Why does AI break hourly and per-token pricing?
Because AI shrinks the meter. A return that took a day now takes an afternoon, so a firm billing by the hour earns less for the same work, and clients push for discounts. The Thomson Reuters Institute argues that firms should avoid billing less for work done with AI’s help and should instead price on the value and risk of the deliverable, noting that KPMG won a 14 percent fee reduction from its auditor tied to AI savings (Thomson Reuters Institute, 27 October 2026, read 2 November 2026). The profession was never selling hours, it was selling answers. The same pressure hits software: per-seat and per-token prices track inputs, while the buyer cares about what gets done.
Cost-based vs value-based vs outcome-based pricing: what is the difference?
| Cost-based | Value-based | Outcome-based | |
|---|---|---|---|
| Price set by | Seller’s cost plus a markup | Worth of the deliverable to the buyer | A measured result, paid per result |
| AI example | Hours billed, or tokens resold at a margin | Fixed fee for a filing package or a processed invoice batch | A fee per resolved support ticket or per qualified lead |
| Who carries delivery risk | Buyer | Shared | Seller |
| What AI efficiency does | Cuts the seller’s revenue | Raises the seller’s margin | Raises margin if the outcome rate holds |
Outcome-based pricing is the strictest form of value pricing, and it only works when both sides can count the outcome the same way.
What is a value-based pricing example in AI?
An accounts payable agent sold per invoice processed, not per user seat: the buyer compares that price with what a processed invoice costs today, and the vendor keeps the gain if its automation improves. A law firm quoting a fixed fee for a contract review package instead of hours is the services version. The price anchors to the buyer’s baseline, so you need that baseline first, the discipline AI ROI measurement describes.
How should a buyer judge a value-based AI price?
Convert it into your own cost per task at expected volume and set it against today’s loaded cost. Define the unit precisely: what counts as resolved, processed or qualified, and who audits the count. A value price without an agreed definition of the outcome is just a bigger invoice. Ask for caps at high volume, and put the terms inside the same AI vendor assessment and AI business case as any other purchase. Pay for outcomes you can verify, not outcomes the vendor reports.