Strategy & Planning

How to run a scenario analysis for a strategic decision with AI

Strategy & Planning3 AI tools7 steps6 friction points

Want an agent to handle this on your live data? See the Starch version →

Scenario analysis is the practice of modeling multiple versions of the future — what happens to your business if revenue grows 15% slower than plan, if you add three engineers this quarter, or if you delay a fundraise by six months. Every operator faces moments where a decision has real financial consequences, and the only honest way to evaluate it is to run the numbers across a few credible futures before committing. Most teams do this in a spreadsheet, which works until the spreadsheet breaks or the assumptions get stale.

AI feels like a natural fit here because the mechanical parts — building formulas, structuring assumption tables, writing out scenario narratives — are exactly the kind of structured reasoning that large language models handle well. You can describe your business in a prompt, specify the levers you want to test, and get a working model structure back in minutes instead of hours. That's genuinely useful, especially if you don't have a finance hire who lives in Excel.

ChatGPT, Claude, and Gemini can all contribute meaningfully to scenario analysis. They'll generate scenario frameworks, suggest relevant assumption variables for your business model, write out the logic for a three-scenario model, and produce structured tables you can paste into a spreadsheet. Claude tends to be particularly strong at multi-step financial reasoning. What they can't do is connect to your actual data — the numbers you hand them are whatever you copy in at that moment.

Skip the manual work → See the Starch version

Strategy & Planning3 AI tools7 steps6 friction points

AI walkthrough

How to do it with AI today

A practical walkthrough using ChatGPT, Claude, and other off-the-shelf LLMs — what they're good at, what you'll have to do by hand.

Tools that work for this

ChatGPTClaudeGemini

Step-by-step

1 Export your current financials from wherever they live — Stripe for revenue, Plaid or your bank for burn, QuickBooks or a spreadsheet for headcount costs. You'll be pasting these in manually, so pull the most recent 3-6 months at minimum.

2 Open Claude or ChatGPT and paste in your baseline numbers with a prompt that describes your business model, current burn, and the decision you're evaluating. Give it context: stage, revenue type (recurring vs. project), and the specific variable you're stress-testing.

3 Ask the model to define three to five discrete scenarios — for example, base case, upside, downside, and a specific strategic variant like 'delay hiring by one quarter.' Have it output the assumption set for each scenario as a structured table you can copy.

4 Prompt the model to extend each scenario into a 12 or 18-month cash projection, given your starting cash balance. Ask it to show month-by-month burn, cumulative cash, and break-even date for each scenario in a format you can paste into Google Sheets.

5 Copy the output into a spreadsheet and manually update the cell formulas — LLM output often uses hardcoded numbers rather than linked cells, so you'll need to wire the assumptions to the outputs yourself before the model is actually interactive.

6 Return to the LLM to generate the narrative layer: a plain-English summary of what each scenario means, the key risks, and a recommendation on which decision the numbers favor. Ask Claude to write this as a one-page memo you can share with co-founders or board members.

7 When assumptions change — a new month closes, a deal slips, burn increases — repeat the process from step one. There's no persistent state; the model doesn't remember last month's version.

Prompts you can copy

We're a SaaS company with $180K MRR growing 8% monthly, $95K monthly burn, and $1.4M cash. Model three scenarios: base case (current trajectory), a hiring freeze that cuts burn by 20%, and a price increase that lifts MRR by 12% immediately. Show 18-month cash and break-even for each.

I'm deciding whether to raise a bridge round now or in 6 months. Current burn is $110K/month, cash is $900K, revenue is flat. Model the scenario where I raise $500K today versus wait 6 months assuming burn stays flat. What does each path look like at month 18?

List the 8 most important assumption variables for a B2B SaaS scenario model at the seed stage. For each, suggest a base case value, an upside case, and a downside case, assuming $150K MRR and 15% monthly churn.

Here are my last 6 months of P&L data: [paste data]. Build a scenario where I double the engineering team in month 3 versus keep headcount flat. Show the monthly cash impact through month 12 in a table.

Write a one-page scenario analysis memo for my co-founders comparing two hiring plans: adding 2 salespeople now at $80K each versus waiting until MRR hits $250K. Include assumptions, cash impact, and a clear recommendation.

Reality check

Where this gets hard

The walkthrough above works — until your numbers change, the LLM hallucinates, or you have to re-paste everything next month.

No live connection to your Stripe or Plaid data — every session starts with a manual export and paste, so the numbers are already a few days old before you've asked a single question.

Nothing persists between sessions. The model you carefully prompted this month doesn't exist next month; you reconstruct the full context from scratch every time an assumption changes.

LLM output is rarely formula-linked. You get hardcoded tables, not a working model — so after the first assumption shift, you're back to the LLM or manually editing cells yourself.

Large datasets hit context limits. If your transaction history runs long or you're modeling at a per-SKU or per-department level, the model truncates or loses track of earlier rows mid-analysis.

Output structure drifts across runs. The table format, column labels, and scenario names you got last week don't reliably match what you get this week, which makes version comparison tedious.

The model can't tell you when the underlying numbers change. If your burn spikes in week three of the month, you find out when you next think to run the prompt — not proactively.

Tired of the friction?

Starch runs the whole workflow on live data — no copy-paste, no hallucinated numbers, no re-prompting next month.

See the Starch version →

Starch alternative

The same workflow on Starch

Starch is an agentic operating system. For scenario analysis, that means an agent builds a persistent financial model connected to your live Stripe and Plaid data — one that updates when your numbers change, not when you remember to re-run a prompt.

Start with the Scenario Analysis app from Starch's App Store — it connects Stripe and Plaid on a scheduled sync so the baseline reflects your actual revenue and burn, not a copy-paste from last Tuesday.

Adjust assumptions in plain English and see the impact immediately. Tell Starch 'show me what happens to runway if we add three engineers in month two and revenue growth slows to 5% monthly' — the model recalculates across all scenarios without you rebuilding anything.

Each scenario shows runway, burn rate, and break-even under its specific conditions, all in one view — so you're comparing futures side by side instead of toggling between spreadsheet tabs.

Pair it with the Runway Analysis app to see your baseline cash position updated daily from Plaid and Stripe — the scenario model always starts from a number that's actually current.

Describe a custom variant in natural language and Starch builds it into your app. Tell it 'add a scenario where we delay the fundraise by four months and cut burn by 18%' and the agent adds it without you touching a formula.

The app persists and runs continuously. When next month closes and your Stripe and Plaid data refresh, your scenario model reflects the new actuals automatically — no re-prompting, no manual export.

Get closed-beta access →

Toolkit