Less time assembling a KYC review, more time on the judgement

The work

A periodic review of an existing business customer is mostly assembly. The analyst collects the documents that the review requires, compares them with the commercial register and the screening results, notes what is missing or does not match, and only then applies judgement to what remains. The assembly is the same for every file and consumes most of the time; the judgement is where the analyst's experience matters and where supervisors expect it to be applied.

The proposed scope is deliberately narrow: periodic reviews of existing business customers, prepared from the documents and screening results you supply. Onboarding decisions, screening itself and anything touching live records stay outside the pilot.

The proposed workflow

The package is built by a network of Agents: a Lead Agent holds the review procedure, and an Agent per step reads one source or compares two. Steps that are pure retrieval are carried out without any model involvement; a model is consulted only where a comparison or a judgement is required, and that judgement is shown with its evidence, never silently applied.

A sample review package

Illustrative sample on a synthetic case, not a measured result

The case below is a synthetic review from our benchmark environment, prepared in the format your analysts would receive. It illustrates the output; it is not a measured product result.

Where this was exercised

The sample comes from our KYC benchmark environment: a case file service with eleven tools, from document retrieval and register lookup to sanctions and PEP screening, and eight synthetic customer files covering a clean file, a sanctioned owner, a PEP linked owner, a document to register mismatch and stale incorporation documents. Every system under test receives the same files and tools, and a deterministic grader scores the final state. Realistic in its procedures, synthetic in its data.

What the pilot measures

• Analyst minutes to prepare a file and minutes to review the prepared package, against your current process on the same agreed cases.

• Evidence completeness: required documents found, missing evidence detected, and cases that needed no further lookup by the analyst.

• Errors, corrections and appropriate escalations, with critical errors reported separately.

• Total cost per case meeting the acceptance criteria, including review, software, infrastructure and operating support.

• Case data transmitted outside the agreed environment, measured in bytes per case.

Released staff capacity is reported apart from cash savings; the two are not the same number. Critical errors are counted and reported separately from everything else.

How the pilot earns a go

Before testing begins, the institution and Eigen Networks agree acceptance thresholds for review time, evidence completeness, critical errors and appropriate escalation. Progression depends on meeting the agreed quality requirements and demonstrating operational value. The numbers stay open until they are agreed with you.

Who evaluates

Your reviewers judge every prepared case against your policy. Eigen Networks sees the verdicts, not the other way round.

What counts as a critical error

A wrong figure or a missing exception that a reviewer would have had to catch, or a referral missed where your policy requires one.

When testing pauses

On any critical error, until its cause is explained in writing and the fix is agreed. The pause is part of the method, not a failure of it.

Data handling and human approval

• Every component runs inside your environment, on premise or in your private cloud: the planning of the workflow, the coordination of the Agents, the models that execute the steps, the tools that read your systems, and the records of what happened.

• No case data is sent to an external model service. Any external dependency would be named in the scoping document and approved by you before a single case is processed; the current configuration is listed in the scoping document.

• Each step that reads a system leaves a receipt: which service was called with which arguments, whether it succeeded, and what it returned. Each figure in a prepared case points to its receipt. Run records with the full prompts and replies are stored inside your environment and only there.

• The Agents prepare; people approve. The pilot changes nothing in live records and communicates nothing to customers. Decisions remain subject to your existing controls.

What participation requires

Duration

Eight weeks: two for scoping and integration, five for the evaluation, one for the report.

Commercial basis

Fixed fee pilot, figure agreed in scoping. No usage based charges during the pilot.

Case sample

30 to 50 historical periodic reviews with their original documents, register extracts and screening results, supplied as files, pseudonymised where your policy requires it.

Integration boundary

Files and read only lookups only. No write access to any system, no live customer records, no customer communication.

Your roles

An operational owner from KYC operations, two reviewing analysts, one IT contact for the environment.

Our roles

Deployment in your environment, the review procedure configured with your policy, the evaluation report.

Proposed next step: select one workflow, an operational owner and an agreed case sample for a controlled evaluation. Run initially without changes to live records, and agree acceptance criteria before testing begins. The resulting report shows measured benefits, observed failures, remaining dependencies and the work required before production deployment could be considered.

Contact: contact@eigennetworks.com

Eigen Networks UG (haftungsbeschränkt), München. Eigen One, Eigen N and EigenBrain are products of Eigen Networks. No figure in this brief is a measured result unless it is labelled as one.