AI Guides › Money & Business

Using AI To Model A Pricing Decision

By Nigel Guy · 2 min read

Pricing decisions usually get made one of two ways: a gut-feel number that "feels about right" compared with competitors, or an elaborate spreadsheet nobody trusts enough to actually act on because it's full of assumptions dressed up as inputs. The evidence ledger used for testing a business idea — sorting what you know into fact, assumption, unknown, and constraint — solves the same problem here. A pricing decision is really a bet, and the ledger is how you find out how much of that bet you're actually informed about before you make it.

The mechanism

1. List what's driving the pricing decision — your costs, what competitors charge, what you believe customers will tolerate, what margin you need.
2. Label each one: a fact (your actual cost per unit, verified), an assumption (customers will accept a 10% increase), an unknown (how many will actually leave if you raise it), a constraint (a contract that fixes price for existing customers until renewal).
3. Have the assistant build the model around the unknowns, not around a single guess. Ask it to show you the P&L at your assumption's low, middle, and high case for churn, side by side — not one tidy number that hides how much the outcome depends on a figure you haven't verified.
4. Set your threshold before you look at the range — what churn rate would make the price rise not worth it. Decide that while you're still neutral, exactly as you would before running any test.
5. Test the biggest unknown at the smallest scale you can — a price change on new customers only, or in one segment, before it goes across the whole book.

What to skip

Skip asking the assistant for "the right price" as a single output — that question invites it to sound confident about exactly the number that's actually your biggest unknown. And skip building the model around industry averages you found in a general search as if they were your facts; a competitor's public price is a fact about them, not a fact about what your customers will pay you.

Guardrails

• A model is only as good as its labelled inputs — treating an assumption as a fact inside the spreadsheet defeats the exercise before the numbers even run.
• The assistant can build the arithmetic fast and cleanly; it cannot tell you your actual churn risk, your market position, or your customers' patience — those come from your own knowledge of the business.
• Run the low-case scenario past your actual cash position, not just your margin — a price change that pencils out on paper can still be dangerous if a bad quarter of the low case would break your cash flow before the model plays out.
• Revisit the model once real data comes in from the small test; the whole point of testing at small scale first is to convert an assumption into a fact before you commit at full scale.

All 751 AI guides · JulieMango plans from £17/mo