Why onboarding a business customer still takes days

Onboarding a retail customer to a financial product has, for many institutions, become a matter of minutes: capture a document, match a face, check the name against a list, open the account. Onboarding a business customer routinely takes days or weeks. The gap is not explained by effort or by software budgets. It is explained by the nature of the object being verified.

A person is a leaf; a company is a graph

Verifying an individual is a bounded problem. There is a document, a biometric, an address and a set of watchlists. The checks are independent and can run in parallel.

Verifying a business is a traversal. The institution must establish that the entity exists, that the person applying is authorised to act for it, and who ultimately owns or controls it. That last item is the expensive one, because ownership can run through several layers of holding companies, across jurisdictions, before reaching a natural person.

Each hop in that chain may require a different registry, in a different language, with a different filing cadence, some of which are not machine-readable and some of which charge per lookup.

The four checks that consume the time

Entity existence and status. Straightforward in jurisdictions with good open registries; slow where filings are paper-based, delayed, or behind a fee.

Authorised representative. Establishing that the person on the application can bind the company means reading constitutional documents or board resolutions, which are unstructured documents in inconsistent formats.

Beneficial ownership. Tracing to natural persons above a control threshold, then verifying each of them individually. Trusts, nominee arrangements and cross-border layering each add a step. Registry coverage of beneficial ownership varies widely by jurisdiction and has been contested in several of them.

Purpose and expected activity. Understanding what the business does, where its counterparties are, and what transaction pattern is normal, so that later monitoring has a baseline. This is fundamentally an interview, and it resists automation.

What automation actually removes

Vendors in this space are frequently sold as "instant KYB". What good tooling genuinely removes is the retrieval and collation work: pulling registry records, parsing filings, resolving entities across sources, assembling an ownership graph, and screening every node against sanctions and adverse media in one pass. That is real, and it removes a large share of elapsed time.

What it does not remove is the judgement at the end. An analyst still decides whether an unusual structure is legitimate, whether the stated purpose is plausible, and whether the risk is one the institution wants. Automating retrieval turns a five-day process into a one-day process with an hour of review. It does not turn it into a five-minute process, and claims that it does usually mean the review step has been moved somewhere less visible.

The measurement that matters

Time-to-decision is the headline metric and the wrong one to optimise alone. The pairing that describes a functioning onboarding operation is time-to-decision alongside the rate of post-onboarding exits: customers offboarded shortly after approval because something was missed. Improving the first while quietly worsening the second is not an improvement; it is a deferral, and it usually surfaces during an examination.