Skip to content
CodelessOps

CodelessOps · Services

AI process implementation.

Are you getting your money's worth from the AI you already pay for?

Most firms aren't yet. In PwC's 2026 Global CEO Survey, 56% of CEOs said AI had brought neither lower costs nor higher revenue. Often it gets used to tidy emails and spreadsheets while the month-end still runs by hand.

I configure Claude, ChatGPT or Microsoft Copilot to work the way your finance team does, with no new tool to adopt. I connect it to the systems you already use and build reliable automations it runs in the background, under your own accounts. Month-end reporting, the finance inbox and your FP&A processes then run without anyone babysitting them.

Keep using the AI you have. I make it capable of running your real processes.

Master Anaplanner · CIMA-qualified · 15 years in finance operations

01 · Use cases

It starts with the problems finance teams already have.

In each of these, the AI gets used for bits of the work today. The process itself is still manual.

The close, the bank rec, the variance report and the pack run today in my public close kit, on a fictional company you can inspect. Yours isn't here? If it comes up again and again and someone already reviews the result, send it to me.

02 · What gets built

What AI process implementation does.

AI process implementation is configuring the AI you already have to run specific finance processes, reliably, under your own accounts. For each one I do the same six things. They're what you'd give a new starter:

You keep the platform. I implement the processes that matter.

03 · Proof

Work I've done, with the numbers.

The Exscientia close used no AI: it was scheduled data loads and a planning model. It's here because the job was the same. Find where the days go, then rebuild the process.

04 · Running and controls

It runs without babysitting, and your controls stay in place.

  • It reports where it is: running, waiting for sign-off, done or failed.
  • If a run fails, a named person is told and nothing downstream runs until it's fixed.
  • People keep the access they have today. If two people run the bank rec now, they're the only two who can run it with the AI.
  • A past month with known answers is rerun on a schedule and after every change. The figures have to match to the penny.
  • Changes to a procedure are reviewed and versioned before the team gets them.

The technical side is mine to deal with: the connections, the logins, the monitoring. You also get a one-page policy your CFO signs: what the AI may touch and what it never does.

05 · Why this way

Don't buy another AI product. Use the one you have.

Products built around AI keep getting overtaken by the AI itself. Each release does more out of the box. I've built that kind of product myself, and more and more of what it did now comes with the AI you already pay for. So the better investment is the one you've already made.

You could set it up yourselves. Most teams would rather hand it to someone who has done it before, the same way you'd get a decorator in to paint a room you could paint yourself.

What the AI doesn't come with is your process. I've spent fifteen years in finance operations, so I know what a close has to respect. These are the kinds of rules I build in:

  • Last month's reconciling items are brought forward, so nothing drops off between periods.
  • A stale item goes to a person as a question. The AI never writes it off.
  • Commentary is written only for the lines over your threshold, and only with a cause that can be sourced.
  • The person who prepares a step can't be the one who approves it.

None of that comes with the licence. I tested Anthropic's free finance skills to see where they stop: what Anthropic's finance plugin leaves to you.

06 · How it works

One process first.

  1. Step 1

    Scoping sprint

    You pick a process that comes round every month and takes too long. I sit through it with the person who does it, check what your AI tool, your plan and your IT team allow, and scope the build, or give you a written reason not to build.

  2. Step 2

    Build

    I separate what code has to calculate from what the AI can interpret, connect the systems, write the procedure in and test it on a past month until the figures match. Each phase has a fixed price.

  3. Step 3

    Retainer

    Your team runs it and signs it off, with a runbook that makes the process theirs to change. I keep the checks running, handle changes and add the next process when you want it.

See the steps and prices

07 · The honest boundary

Where this stops.

Your data goes where it goes today, to the AI provider you already have an agreement with. If it has to stay inside your own environment and your current tool can't do that, this is the wrong route and I'll say so.

No new tool for your team still means some code. The automations I add are small scripts that run in your own environment, and your IT team has to agree where. What the AI can connect to and enforce differs by tool and by plan, and the kit you can inspect today runs on Claude. I check all of that before anything is built.

Some processes aren't ready. If nobody can describe the steps, or they change every month, I'll tell you before you pay for a build. And if your tool already does the job out of the box, I'll tell you that too.

08 · Common questions

Questions about the AI you already have.

What is AI process implementation?

Configuring the AI you already have to run specific finance processes. I connect it to your systems, write in how your team does the work, build the automations that run it in the background, and add the controls and the record of each run. You keep the platform.

Do we need to buy anything new?

No new tool for your team. You keep the AI licences and systems you have. What I add are the automations behind the AI, and they run in your own environment.

Where do the automations run?

The AI runs where it runs today, in your provider's cloud under your accounts. Anything I add runs in your own environment. Nothing runs on my machines, and nothing I add holds your data anywhere else.

Which tool do we need?

Whichever you already have: Claude, ChatGPT or Microsoft Copilot. What differs is what the tool can connect to and how much it can enforce by itself. Where it can't, your existing folder and system permissions do that job. I check this in the scoping sprint, before you spend anything.

What will our auditors see?

A record per run: what went in, which version of the script ran, what came out, who approved it and when. They can rerun it and get the same number.

Does our data train the model?

That's set by your agreement with your AI provider, and it differs by provider and plan. I'll point you to the current terms and won't paraphrase them here.

More questions, on numbers, access and handover

09 · Get in touch

Send me one situation. I'll reply within two working days.

A few lines are enough: the workflow or the estate, the systems involved, how often it runs and who owns it. You'll hear what I'd do first, what it would take and what I'd leave alone, whether or not we work together.