Finom · employed product role

AI transaction monitoring

An AI assistant brought transaction context and decision support into a consequential compliance workflow.

Innokenty Bodrov · Product lead

← All work

01 / The task

Analysts needed to assemble the context of a suspicious transaction and decide how to handle it. The product challenge was to make that work faster without losing a clear boundary between automatic processing and expert judgement.

The assistant was part of a wider Finom platform programme spanning data, integrations and AI-assisted operations. The transaction-monitoring result belongs to the assistant; the platform-wide integration result is a separate measure.

02 / My contribution

I led the product delivery with product, engineering and operations teams. My work covered the analyst workflow, data requirements, decision logic, priorities and rollout. The wider programme built four platform products from zero across three cross-functional teams.

This was an employed product role, not a separate client consulting engagement or a claim that I personally built the model.

03 / Reported result

65% of suspicious-transaction monitoring cases were handled fully automatically. The denominator is monitoring cases, not all transactions. Across the separate four-product platform programme, integration time-to-market fell by 40%.

The wider platform programme

Three cross-functional teams built four platform products from zero across data, integrations and developer-facing capabilities. I led product strategy, priorities and delivery in this employed role. The reported 40% reduction concerns integration time-to-market across this wider programme, not the AI assistant or model quality.

What this does not prove

These figures do not state model accuracy, false-negative or false-positive rates, the measurement period, or financial savings. Client data and internal thresholds are not disclosed.

Source and verification

The figures come from my professional record and my earlier self-published case. Finom has not published the 65% or 40% figures in a source linked here; they are not independently verified on this page.

Earlier author publication ↗

Working on a similar problem?

Tell me about the workflow and the decision your team needs to make. I will read your note myself.

Discuss a regulated AI workflow
Back to top