This website uses cookies

Read our Privacy policy and Terms of use for more information.

Most AI adoption stalls in the same place: not at the demo, but at the point where someone has to decide whether a multi-step output is actually correct, compliant and safe to rely on. That decision is our specialism.

1 · AI workflow design & evaluation

For long-horizon business, finance and GRC workflows — the multi-step processes where an error compounds quietly rather than failing loudly.

What this looks like in practice

  • Mapping and documenting the workflow as it really runs, not as the process document claims.

  • Designing the AI-assisted version: where the model helps, where it must not act alone, and where a human decision belongs.

  • Building the evaluation — the rubric or reward signal that defines what "good output" means for your workflow, so quality becomes testable rather than felt.

  • Writing the governance layer: disclosure points, audit trail, escalation, and the pre-launch gate.

Who it fits:

Teams scaling AI who have the tooling and lack the bar; finance and operations functions where compliance isn't optional; anyone who has to defend an AI-assisted output to an auditor, a regulator or a board.

2 · Governed MCP integration & evaluation

Before you connect an AI agent to a system that can move money, change a record or send something on your behalf.

Tool access is the point where AI stops being a drafting aid and starts being something that acts. Most of the risk arrives in that transition, and most checklists were written for the version before it.

We run a readiness audit against fifteen controls covering: what the agent is allowed to reach and what it is denied by default · how each connected server handles your inputs · credential scope and rotation · defence against instructions smuggled in through tool output · staged enablement from read-only to write · human approval on anything that changes state · the audit trail · and a kill-switch that has actually been tested.

Six of the fifteen are must-pass. Any one of them failing is a stop, whatever the rest of the score looks like. A setup can score well and still be a stop — and if that sounds harsh, the rule is doing its job.

What you get

  • The findings, control by control, with the evidence we sighted rather than the assurances we were given.

  • A decisive GO / GO-WITH-CONDITIONS / NOT-READY — not a traffic-light dashboard you have to interpret.

  • A remediation list in priority order, with the cheap high-impact items named first. In our own audit, three actions cleared the critical failure in about forty minutes.

Three ways in

Readiness audit

Fixed fee, about a week. We run the checklist on your proposed setup and hand over the findings, the verdict and the remediation list. The audit is the deliverable — there is no upsell built into it

Build & evaluate

We stand up least-privilege access and approval gates, then evaluate the agent on your real workflows against a rubric written for your definition of a good output

Governed operations

We hold the register, re-audit when something changes, and add new tools under change control

Who it fits:

Finance and operations teams connecting agents to ledgers, banking or client records · anyone who has to explain an AI-assisted action to an auditor, a regulator or a board · teams who have the tooling and haven't set the bar.

Where this instrument stands, stated plainly. It has been run live once — on our own setup, where it returned a NO-GO. We are not going to call that "proven in production", because it isn't yet. Read what it found →

Or see the fifteen points in full — the checklist is published, free, and yours to run yourself.

Book a first conversation · or start smaller: Get the free scope check mailto:[email protected]?subject=Scope%20check

Scoped and priced on a first conversation — we don't publish a rate for work whose shape we haven't seen.

3 · Governed AI talent & data partnerships

Two bottlenecks come up in nearly every conversation: vetting AI talent well enough to trust the hires, and finding data or agentic-workflow partners whose pipelines would survive an audit.

We handle the governance-fit side of that conversation — free, and separate from anything transactional. Where a route we use ourselves genuinely fits, we introduce it and disclose the relationship plainly.

We only list what we've used, and we don't quote other people's terms secondhand.

4 · Finance operations

Delivered through Striking Accounts: bookkeeping-to-advisory, systematised and AI-assisted, with a qualified finance function accountable for the output. Suited to small businesses, startups and independent professionals who want their numbers to be decision-grade rather than merely filed.

Finance and accounting advice is delivered inside a formal professional engagement, with its own scope and terms. Nothing on this website is financial, legal or professional advice.

5 · Education & knowledge products

The teaching that comes out of the work, published through AI the Brief — the honest version of what AI does and doesn't change for a business.

A structured governed-AI setup workshop and a longer-form book are in development. They'll be announced when they're ready and tested, not before. If you'd like to hear when: [email protected].

How an engagement runs

1 · A real conversation. What you're actually trying to do, and whether we're the right people. "Not a fit" is a perfectly good outcome and we'll say it early.

2 · Vet. Before we recommend a tool, a partner or an approach, we check it against your data policy, your risk position and your budget.

3 · Build, with disclosure. AI-assisted where that's genuinely better, human-decided where it matters, documented either way — so you can see exactly how the thing works.

4 · Verify, then hand over. Nothing ships on an unchecked model output. You get the workflow, the evaluation and the governance layer — meaning you can run it without us.

What we won't do

We won't recommend a tool we haven't used. We won't put your client data or personal data into an AI system. We won't publish an AI-assisted claim we haven't verified. We won't quote a partner's pricing or terms secondhand. And we won't take an engagement where the honest answer is that you don't need us.

Start a conversation

Email us

General enquiries [email protected]

Partner enquiries → [email protected]

Know someone wrestling with this? Send them here — that's how most of our conversations start.

Nothing on this page is financial, legal or professional advice. Educational and informational content only.