Bringing on a new counterparty, whether a customer who has to be credit-checked, a vendor who has to be set up for payment, or a business client who triggers know-your-customer and know-your-business checks, is a document-collection and verification process, and it is slow for a reason that can be measured. Fenergo's 2025 survey of financial institutions found that seventy percent had lost clients in the prior year because onboarding was inefficient, up from forty-eight percent in 2023, with onboarding abandonment running near ten percent. Those clients were not lost because they failed the checks. They were lost because the checks took too long to finish, and most of that elapsed time has nothing to do with the rules being applied.
Where the elapsed time actually goes
Fenergo's earlier research, reported by Corporate Compliance Insights, put the average cost of a single corporate know-your-customer review near $2,598 and the time to complete one at about ninety-five days. Ninety-five days is not ninety-five days of analyst attention. A 2015 Forrester Consulting study commissioned by Fenergo, covering thirteen corporate and investment banks, found that fully manual client onboarding ran anywhere from two to thirty-four weeks, and the width of that range is the tell. A review that takes two weeks and one that takes thirty-four are applying the same regulatory standard to the same kind of entity. What separates them is how many times the counterparty had to be contacted again, how many documents were requested a second or third time, and how long each round trip sat in someone's inbox before it moved.
If you watch a single onboarding from the inside, the calendar fills with waiting rather than deciding. A request goes out for certificates of incorporation, ownership structure, and proof of address. Some arrive in the wrong format, some are missing a signature, some are for the wrong legal entity in a group. Each gap triggers another email to the counterparty, and each email resets the clock by however long that counterparty takes to respond. The same incorporation document is then rekeyed into the onboarding system, the screening tool, and the system of record, by hand, because none of those systems talks to the others. The actual verification, comparing the entity and its beneficial owners against the required sources, is fast. The gathering and the rekeying are what consume the months.
Collect once, then stop asking
The first change in the redesign is to collect each document exactly once and never ask for it again. A counterparty uploads its incorporation papers, ownership chart, and identity documents through a single intake. From that point the agent works from what was submitted rather than reopening the request. Where a document is missing or malformed, the agent flags the specific defect, the missing page, the unsigned section, the entity-name mismatch, in one consolidated request, so the counterparty answers once instead of being pulled into five sequential rounds. The number of times you contact the counterparty is the largest single driver of elapsed time, because every contact carries the counterparty's own response latency, which you do not control.
Extract and validate the entity and ownership data
Once documents are in, the agent reads them and pulls the structured facts: legal name, registration number, registered address, the ownership chart, and the names and stakes of beneficial owners. Onboarding documents are not uniform. A certificate of good standing from one jurisdiction looks nothing like another, and ownership can be disclosed in a board resolution, a shareholder register, or a free-text letter. The variability is real, and handling it is a known problem rather than a blocking one, as we describe in the long tail of document variability. The agent extracts the fields, checks them for internal consistency, the name on the certificate matching the name on the ownership chart, the registration number being well-formed for its jurisdiction, and normalizes them into a single structured record instead of three hand-typed copies.
Check against the required sources, route only real exceptions
With a clean record, the agent runs the checks the policy requires: confirming the entity against the relevant registry, screening the beneficial owners against the lists the institution is obligated to screen, and reconciling the declared ownership against what the sources return. Most counterparties clear cleanly, and the agent assembles the supporting evidence, which source said what, with what timestamp, so a human can confirm it in minutes rather than reassembling it from scratch. The cases that do not clear, a name that partially matches a screening list, an ownership chain that does not reconcile, a registry record that contradicts the submitted document, are routed to a person as a ranked exception queue with the conflict already laid out. The verification decision stays with a human throughout. The agent's job is to do the mechanical gathering, extraction, and checking, and to hand the analyst a short list of the cases that genuinely need judgment, which is the same division of labor we walk through in rebuilding the month-end close.
Write the result back where it has to live
A cleared counterparty has to exist in the systems that act on it: the ERP vendor master, the customer record, the screening tool's case file. This is usually the step where automation projects quietly fail, because the system of record has no usable interface and the institution falls back to manual entry, reintroducing the rekeying the redesign was meant to remove. The agent writes the validated record into those systems the way a trained person would, a problem we treat directly in writing back into a system with no usable API, so the structured record produced once at intake is the same record that ends up in every downstream system, entered without a person retyping it.
A reasonable counter, answered
A reasonable counter is that compliance teams are slow on purpose, that the deliberate pace is a control against waving through a counterparty that should have been stopped, and that compressing the timeline trades away rigor. There is something to this, and the verification standard is not where the time should be cut. But the ninety-five-day average and the two-to-thirty-four-week range are not the cost of rigor. They are the cost of requesting the same document three times and typing it into three systems by hand, none of which adds a single check. The rigor lives in the decision, and the redesign leaves that decision with a person, who now reaches it with the evidence assembled rather than spending weeks waiting for it to arrive. The same survey that recorded rising client losses recorded the use of AI tools in these checks climbing from forty-two percent to eighty-two percent year over year, and LexisNexis put US and Canada financial-crime compliance cost near $61 billion in 2024, so the pressure to reclaim the gathering time is already moving the market. The pattern generalizes well beyond financial services: any time you onboard a customer or vendor by collecting a file, validating it, and entering it into a system, the same redesign applies, and we describe what an engagement built on this division of labor looks like in what an aligned AI engagement actually looks like.