Simplifying identity verification
I took a stalled registration funnel from diagnosis to provider negotiation and executive approval. We reused the phone number already collected at signup, reduced basic verification to two fields, and changed the provider behind advanced video verification.
What changed?
first-week registration conversion
Registration entry → success; first-week increase. The earlier average was about 23%.
registration to trade
Registration → first trade; approximately 8–9%, with rounding in the recalled figure.
video approval after recording
Advanced video approval after recording, not the full registration journey.
lower verification cost per user
Reported cost reduction exceeded 40% per verification.
The problem & the evidence
Where the problem came from
Funnel analysis showed 14% conversion from registration entry to successful registration and more than 80% drop-off in the verification flow. Basic verification alone had four stages. Reluctance to share identity data was often treated as an unavoidable trust problem; I argued that a faster, simpler process could help earn trust instead of accepting that friction.
- Registration drop-off
- Incomplete verification
- Analytics + Metabase
- How it was measured
- Analytics and Metabase tracked registration entry → successful registration, registration → first trade, verification drop-off, and verification cost. The first week showed a 30% conversion increase; an earlier average was reported at about 23%. Advanced video approval took roughly 2–3 seconds after recording.
- A/B testing
A step-by-step verification flow was compared with a compact national-ID and bank-card form.
AStep-by-step verification
The multi-stage flow asked for the phone again, national ID plus an ID-card photo, and both bank-card number and Sheba (IBAN).
BTwo fields, no repeated phone entry
Ask only for national ID and bank-card number. Reuse the registered phone, derive Sheba from the card, and retrieve the identity photo through the Civil Registration inquiry. Request a phone edit only on ownership mismatch.
The compact form with phone reuse won the comparison. Users could add multiple bank cards for different payment methods. The designer associates removing repeated phone entry with better conversion; no separate, precisely isolated field-level lift is asserted here.
The experience was presented as getting ready to use the service, rather than asking users to complete “verification.”
Measurement context
The first-week result and the earlier average refer to different reporting windows; the exact average window is not recorded here. The 8–9% downstream figure reflects rounding in the designer’s recollection. Reported increases are not stated as percentage-point changes, and the experiment sample size and allocation are not documented. Video approval time is separate from the time needed to enter basic identity details.
Decisions & leading the team
Challenge an accepted source of friction
I found the bottleneck by reviewing the funnels. Rather than accepting identity-sharing reluctance as something every service had to live with, I made the case for a short, clear interaction. Simplicity and speed were the design hypothesis for building confidence, not a separately measured trust score.
Reuse information before requesting more
Phone entry was already mandatory at registration, yet verification asked for it again. I proposed reusing that number, matching it to the national ID and card owner, and asking for an edit only on mismatch. Deposits and withdrawals required the phone, identity, and bank card to belong to the same person.
Move retrieval work into the service
Previously, users entered both card number and Sheba, a long bank identifier they often had to look up in their banking app or on the card. We derived Sheba from the card number and matched ownership instead. For identity, we requested only the national ID and retrieved the photo through Civil Registration rather than asking for an upload. This reduced manual-entry errors; no separate numerical reduction is documented.
Own the service decision end to end
I found and compared providers, negotiated pricing and contract terms, and held several meetings with the CEO and CTO to secure approval despite the engineering cost. At my suggestion, we also explored an in-house AI service; the external provider was less expensive. The new provider changed the liveness-detection method and shortened advanced video approval to about 2–3 seconds after recording.
Treat reliability as part of the product
Provider instability, error rates, weak detection, and fraud were explicit concerns. I worked with Legal on additional contractual protections: provider uptime below 99% would trigger compensation. That was a contractual threshold, not a claim that measured uptime was 99%.
The flow & the design logic
Before: repeated inputs in basic verification
- User
Phone number again
Already collected during registration
- User
National ID + card photo
Enter the ID and upload the identity-card photo
- User
Bank card + Sheba
Look up and enter both numbers manually
After: reuse the phone, request only two fields
- System
Reuse the registered phone
No repeated entry
- User
Two-field setup
National ID + bank cards
- System
Retrieve, not retype
Derive Sheba from the card; retrieve the photo from Civil Registration
- System
Provider validation
Match phone, identity, and card ownership; request a phone edit on mismatch
- Outcome
Basic verification completed
Higher withdrawal limits require advanced verification
Advanced verification: a separate video check
- User
Record video
Advanced identity verification
- System
Provider assessment
Updated liveness method; about 2–3 seconds after recording
- Outcome
Higher withdrawal limits
Available after advanced verification is approved
These diagrams reconstruct the process and design logic; they are not product screenshots.
My direct work
- Diagnosed the funnel and reframed the problem as service architecture.
- Found and evaluated providers, negotiated pricing and contract terms, and compared the external option with an internal AI service.
- Secured CEO and CTO support despite engineering cost, and worked with Legal on reliability protections.
What I learned
Remove work from the user before polishing the work you still ask them to do.
More on the decisions
The screen was not the real bottleneck
Registration converted at 14%, and verification drop-off exceeded 80%. Basic verification had four stages, including requesting a phone number that registration had already collected. Advanced verification had a separate video requirement. I challenged the assumption that this friction was unavoidable because people were reluctant to share identity data.
Change the service, then the flow
I found providers, compared costs, negotiated the commercial terms, and secured executive support through meetings with the CEO and CTO. We explored an internal AI option too; the external provider was cheaper. With Legal, I strengthened the contract around provider reliability and compensation.
Less interface, more certainty
The winning approach collected national ID and bank-card number together and reused the registered phone. The service derived Sheba from the card and retrieved the identity photo through Civil Registration, replacing manual Sheba entry and ID-photo upload. This also reduced entry errors. Users only edited the phone on ownership mismatch. Advanced video approval remained a separate check taking roughly 2–3 seconds after recording.