Services · AI for Finance

Practical AI for the finance function

Hands-on training and guarded pilots with Claude and Copilot. I help life sciences finance teams adopt AI usefully — zero to one, measured, and auditor-aware.

Get in touch

Where this helps

  • Engineering edge

    I build closed-loop Finance systems that expand business intelligence and get smarter each quarter.

  • Workflow discovery

    Find the 3–5 finance tasks where AI saves real hours without compromising control.

  • Team training

    Zero-to-one coaching so analysts and controllers can use Copilot and Claude confidently.

  • Guarded pilots

    Scoped rollouts with data rules, human review, and success metrics.

  • Control design

    Patterns auditors can understand — because finance AI without guardrails is a non-starter.

  • No hype

    If a use case isn't useful, we kill it. Tools serve the work, not the other way around.

How we work

Hands-on, not board decks

  1. 1

    Inventory finance workflows and sensitivity — what can be assisted vs. what stays human-only.

  2. 2

    Stand up a small pilot (often 3–5 users) on Copilot and/or Claude with clear guardrails.

  3. 3

    Train hands-on, measure time saved and error risk, iterate prompts and review steps.

  4. 4

    Document controls and expand only what proved useful.

What you walk away with

  • Trained finance users with repeatable prompts
  • Documented AI use policy and review steps
  • Pilot results you can show leadership or auditors
  • A shortlist of scaled use cases — or a clear 'not yet'

Tools we train on

Claude + Copilot — practical, not hype

I apply AI to finance work the way an engineer would: start small, measure usefulness, add guardrails, and only then expand. No “AI-first finance” theater.

Claude

Claude

Analysis, drafting, synthesis

Microsoft Copilot

Microsoft Copilot

Excel, Outlook, Teams workflows

Why now — and not a year ago

What changed is not the model

Two years working with these tools. What changed is that the layer a company actually needs finally became practical — procedure, guardrails, and evidence you can put in a close.

Each layer contains the one below it

Only the outer layer makes it a process you can put in a close.

Evidence & Control

Can we prove it to an auditor?

Procedure & Guardrails

Does it do it right every time?

Context

Does it see our data?

Prompting

How do I phrase the question?

My own path through it

  1. 2024 · Prompt

    Chat and drafting. Useful for research — nothing I would have put in a close.

  2. 2025 · Context

    Our own data in context. Better answers, but still one-off. Not repeatable, not evidenced.

  3. Early 2026 · Harness

    Written procedures with guardrails and forced tie-outs. The first output I was willing to put in a workpaper.

  4. Today · Loop

    Used in a live close, with the evidence package built alongside it — not bolted on afterwards.

Zero-to-one hands-on training

I sit with your finance team and build real workflows together — variance narratives, contract summaries, board draft sections, reconciliation assists. You leave able to run them without me.

  • ✓ Live working sessions, not webinars
  • ✓ Prompt patterns tailored to your charts and policies
  • ✓ Clear do / don't rules for sensitive data

Masked pilot

4 finance users · Copilot + Claude · Guardrails

Ran a scoped pilot with four finance users combining Microsoft Copilot and Claude. We defined data boundaries, review steps, and escalation paths up front. External auditors later commended the control design — practical AI that didn't scare compliance.

“I never thought my engineering background would come in handy — now it's the distinctive advantage.”