system matches the document photo to a live selfie and confirms a real person is present. Human reviewers then handle only the genuine grey zone, where their judgement adds value. Our liveness product page explains the approach, and our deepfakes article covers how it holds up against spoofing.

Asking for more than the product needs

Over-collection is a quiet conversion killer. In the European study above, 21% of abandoners blamed the amount of personal information requested. Privacy-conscious applicants notice when a form wants more than the product could plausibly require.

Data minimisation is the fix, and it doubles as good compliance. Collect what your risk assessment justifies, and nothing more.

Mobile driver licences make this concrete through selective disclosure. A verifier requests specific attributes, not the whole document. We explain the mechanics in our selective disclosure article and the basics in what is an mDL.

False rejections: the invisible loss

Some genuine applicants fail verification through no fault of their own. Worn documents, poor lighting or unusual capture conditions can all trip an automated check. These losses rarely surface as complaints, because most people simply leave.

Independent testing suggests the problem is widespread. The US Department of Homeland Security tested seven document validation systems in its 2025 Remote Identity Validation Rally. Only one met the target for false rejections of genuine documents. Accuracy clearly varies, so ask any provider for their measured false rejection rate.

Tuning matters as much as the engine. Rejection thresholds should reflect the risk of the product, not one global default. A soft fail can route to manual review rather than a hard decline. Genuine customers stay in the funnel, and fraud still gets stopped.

Conversion versus compliance is a threshold decision

It is tempting to frame all this as growth versus regulation. In practice it is rarely either/or. The same verification stack can run strict for high-risk products and lighter for low-risk ones.

What changes is the threshold, the escalation path and the data requested — not the obligation. Regulators generally expect risk-based controls, and a well-tuned funnel is exactly that. Treating every applicant as maximum-risk is not caution; it is an untested default.

Measure before you optimise

You cannot fix a funnel you cannot see. Instrument every step: form start, document capture, selfie, decision and funding. Then watch where abandonment clusters, and segment by device and document type.

• Step-level completion rates, not just end-to-end conversion
• Median time per step, plus the long tail where people give up
• Retake counts on document and selfie capture
• Share of applications routed to manual review, and queue wait times
• Support contacts or appeals that follow a decline

Cohort the data by acquisition channel as well. Paid traffic often behaves differently from organic, which changes the economics of each fix.

Small losses compound quietly across a five-step flow. The European study found applicants gave up after about 19 minutes on average. The window to earn a completed application is short.

Start with one friction point

If onboarding drop-off is on your roadmap, begin with measurement and a single fix. OCR autofill and capture guidance are usually the fastest wins. From there, automating the clear cases tends to deliver the biggest queue reduction.

Our KYC onboarding use case shows how these pieces fit together in one flow. Want to see it against your own funnel? Book a demo and bring your drop-off data.