Business Verification
for Account Recovery
A two-phase global research programme to identify the right verification documents for small businesses in key markets — directly reducing account recovery failure rates.
The Problem
Business verification was a critical step in the account recovery process. It relied heavily on website domain verification and US-skewed business registration documents. The problem: many businesses — especially small businesses in developing countries — didn't have a formal website or equivalent documentation.
in Brazil & Indonesia
globally
investigated
This wasn't an abstract product metric. Business owners were losing access to their accounts entirely, forced to start over from scratch — causing severe, real-world harm to their businesses and livelihoods.
What I Set Out to Learn
- Understand how businesses think about verification and which methods they could actually access
- For key markets, determine exactly which business registration documents to prioritise for implementation
- Explore where businesses might be open to preemptive verification — ideally before an account recovery situation arose
- Quantify access rates and comfort levels to provide the precision engineering needed to prioritise work
A Two-Phase Programme
I designed a qual-first, quant-second programme. The generative qualitative phase was essential to uncover factors we hadn't considered — like comfort with sharing — and to discover that users were confusing business vs. individual verification. These insights directly shaped the quantitative instrument.
In-person interviews and site visits across multiple countries. Goal: understand motivations and challenges, map the full range of potential verification documents, and identify opportunities for design improvements or pre-emptive verification paths.
Survey targeting small businesses in key markets. Goal: quantify document access rates by country, measure comfort with sharing each document type, and determine exactly which documents to support in each market.
I identified and recruited small business owners across each target market for in-depth in-person interviews and site visits. The site visits were particularly valuable — observing the physical context of how businesses operated gave important colour to self-reported responses about documentation access.
During interviews, I specifically explored: what documentation businesses typically held, their mental model of "business verification" vs. "personal verification", their experience with the existing recovery flow, and their openness to verifying proactively at moments of value (e.g. opening a Shop, getting rate limits lifted on Business Messaging).
Key qual discovery: A significant proportion of users were confusing business verification with individual identity verification. This was a UI/content problem that could be addressed immediately, independent of documentation changes. This finding was escalated and acted on while the quant work was still in field.
Document range mapped: Domain verification, bank statements, utility bills, business registration certificates, tax registration documents, trade licences — with meaningful variation by country in what was commonly held and what businesses were willing to share.
| Decision | Approach & Rationale |
|---|---|
| Document availability measure | Simple binary yes/no availability question per document type. Chose binary to reduce survey fatigue — the intent was to show proportion available, not nuanced gradations. |
| Comfort-sharing measure | Continuous 5-point bi-polar scale, to capture the full spectrum from discomfort to comfort. Bi-polar rather than unipolar because discomfort was a meaningful signal in its own right. |
| Sampling design | Stratified sample using a market research report to define population targets. Used product data to identify which subgroups to oversample. Oversampled very small businesses in Brazil & Indonesia — our platform under-represented this group vs. market data. |
| DS partnership on data pull | Iterated with Data Science on product variables to accurately match the product's view of business accounts to market reality. Multiple threshold/variable iterations before finalising the frame. |
| Margin of error | 5% MoE chosen given the product decision did not require extreme precision — differences in document access rates needed to be large enough to justify implementation priority. |
| Fielding timing | 2 weeks in field, aligned to typical product usage behaviour (most small businesses in these markets fell into the L10 active-user group). |
| Indonesia resampling | Upon reviewing sample composition for representativeness, Indonesia required an additional +1 week of fielding to achieve sufficient coverage of the smallest business segment. |
Analysis approach: For availability (binary), ran descriptive statistics and chi-square tests for statistical significance across country/segment comparisons. For comfort sharing (continuous), reported means by document type and used t-tests to compare means across document types. Ran ANOVA to test whether business size, country, or other factors predicted willingness to share — found no significant patterns, which simplified the final recommendations.
On sample size: Sample sizes were kept relatively small intentionally — we only needed to detect large effect sizes. Differences in document availability needed to be large enough to justify prioritising one document type over another. Detecting small differences would not have changed the product decision.
Determining Key Markets
This was a global problem that required, in many cases, unscalable country-specific solutions. The product team couldn't build unique solutions for every market — so choosing which countries to focus on was itself a significant research and strategic question.
I began by generating a list of high-priority-need countries based on analysis of verification drop-off rates and compromise rates in product data — the places where failure was most frequent and most painful for users.
I then brought this user-need framing into a collaborative session with PMM and PM to layer in business context. A major factor at the time was the upcoming launch of Meta Verified — which created commercial urgency around certain markets that overlapped (but didn't perfectly align) with user need.
Where I pushed back: Brazil and Indonesia ranked very high on user pain — but neither was included in the initial first wave of Meta Verified launch countries. I successfully advocated for their inclusion in the research scope, making the case that failing to address these markets would leave the highest-severity user harm unaddressed, and that the research investment was warranted given the scale of drop-off there.
This cross-functional framing — user need × business priority — became the shared language that the broader team used when making tradeoff decisions throughout the project.
Document Availability vs. Comfort to Share
The core finding was that availability and comfort were distinct and sometimes misaligned dimensions — a document could be widely available but feel uncomfortable to share, or vice versa. Both dimensions mattered for product decisions.
- Brazil & Indonesia: Identified the specific business registration document types with the highest access rates in each country — enabling targeted engineering decisions rather than a one-size-fits-all approach.
- India: No universally available document type found. The team ultimately decided to hire a local third-party verification firm rather than attempting a self-serve technical solution.
- UI confusion: A significant proportion of users were confusing business verification with individual identity verification — a content/design problem addressable immediately, independent of documentation changes.
Verification Dropoff Reduction
Support for widely available documents in Brazil & Indonesia. UI and content improvements clarifying the business vs. individual verification distinction. Preemptive verification path piloted for Business Messaging rate limit removal.
Bank statements and utility bills identified as a promising lead, but name-matching complexity required further engineering scoping. India verification routed to a local third-party firm given absence of a universal document type.
Responsibilities
Data Science partnership: The data pull for survey sampling required close collaboration with DS. We iterated through multiple versions of the product variables used to identify and classify business accounts, ensuring that the "small business" population we were surveying reflected market reality rather than just our platform's internal categorisation (which under-represented micro-businesses in Brazil and Indonesia).
PMM alignment: Market selection required navigating a real tension between user-pain rankings and Meta Verified launch priorities. I worked directly with PMM to surface this tradeoff explicitly, ensuring the team was making an informed decision rather than defaulting to the commercial roadmap without considering the user impact data.
PM and ENG: After fieldwork concluded, I worked closely with PM to translate quantitative findings (access rates, comfort scores, significance tests) into a format that engineering could directly use for implementation prioritisation — essentially co-writing the product spec for which documents to support first and in which markets.
Qual-to-quant handoff: The qualitative phase surfaced the UI confusion issue early, which I escalated separately and in parallel with the quant work. This meant that one key improvement (the business vs. individual verification distinction in copy) was already in flight before the survey results were in — shortening the overall time-to-impact.