Automating Identity Verification / by Sara Yang

Context

As we launched Tala in the Philippines, we had to introduce an identity verification flow for the first time, where users submit photos of their ID & a selfie. In the spirit of MVP, we launched a basic Know Your Customer (KYC) flow to get-to-market.

We soon realized that KYC would be a top barrier to scale in this new market — as both a major drop off point in our activation flow (approximately 50% of users drop off before or after submitting their ID photo & selfie) and as a driver of operational overhead & complexity (hiring projections estimated a 3x increase of headcount to handle volumes for Dec 2018).

PURPOSE

Our goal was to address the operational complexity & better fulfill our “instant” value proposition to our users.

Primary goal: Handle KYC volumes at scale.

  • Automated decisioning should reduce load on the manual review team, to manage our steep growth curve.

  • Success metrics: automation rate, KYC approval rate, cost

Secondary goal: Improve conversion through KYC funnel.

  • Drop off occurs when submissions aren’t reviewed within a window; a faster app experience should improve conversion and increase the number of people who convert to users.

  • Success metrics: decision time, conversion

Defining Success

To kick off the project, I created a brief summarizing the context, goals, and key metrics. For each metric, I estimated targets to keep ourselves accountable. These included primary success metrics and guardrails:

  • Automation rate: % of submissions that are automatically decisioned

  • Approval rate: % of submissions that are approved

  • Cost: blended cost per review

  • Decision time: # of minutes before decisions are returned

  • Conversion: % improvement to the conversion funnel

  • Repayment: % guardrail to ensure KYC quality

Partner Evaluation

We evaluated options to scale our manual review, or automate the process; and decided to evaluate third party companies with automated identity verification solutions. We narrowed to our top three contenders & I led a deeper evaluation process (test integration & data analysis) to select our final partner for the integration. This involved shipping a proof-of-concept integration; then analyzing the data against the metrics we cared about.

Product Design & Delivery

I designed the flow & shipped the integration in close collaboration with my tech lead; then monitored the data in dry run with our credit & business intelligence teams. The dry run allowed us to monitor our success metrics and the integration performance, without putting customer decisions at risk. We soon exited the dry run and turned the integration live for our customers in Philippines and Mexico.

A high-level flow describing the flow with automated review. When a user submits their KYC submission, the majority are sent through automated review while a small portion goes directly to our in-house team as a downsample.

A high-level flow describing the flow with automated review. When a user submits their KYC submission, the majority are sent through automated review while a small portion goes directly to our in-house team as a downsample.

Outcomes

We analyzed the performance against our target success metrics, set at the beginning of the project. Based on actuals, the project met all targets and performed same or better than our in-house manual review.

Some highlights:

  • Automation: Over 30% of decisions were automatically verified (which meant that ⅓ of users received their KYC decision in less than 3 minutes!)

  • Decision time: 95% of submissions were decisioned in under 5 minutes (a 10X reduction)

  • Conversion: the overall cohort showed over 1 pp (absolute) improvement in conversion.

For our users, the processing time beat expectations; for some, the fast process provided mental relief & solidified their perception of Tala as a reliable partner for emergencies.

It’s fast and within the day you can get approved wherever you are ... I was climbing a mountain when I got a confirmation text that I was approved. My expectation was after 1 to 2 days.

I felt relief after … now I have something I can get emergency funds from.
My expectations at that time is that it will take 2 days before I will get approved.

My ID was verified within minutes of uploading.

I was really thankful … it reduced my stress because the process is not too long or tedious.

With these proof points, we extended our contract with our partner. We’re continuing to invest in KYC to increase our automation rate and keep making the onboarding experience an easy, fast process for our users.