Three Claude Skills that run a monthly close: my workshop toolkit

In April I ran a workshop with Packt on Claude Skills for FP&A. Instead of slides about what AI could theoretically do, we spent the time actually running a monthly close with three pre-built skills and a realistic case dataset. The toolkit we used is the download at the bottom of this page, exactly as the workshop participants got it.

Let me tell you what’s in it, and how the pieces fit together.

Three skills, one close

variance-analysis. The workhorse. It runs the PVM decomposition (volume, price, mix), flags the lines that matter based on materiality, and writes a one-sentence commentary per flagged line. Not an essay. One sentence, the way a reviewer actually wants it.

simple-forecast. A three-month forward view with adjustable growth rates. Deliberately simple, because the workshop point wasn’t sophistication. It was showing how a forecast skill picks up where the variance skill leaves off.

reporting-standards. The quiet one that makes the other two look good. P&L formatting rules, sign conventions, number formats. Both other skills reference it, so every output lands in the same house style without anyone asking.

That last part is the design lesson I most wanted the workshop to teach: skills can build on each other. You don’t write formatting instructions into every skill. You write them once, in one skill, and the others point to it. That’s how a pile of prompts becomes a system.

The practice case: Precision Manufacturing

The toolkit ships with the case dataset we used live: Precision Manufacturing’s March actuals and budget as clean CSVs. Fictional company, realistic numbers, which means you can practice the full close end-to-end without touching your company’s data or asking anyone’s permission. Load the CSVs, run the variance skill, run the forecast, and watch the reporting standards keep everything consistent.

Practicing on a case with a known shape matters more than people think. You learn the workflow’s behavior before you trust it with real numbers, and you have something safe to demo when your boss asks what this AI thing actually does.

From three skills to a full workflow

The natural question after the workshop was: fine, but how does this scale beyond the close? This video is my answer. I show how I design workflows now, drawing them on a canvas first, chaining skills from the general ledger all the way to the board presentation, with the human checkpoints planned in from the start.

Notice the manual step sitting in the middle of that workflow. The meeting with the team, the human judgment, deliberately placed. Don’t leave that out of your designs. The goal was never a close that runs without you. It’s a close where your hours go into judgment instead of formatting.

How to use the toolkit

Start small and in order. Install the three skills, load the case data, and run the variance analysis first. Read what it flags and check it against the CSVs, because trusting a workflow you haven’t verified is how AI adoption dies in finance teams. Then run the forecast. Then look at how the reporting standards shaped both outputs.

After that, the real work begins: swap Precision Manufacturing for your own numbers, replace my materiality thresholds with yours, and adjust the commentary style to how your CFO likes it. The skills are a starting point built to be edited. That’s the whole idea.

Get the workshop toolkit

The download is the complete Packt workshop toolkit: the three skills (variance-analysis, simple-forecast, reporting-standards) and the Precision Manufacturing case dataset. Everything you need to run a monthly close end-to-end, on safe data, this week.