Twice a Year, Software Companies Admit They Got the Price Wrong

For most of the SaaS era, the pricing page was the most stable page a company owned. Three tiers, an aspirational name on the middle one, a number next to a seat. You shipped it and left it alone for a year and a half.

That page now goes stale in six months or less. The cause isn't inflation or a competitor undercutting you. What changed is the thing being sold. Vendors are moving off charging for access to software and onto charging for work that software and AI agents finish together.

Kyle Poyar put the scale of it on one page in a chart posted to LinkedIn in July, tallying eighteen separate pricing changes in 2026 alone across Salesforce, HubSpot, Google, Adobe, Figma, Canva, Notion, Clay, ServiceNow, SAP, OpenAI, Lovable and Anthropic. No two vendors landed in the same place. Poyar, who writes the Growth Unhinged newsletter, has advised more than a hundred SaaS and AI founders on monetisation, over a dozen of them this year, and reports pricing anxiety running higher than he has ever seen it, with almost nobody satisfied with the model they are currently on.

Four companies, four incompatible answers

Salesforce is on its fifth AI pricing model in under two years. The current one bills a flat $2 when the Agentforce Help Agent resolves a customer issue without human intervention, and nothing at all if the customer escalates to a human or leaves unhappy. It replaced a $2-per-conversation model that drew enough criticism to prompt the launch of Flex Credits, a usage currency where one credit costs half a cent and a single agent action burns twenty of them. Three different mental models for one product line inside twenty-four months.

HubSpot went further toward outcomes and undercut the field doing it. Since April 14, Breeze Customer Agent bills $0.50 per resolved conversation and Breeze Prospecting Agent bills $1 per qualified lead. That resolution rate sits at roughly half of Intercom's Fin, which starts at $0.99. HubSpot made the ideology explicit in the announcement, arguing that AI should be priced on the value it delivers rather than the compute it consumes.

SAP was the least likely company on the list to touch per-user licensing, and it is now unwinding it. CEO Christian Klein told Bloomberg in March that SAP would start charging on AI consumption, a change he framed as the biggest to the business model since the move to cloud subscriptions. The mechanism is a prepaid credit called an AI Unit, and the fine print is where the brand exposure sits. Those units expire at the end of the contract year, with no standard notification beforehand, no credit for what goes unused and no obligation to roll anything over.

Anthropic unbundled in the opposite direction. Token allowances came out of the Enterprise seat entirely, the headline price fell toward a flat $20, and the 10 to 15 percent API discounts that used to ride along with $40 to $200 seats disappeared with them. The seat is a base fee now, and usage bills at standard API rates with no bundled allotment and no cap. Lower sticker, uncapped meter.

Consumption. Credits. Resolutions. Qualified leads. Four of the most sophisticated pricing teams in enterprise software looked at one problem and produced four answers that don't reconcile.

Marketing owns this now, whether it wants to or not

A pricing model is a claim about what you are worth and what you will be held responsible for. It is the most legible promise a company makes, and buyers read it more carefully than anything on the homepage. Poyar reaches the same conclusion from the operator side, arguing that "pricing is ultimately about positioning" and that the job is to tell a story about who the product serves and why it beats the alternative.

When Salesforce says it doesn't get paid unless the agent finishes the job, that is a statement about confidence in the product. When SAP moves to credits that evaporate at year end, that is also a statement, and what it says is that forecasting risk belongs to the buyer. One analyst summarised the market mood after the Agentforce launch by noting that customers want to pay for work completed rather than software accessed.

So every repricing is a repositioning, whether or not anyone from brand was in the room. And the industry is repositioning itself about twice a year.

Three costs accumulate while that happens.

The first is legibility. Outcome pricing sounds simple and is anything but. What counts as resolved? Who decides? What happens when a ticket reopens a week later? A price the buyer cannot verify independently isn't a simple price, however clean the headline number looks.

The second is forecast anxiety. Bloomberg reported SAP customers already struggling to predict what they would spend under consumption pricing. Budget unpredictability does not stay filed under finance in a buyer's head. It becomes a feeling about your company.

The third is the tax framing. When AI arrives as an uplift rather than a choice, as it did when Microsoft raised M365 commercial plans between 9 and 33 percent from July 2026, buyers experience it as a levy on something they already bought. Once that frame sets, it is very hard to shift.

The reversals are the real story

The pattern nobody talks about is the snap-back. A company ships a new model, meets the customer reaction, and quietly restores the old one a few weeks later. Poyar's chart catches one in progress, with Anthropic testing Claude Code as a Max-plan exclusive and running into immediate backlash.

Reverting is often the correct commercial call. It costs far more in brand terms than a straightforward increase does, though, because of what each move communicates. A price rise tells customers you cost more now. A reversal tells them you were guessing. The second version lodges deeper, and it hardens into a belief that your pricing is provisional, which turns every renewal conversation into a negotiation.

Worth holding the opposite thought at the same time. Poyar's view, set out in the newsletter edition behind that chart, is that a pricing change nobody pushes back on probably didn't go far enough, and that the useful response to complaints is to identify the specific problem and fix it rather than retreat wholesale. Silence is not success. The distinction that matters is between a company absorbing feedback and adjusting, and a company that abandons a model before it has had time to work.

The market has already started pulling back toward the middle. Analysts describe 2026 as a year of hybridisation and partial reversion, with some vendors keeping per-seat pricing and simply redefining what a seat means.

The survey data captures the confusion well. A Cruxy poll of 300 SaaS CEOs in April found 97 percent planning to retire seat-based pricing within two years, while 94 percent said seat-based pricing currently matches the value their product delivers. Separately, ICONIQ found 37 percent of companies intending to change their AI pricing model within twelve months.

Almost everyone thinks the model is finished. Almost everyone thinks it currently works. That is a stampede rather than a strategy.

Four things worth getting right

Price the outcome you can evidence, not the one that demos well. If you cannot show a buyer the audit trail behind a resolution, you have created a dispute rather than a differentiator.

Treat the pricing page as owned media. Serious buyers give it more scrutiny than any campaign you will run this year, so write it like the positioning document it has become.

Disclose the awkward parts yourself. Expiring credits, uncapped meters and removed discounts all surface eventually. Volunteered, they cost you a paragraph. Discovered, they cost you the renewal.

Build for change without broadcasting instability. Nobody has identified the correct model yet, so the advantage goes to whoever designs a pricing architecture that can evolve without every revision landing as an admission of error.

Pricing used to be the last thing a brand team thought about. In a market that reprices itself twice a year, it has become one of the first.