Early access: manage your cases from Claude. Find out more →
Home›Blog›Set up a compliance platform from your own policy
policy set-uprisk assessmentAIcompliance

Set up a compliance platform from your own policy

12 November 20265 min readFirst Mile Labs

Every regulated firm already has its risk methodology written down. There's the risk assessment, often a spreadsheet your MLRO signed off, and the AML policy your board approved. Then a new compliance platform arrives, and someone re-keys all of it by hand: every factor, weight, band, question and document requirement.

That can take weeks, and every manual step is a chance for the platform to drift from the documents your regulator will ask about. Policy set-up is our answer to that. You upload the documents you already have, and the AI drafts the platform's configuration from them. Your team confirms every setting, and a different MLRO approves the model before anything goes live.

This post explains what policy set-up drafts, how the checks work, and what we keep. Policy set-up is in early access.


Your methodology is already on paper

The Money Laundering Regulations 2017 make sure of that.

  • Regulation 18 requires a firm to identify and assess the risks of money laundering and terrorist financing to which its business is subject. It must take into account risk factors relating to its customers, the countries or geographic areas it operates in, its products or services, its transactions and its delivery channels. And it must keep "an up-to-date record in writing" of the steps it has taken, unless its supervisor tells it in writing that it needn't. (Source: MLR 2017, regulation 18)
  • Regulation 19 requires policies, controls and procedures to manage the risks that assessment identifies. They must be "approved by its senior management", regularly reviewed and updated, and recorded in writing, with any changes made to them. (Source: MLR 2017, regulation 19)

So the thinking has been done, and approved, once. The platform should follow those documents, and you should be able to show that it does.


What the AI drafts

From your risk calculator: the risk model

Upload the calculator you already score customers with: an Excel workbook (.xlsx or .xlsm), a PDF or a Word document, up to 25 MB. The AI reads every sheet, hidden ones included, with their formulas and named ranges. It drafts the risk model: its categories, factors, weights, scores and bands. On a large calculator that can take a couple of minutes.

Where your spreadsheet is ambiguous, it asks you rather than guessing.

From your AML policy: the rest of onboarding

Upload your AML policy and procedures as a PDF or a Word document. For your business customers, the AI drafts:

  • the onboarding questions your policy asks;
  • the documents it requires;
  • the triggers for enhanced due diligence.

For individuals, it drafts questions on the onboarding profiles you choose when you upload the policy. An individual's documents and enhanced due diligence stay with the identity checks and switches on their profile.

Either document can rate countries too: the levels in your country risk list, and the countries in each. The model's own factors read those levels, so moving a country from one level to another moves its score.

Add your policy to the model your calculator drafted, and the two make one model. The calculator sets the scoring, and the policy can't change it. Where the two disagree, the AI doesn't settle it: it asks your team, naming both documents and both values. The policy can still feed a factor from one of its own questions. Upload a policy on its own, and it drafts the risk model too.

Every setting says where it came from

Each setting records its source: a sheet and row in your calculator, or a section of your policy. Each upload records the document's version, the date your board or committee approved it, and who owns it. So the record can always say which approved document a setting came from.


What stays with your team

The AI drafts. People decide what goes live.

See the effect first. Before anyone approves, you can see how your past cases would have scored under the draft.

Confirm every setting. Someone on your team confirms the model one setting at a time, with a record of who confirmed what. If a setting doesn't match your policy, they mark it and say why, in plain language. The AI proposes a correction, showing each changed setting before and after, and nothing changes until your team accepts it.

A second approver. A different MLRO approves the whole model before it goes live. No AI can switch a model on, and neither can any AI tool connected to your workspace.

A model has to be able to rate a case. Our engine rates a case only once enough of its model has a value. A model that could never get there stays a draft, and the screen says how many of its factors have nothing to feed them.

Some factors are a person's judgement. Where your policy leaves a factor to an analyst's judgement, that factor goes to your analysts. They answer it on the case, choosing one of the factor's answers with an optional note, and the case is scored again.

When your policy changes. Upload the new version, and the AI edits the model in place. You see what changed. Settings that haven't changed keep their confirmations, and your team confirms the ones that did.

Every year, again. Every 12 months, someone confirms every setting of the live model again, and an MLRO who confirmed none of them signs the review. Nothing in the model changes, and it keeps working while the review is due. A revised policy that goes live takes the review's place.


What we keep

Never the file. Your documents are read to draft the configuration, and they aren't stored.

We keep the configuration, where each setting came from, who confirmed each setting and who approved the model. We also keep every conversation your team has with the AI to correct a setting, which may quote your document. That record is kept for five years, including after you leave, so you can always show how a setting got there.


Prefer your own AI agent?

Point it at our MCP endpoint, and it drafts through the same tools, reading your workbook on your side. It can't switch a model on either: that stays with your approvers.


What it costs

Policy set-up is an add-on on every plan, in early access. It costs £500 for each document our AI drafts from, charged once, the first time a model built from it goes live. Corrections and re-approvals cost nothing, and a revised document counts as a new one. A draft that never goes live costs nothing, and neither does a document your own agent read.


Start from the policy you already have

We're opening policy set-up to a small number of teams. Request early access, or read how policy set-up works.

Previous
Onboarding is today. Compliance is for a lifetime.

See automated KYB in practice

Book a demo and walk through a live KYB case from application to decision.

Request a demo →