Industries · Financial services

An AI pilot in financial services, with the controls written down before the build

If your financial-services team wants an AI workflow pilot that your risk function can inspect, I scope and build it so every AI step has a governance tier, a named approver and an audit trail, then hand it to your team. The engineering comes from building agent infrastructure at a UK fintech. Legal, compliance and model-risk sign-off stay with your own functions.

This is a good fit if…

  • You run a bank, insurer, lender, payments or wealth business and want one AI workflow (operations, servicing, analyst support) piloted with controls your risk function can inspect.
  • Your compliance or model-risk team has asked engineering to show how AI actions are bounded, logged and reversible, and nobody has written it down yet.
  • A pilot worked in a demo but stalled in internal review because approvals, data lineage or audit evidence were missing.
  • You need senior engineering for an agent or retrieval system in a regulated setting and want to deal with the person doing the work.

Look elsewhere if…

  • You need regulatory advice, a legal opinion or assurance that a system meets FCA, PRA or EU AI Act requirements. That belongs with your compliance function or a regulated adviser.
  • You want automated credit, fraud or suitability decisions with no human approval as a first pilot. I will not build that.
  • Your requirement is turning an agreed policy into staff workflow controls outside a customer-facing financial process. Use AI enablement for regulated operations instead.
  • You only need a contract engineer to work your existing backlog. Use the contract AI engineer route instead.

What you get

Scoped engineering or enablement for regulated workflows

  • One workflow running as a pilot, with every AI step assigned a governance tier: read-only, advise, or act under supervision.
  • A control register linking each step to its approver, log, data source, evaluation and rollback.
  • Evaluation results your model-risk reviewers can rerun themselves, not a demo recording.
  • Audit evidence produced by the system: who asked, what was retrieved, what was proposed, who approved and what happened next.
  • A written go, revise or stop recommendation and a named internal owner for whatever continues.

How it runs

  1. 01

    Workflow and control brief

    You describe the workflow, the data and which internal functions must approve. I return a scoped brief listing controls, approvers, dependencies and exclusions before anything is signed.

  2. 02

    Tiering and data review

    Each AI step is assigned a tier, every data source is traced to its classification, and approval dependencies are listed with owners and realistic lead times.

  3. 03

    Build and evaluate

    I build the pilot in your environment with approval gates, logging and an evaluation set agreed with your reviewers. Changes go through your change process.

  4. 04

    Review and handover

    Your risk and compliance owners review the evidence pack. Your team receives the code, runbook, control register and a go, revise or stop recommendation.

What needs to be in place

  • A named business owner for the workflow and a named engineering owner who will run it afterwards.
  • Access to the compliance, risk or model-risk reviewer for this workflow, with their review steps and lead times known.
  • A data classification for every source involved and an agreed environment where that data may be processed.
  • Approved AI providers or hosting options, or a decision on who approves them and when.

Not included

  • Regulatory, legal or compliance advice, and any statement that a system is compliant.
  • Automated decisions on credit, fraud outcomes, suitability or customer eligibility without human approval.
  • Representing your firm to a regulator.
  • Independent model validation sign-off; that remains with your model-risk function.
  • Out-of-hours production support unless agreed separately in writing.

The situation in most financial-services teams

The appetite for AI is rarely the problem. Operations teams can see where analysts spend hours collating documents, servicing teams know which customer queries are repetitive, and someone has usually built a promising prototype. What stops the work is the question every second-line function asks: what exactly can this system do, who approved it, and how would we know if it went wrong?

If engineering cannot answer that in writing, the pilot stalls in review. If engineering answers it by claiming the system is “compliant”, the pilot fails later and more expensively. The useful middle ground is an engineer who designs the controls into the system, produces the evidence your reviewers need, and leaves the compliance judgement with the people accountable for it.

I designed and built performance-critical components of an AI agent stack for regulated financial services as Principal AI Architect at a UK fintech. This page describes how I apply that work to a scoped pilot for your firm.

What the work involves

Every engagement starts from one workflow, not a platform. Typical candidates are analyst support (preparing a KYC or due-diligence file for a person to review), servicing (drafting responses to routine customer queries for an agent to approve), fraud and alerts triage (summarising the evidence behind an alert, not deciding it), and internal knowledge retrieval over policies and procedures.

For that workflow I do four things.

  • Tier every AI step. I use the published Tiered Governance Model: T0 read-only (summarise, classify, search), T1 advise (the system drafts, a person approves), T2 act under supervision (actions are logged, attributed and reversible), T3 act autonomously. First pilots almost always sit at T0 or T1. The tier determines the controls, not the other way round.
  • Trace the data. Each source is mapped to its classification, its lawful basis as your data protection owner has recorded it, and the environment it may be processed in. Personal data that does not need to reach a model does not.
  • Make approvals mechanical. Where a person must approve, the system cannot proceed without that approval, and the approval is recorded. I follow the Substrate Pattern: permissions and approval boundaries are enforced in the runtime, not stated in a prompt.
  • Evaluate in terms reviewers recognise. An agreed set of real or representative cases, expected outputs, and failure categories that matter to your firm, such as a missed adverse-media hit or an unsupported statement to a customer.

The signature deliverable, illustrated

You receive the working pilot, plus a control register your reviewers can read without reading code. Illustrative example, not taken from a client:

StepTierControlEvidence producedApprover
Retrieve customer file and screening resultsT0Read-only service account scoped to caseRetrieval log with document IDsData owner (one-off)
Draft due-diligence summaryT1Every claim linked to a source documentDraft plus citationsAnalyst, per case
Flag missing documents to customerT1Template-bound message; no free-text sendApproved message and senderTeam lead, per message
Update case statusT2 (later phase)Reversible within 24 hours; daily sample reviewAction log with reviewerOperations risk owner

How acceptance is judged

Acceptance is agreed in the scope before the build. It normally has three parts: the pilot meets the evaluation thresholds your reviewers set on the agreed case set; every control in the register can be demonstrated live; and the evidence pack has been reviewed by the named risk and compliance owners. Their decision to approve, revise or stop is theirs. My acceptance criterion is that they have what they need to make it.

Ownership and handover

The workflow belongs to a named business owner from the start, and the system to a named engineering owner who pairs with me during the build. At handover they hold the code, the runbook, the evaluation harness and the control register. If the pilot proceeds, those owners decide whether to scale it, and any tier change goes back through your own approval process.

When to choose something else

If the need is turning an agreed AI policy into staff workflow controls across operations, see AI enablement for regulated operations. If the pilot is an agent that must take actions in your systems, governed AI agent implementation goes deeper on that build. If you want senior capacity inside your own team, use the contract AI engineer route. Other sectors are compared on the industries overview.

Questions buyers ask

Can you tell us whether the pilot is compliant?

No. I can show exactly how the system behaves: what data it reads, what it can do, who approves each action and what is logged. Your compliance, legal and model-risk functions decide whether that meets your obligations. Keeping that line clear is part of the work, because a pilot that quietly assumes a compliance view tends to fail review late.

What does 'act under supervision' mean in practice?

The system may take an action, such as updating a case record or drafting an outbound message, but every action is logged, attributed to a person and reversible within a defined window. Most first pilots stay at read-only or advise, where a person reviews every output before it has any effect. Moving a step up a tier is a decision for your risk owners, made on evidence.

How is customer data handled during the pilot?

Inside your environment, under your access controls, using the providers your firm has approved. I do not copy data out. Where evaluation needs real cases, we agree a sample, any masking, and retention before I see it. If no approved provider exists yet, the provider decision becomes an explicit dependency in the plan rather than a workaround.

What if internal review rejects the pilot?

Then you have a clear record of why: which control was insufficient, which data use was not approved, or which evaluation result fell short. Stop is a valid outcome. The control register and evaluation harness remain useful for the next attempt, and the scope states up front what is delivered if a review gate is not passed.

Do you work as a contractor or on a project basis?

This page describes scoped delivery: an agreed outcome with acceptance criteria. If you would rather have me join your team under your engineering manager, that is a personal contract and the contract AI engineer route explains it. Either way, I do the work personally and any specialist is disclosed and approved by you first.

Related engagements

Further reading

Describe the workflow and controls

A short, non-confidential description is enough to start. I read every brief personally and reply within two business days, including when the answer is that I am not the right fit.

Step 1 of 2 · The basics