How to build a cash flow playground with Claude (skill file inside)
Install in 5 minutes.
Hey Finance Engineers,
Confession: this issue was supposed to be simple.
The plan was “how to build your cash flow forecast with Claude.” One issue, one guide, done.
Then I started listing all the cash flow forecasts I’ve actually seen across businesses, and the list refused to end:
→ the CFO/CEO playground - the one you use for decisions and what-ifs (and even this one splits into several types, depending on how deep you want your assumptions to go)
→ the 13-week cash flow forecast
→ the monthly cash flow forecast
→ direct vs indirect
→ and every flavor of the above for SaaS, manufacturing, services - early stage, growth, enterprise…
One forecast, it turns out, is never one forecast.
So instead of one issue, you’re getting a series. Cash flow, type by type, each one as a working build.
Today we start with my favorite: the playground.
What a playground is, and why every CEO wants one
We built this one at Fuel for ourselves first. Then we noticed the pattern: every CEO and CFO eventually asks the same question, usually right before a board meeting.
“What happens to our cash if we...”
Cut this expense. Add two hires. Raise in July instead of September. And they don’t want a report next week - they want to move a number and watch the ending cash balance react. Right now, while the thought is still warm.
That’s a playground. Not the forecast of record, not the 13-week treasury view - the sandbox. Every future expense visible in detail, every line editable, and the monthly free cash flow, end-of-year forecast, and cash ending balance recalculating live as you play.
The problem: building one by hand in a spreadsheet takes days, and it breaks the first time someone drags a formula the wrong way. So we turned the whole thing into a Claude skill. You install it once, hand Claude your numbers, and it builds the playground for you - with the rules that keep it honest baked in.
How the CF playground skill works
Four stages, and the second one is the whole point:
1. Inputs: your historical P&L (monthly), payroll roster, software spend, revenue roster with each customer’s billing frequency and days-to-pay, and your forecast assumptions. Nothing invented: if something’s missing, it asks.
2. Data-quality gate: before building anything, it checks that the P&L ties out, MRR cascades balance, cash reconciles to $0, AR and deferred roll forward. Fails → ask, never fudge. (Regular readers will recognize the religion: the doer doesn’t get to fudge its own inputs.)
3. Build: one self-contained file, six linked tabs: P&L Playground, Software, Payroll, Active MRR, Cash Inflow, Cash Flow. Actuals locked, plan editable, formulas always computed.
4. Verify & ship: it re-runs the checks on the baked file before handing it over.
And the rules baked in are the ones that make finance models trustworthy instead of pretty: actuals are locked (your edits only touch plan months), expenses keep their sign (a typed minus won’t flip a cost into income), cash ≠ net income (AR and deferred revenue tracked properly, never a naive cumsum), and MRR cascades must balance. The stuff that silently breaks hand-built models, made structurally impossible.
The output is one offline HTML file. Send it to your CEO and they can play without touching your source data.
Set it up (5 minutes)
This is a Claude skill - install once, then just ask Claude to build or update a playground.
You’ll need: a Claude plan with Code execution and file creation turned on. Skills don’t run without it.
On Claude web / desktop / Cowork:
Turn on code execution: Settings → Capabilities → Code execution and file creation. (On Team/Enterprise, an owner controls this in Organization settings → Skills.)
Go to Customize → Skills.
Click ＋ → Create skill → Upload a skill (attach the skill file you can find below).
Select the zip.
Toggle it on.
Then start a chat and say: “Build me a cash flow playground from these numbers” - and attach your P&L, payroll, software spend, and customer roster (billing frequency + average days-to-pay per customer).
That’s it. First build takes a few minutes. After that, updating for a new month is one sentence: “here are June actuals, promote June to actual and extend the plan.”
My skill file (built and tested by ~20 finance engineers)
Here it is, the CF forecast playground skill. Install the skill, feed it your real numbers, and then do the thing your CEO would do: go to the plan months and cut your single biggest discretionary expense. Watch what happens to end-of-year cash.