RevTune
SolutionsFeaturesPricingAboutBlogContact
Sign InGet Started

End of dispatch · Ready to tune pricing?

Connect your billing stack and get answers in minutes.

Stripe, Paddle, LemonSqueezy, or Chargebee. Read-only access. No credit card to get started.

Start Free
RevTune

A Monday-morning readout from your billing data — labeled by what the evidence actually supports.

Built for SaaS founders
Questions? support@revtune.io

Product

  • Solutions
  • Features
  • Pricing
  • How it works
  • Dashboard

Resources

  • About
  • Contact
  • Blog
  • Security
  • Changelog

Follow

  • X / Twitter
  • GitHub
  • Instagram

Vol. 1 · No. 47 · The Pricing Desk · 2026

No placeholder docs, no dead ends — every link in this dispatch goes somewhere real.

© 2026 RevTune. All rights reserved.

PrivacyTerms·Set in Fraunces & JetBrains Mono.
All issues
Nº 05·AI & Data

How we use Claude to model price elasticity

Price elasticity used to require an econometrics PhD and a dataset the size of a country. Frontier LLMs changed the math. Here's the prompt structure we actually ship.

RT

RevTune Team

Pricing intelligence

March 15, 2026·8 min read

Price elasticity is one of those concepts that sounds intimidating but is conceptually trivial: if you raise the price 10%, do you lose more or less than 10% of your customers?

If you lose less, raise the price. If you lose more, don't.

The hard part has always been measuring it. You need cohorts that saw different prices under similar conditions, controls for confounders, and enough sample size to trust the result. For most SaaS companies, that data either doesn't exist or is too messy to use.

What changed

Frontier LLMs are surprisingly good at reasoning about messy, partial data. We feed Claude:

  • Cohort retention curves grouped by sign-up price
  • Plan migration history (upgrades, downgrades, churns)
  • Cancellation reason tags from the billing platform
  • Timing information for any prior price changes

…and ask it to estimate how much demand would change at three candidate price points.

The prompt structure

The structure that works well for us has four parts:

  1. Definitional grounding — explain what elasticity means in plain terms, with the formula
  2. The data dump — structured JSON, not prose. Cohort by cohort. Let the model do the aggregation.
  3. Constraints — "your output must include a confidence score and the three data points that most influenced your estimate"
  4. The ask — "given these candidate prices, rank them by expected revenue change"

That last constraint is what separates useful output from horoscope output. Without it, the model will hedge everything and give you a 5-paragraph essay. With it, you get a number, a confidence, and a citation. Three things you can actually act on.

What Claude is bad at

It's bad at extrapolating beyond the range of your data. If your highest cohort paid $99, asking it about $499 is asking for a hallucination. We constrain candidate prices to within ~50% of observed values, and the answers stay grounded.

The Journal

New issues, in your inbox.

One pricing essay every other week. No filler, no spam, unsubscribe anytime.

Requests go to a real inbox until self-serve subscriptions ship.

All issues