Every document upload is a step where people leave. For a large share of your customers the upload is unnecessary, because the answer already exists in sources that can be asked directly.
Not "is a document more reliable than data". Sometimes. The question is which customers need the document at all, and whether your flow can tell before it asks.
Two data points, each confirmed by two independent sources.
Name and date of birth, matched against two sources that do not share an origin.
Name and address, matched the same way.
Sources include credit bureaus, government registries, utilities, telcos and banking data. The independence is the point: two sources that ultimately copy the same file are one source.
Where all four match, the customer uploads nothing. Where they do not, the case steps up to a document or another route, carrying what it already learned.
Sources include credit bureaus, government registries, utilities, telcos and banking data, combined per market.
A synthetic identity is built precisely to pass single-source checks: a real address, a plausible name, a thin but clean history. Requiring two independent confirmations of each data point is what makes it expensive to construct.
Which sources, which combinations, and what counts as sufficient are set per market. Some countries have rich data and this route is broadly usable; others do not. The flow should reflect that rather than pretending the map is uniform.
Most customers upload nothing, which is the drop-off point fixed without touching conversion anywhere else.
Which sources were asked, which matched, and on what.
A single data source is a claim, not a verification. We ask several and require them to agree, which is a different product even when the underlying sources overlap.