# Customers Push Back on Usage Based Pricing as Seats Prove Sticky - Is SaaS Broken? - Week of July 3, 2026

> A Help Scout founder spent a year trying to kill per-seat pricing and customers refused, even when usage was cheaper, while the token bill kept biting and OpenAI kept its inference-cost savings. Our synthesis of the SaaS podcast tape for the week of July 3, 2026.

## Is SaaS Broken?

### Week of July 3, 2026: Customers Push Back on Usage Based Pricing as Seats Prove Sticky

---

Last week the bear case had swagger: consumption pricing is a "one-way door," seats are dead, adapt or die. This week a founder who tried it said the door swings both ways, Help Scout spent a year moving off per-seat pricing, and customers refused to walk through. Meanwhile the token bill kept biting, and OpenAI halved its inference cost and kept the savings.

## TL;DR

- **Buyers are resisting usage pricing.** Help Scout ran per-seat vs. per-contact for 12 months; customers rejected the usage model even when it was cheaper. Seats are stickier than the bears think, a real counter to last week's "one-way door."
- **The token bill keeps biting.** Coinbase cut AI spend ~50%, Ford rehired ~300 inspectors, Priceline meters employees to the CFO. Model prices fell ~50% in a year while token *costs* rose 4.5x (Bain). Usage-priced revenue faces a demand cap.
- **OpenAI halved inference cost, and is keeping it.** SaaS AI-feature margins won't auto-improve as compute cheapens; the labs pocket it.

## What's new

**1. A founder tried killing per-seat pricing. Customers said no.** On [Startups For the Rest of Us, Ep 839, 6/30](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOiCLSRSHqRZFCPJ7-2BpLaBCH2tY2c6BubIWNlyX2L2eh3L7zvprF-2FuEwFGbql9-2Bb93T57F4QxU39fBT9ACHictl06RgV3iv-2F2guSF2y1XOXnAg-3D-3DXlgt_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbWHk4RGVVGQEpZR890xCB2Oy5ywNr8pe2wbRiZVmG6cnSiRVvXHq-2F5Z6zs6TLU8Syfi7H-2BlyWj-2Fw9cThNH67Yi-2BsJxqZHOiAAlOyL9QHnIa3I6XCUmQMg3JeSp-2F-2BLXfjQf-2F7jqncgCOBWK61e0lgILy0-2B-2FFMyslv-2FLN9A0TdQtZuw-3D-3D), Help Scout founder **Nick Francis [Operator]** described moving from per-seat to per-contact in Nov 2024: "we tested... three different variations over 12 months. Long story short, it wasn't there. Even when they would pay less, they just didn't want to do it." Per-contact was "30% less variable than per seat," but buyers saw seats as more controllable, so Help Scout landed on a hybrid, "seats plus you pay for AI resolutions." The seat base is stickier than the "seats are dead" thesis assumes.

**2. The token bill came due again, with a reversal.** On [Future Ready Leadership, 7/1](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOi42qMy3J57XjbMdL55xoYQR5UAFuLmxmwycply72kRUp3Zi3k0v91Pbt03bF69ZU6NlHm-2Be4uCr9iGWL9fZqR-2Bm6XdP2JdJCcSJFU2fE5qEQ-3D-3DbOZq_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbWHk4RGVVGQEpZR890xCB2Oy5ywNr8pe2wbRiZVmG6cna00e2NFcPfR7xlDh3YAcnTtKpjMw9XLr8FSUzko73PJ3933z7UHkFy6UDa8tFL1HDm9kOnlyBTr2uH-2F-2B3O5Z2cYgKSOLZLZ3mNruVoRnWOnqraCTKwdEDB4s9xsWXWE1g-3D-3D), **Jacob Morgan [Analyst]** (citing WSJ/Bain/Goldman) reported Ford "brought humans back" after AI quality checks failed, and per Bain "model prices fell roughly 50%... but token costs grew four and a half times over that same window." On [20VC, 7/2](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOh0KKvRFJkW1nJURZBwq7x6lbAbnrkOy40CdcPabYGkEux0CNYZFwAlW2liU8ulEqKAqCguVcEzLzl4SJLXvh4q1v8-2Fw-2FLOHOfYyBcC-2FKz-2BSw-3D-3D-rtd_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbWHk4RGVVGQEpZR890xCB2Oy5ywNr8pe2wbRiZVmG6cnbgJhyS7NQJq-2BkShOCXhkCCM19EIcyQjSgk-2F-2FH7U3yq-2FpIJNpTt6AOr2HzWUQIm0n8mdCi8ex8jpVgHYJubH-2FJbe-2BY-2BLaktqa1v6oIdkei18w4D-2BoBu5BvBeXh-2FgELBHqw-3D-3D), **Jason Calacanis [Analyst]** said Coinbase cut AI spend ~50% and framed the "where's the lift?" problem: a top-decile portfolio company wanted to double token spend, but "tie it to ROI, and this amazing team couldn't."

**3. Amazon got repriced, hourly to token-based.** On [The AI Daily Brief, 6/30](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOivVD1HY35yQ8-2BlMq5fRQMFvV10Cwzx1yacau3Pb7hnQhfgQ9lidLiLRRyo1bBs4u9htIUj34fmcuUaMDAqR1V6uoJDd-2FFHjKwto4YFMMOzVw-3D-3DrBJ8_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbWHk4RGVVGQEpZR890xCB2Oy5ywNr8pe2wbRiZVmG6cnRDRT8NvqRMjL-2FM7uiAemdbpYjv-2Fe70dUz24IecJUUwPH3DR96YRuIStim9vntXu00fBFW4FXxoRGEclZwa6xtaGYwZKZfNE12c-2FoNDBuqtKxhGovi1nErDC-2FtznLhwb1g-3D-3D), **NLW [Analyst]** (citing The Information) said Anthropic renegotiated the "sweetheart deal" from Amazon's $13B investment: Amazon's Claude bill went from "raw computing hours" to "token-based, similar to every other large Anthropic customer" next year, and Amazon is eyeing OpenAI or its own Nova models to save. "The end of the AI subsidy era is dramatically changing the economics for AI services."

**4. OpenAI halved inference cost, and will keep the margin.** On [The Information's TITV, 6/30](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOi9ZpC-2F8LKHKuKrczX-2FFysdZMlfKikTYu9fPQ68NH0vkqY-2Bztdg6WU2RZy6pxF2esyNtn3lnCeFZAIswUH7xK7Nux0N19r2jtACzE9lB5Jf8Q-3D-3DgPRU_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbWHk4RGVVGQEpZR890xCB2Oy5ywNr8pe2wbRiZVmG6cnTZmYk1twx-2BUHJnDZDy-2BOSGP8OwwPQuonbtCU3ByPC6Ych7QnedJ7M9a196viE-2Ftq8fJP2JOlext7CiImnkx8M8mVdsYAU82y6duNVRJgwLmTCffny3vsw5jgz0Dqf6GSQ-3D-3D), **Stephanie Palazzolo [Analyst]** said OpenAI engineers "cut down on cost by more than a half," then predicted: "they're going to expand their gross margin... I don't think the customers are going to see any of this." The ~52% AI-feature floor won't fall just because tokens get cheaper.

**5. Two silent names surfaced as operators.** On [The Product Podcast, E302, 7/1](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOi2w8X-2BIyi6cdq0tPa7365pB5Hi0ZhvWrWU4J7mhZlAZ7Vf7pngwJQpscIIjPzPgevK0PxhQjKQp-2BFVWcIh85vZy-2FmjP6kQLF6JBoCAFt8lqA-3D-3DyucZ_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbWHk4RGVVGQEpZR890xCB2Oy5ywNr8pe2wbRiZVmG6cnR9oiQ2ixvSf-2Fg3-2FeCpwGAI5QyECQ3hZOPRB7C1v7lJ0NHGJ8VYX5HU-2BNWwSY8Mom-2FE3HLHklqEWtkyHMvlBcMTRJJbFtPQrt0vJVunYzPKZlyuwMUvcyMy7KIJfIOb0Sw-3D-3D), Asana CPO **Arnab Bose [Operator]** admitted it is "difficult for us to provide value to extremely small businesses... 10 [people] or smaller, then you would be able to hack around... with these agentic tools." Seat erosion at the low end, in the CPO's words. On [Tech Talks Daily, 6/29](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOiMbGKb0aY9eK-2F9rFgxDtmcJkkN-2FyMxf-2BM93UOY7LZF6Q6H-2Fkz2q2mXdkyzYIRtMZ0mbHhp4T-2FdpSp8ldRLZOqT-2FNqNHgfs6eKFWkU0JJJ6Uw-3D-3Dmxr8_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbWHk4RGVVGQEpZR890xCB2Oy5ywNr8pe2wbRiZVmG6cnQEhcxNXDf3DOzzMI1LHIs7-2BPPjm0zKc-2Fao89dyspes4j7kSimQaVZPLTWHQC1ioSaa91yoSJ4aakkjFBG8KiTg8ZXaphkiM1DK8uS76rYjDsyqfzyoWcfyrjJB-2BV0OgNQ-3D-3D), Atlassian CTO **Taroon Mandhana [Operator]** pitched context as the moat, a teamwork graph of "more than 154 billion connections," but conceded he's opening it via MCP to "Cloud Cowork... ChatGPT," "not restricted to our own agent Rovo." Openness widens reach, but invites rival agents into the workflow Rovo hoped to own.

## The debate

**Bull, incumbents re-rate, seats prove sticky.** Offered *cheaper* usage pricing, Help Scout's customers chose seats. On [Monetary Matters, 7/2](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOij3rmz8ji-2F2PySniYJk5afeylKrBOWmHCIiik6jI2tJlMb41av43u03jlOqhep8NQ3gY5AD3TAGHhV-2FUTVMCJm9WeycDDmCeh6szdlequjHg-3D-3DKEfl_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbWHk4RGVVGQEpZR890xCB2Oy5ywNr8pe2wbRiZVmG6cnRu8L3-2B-2BgIQsm3c3BAdr0mZW5nE0wL2LQjxWifHFAGQtvvHSPwyf-2BEleJ0Iy2r-2FArXq0oyM9swfliTnfjwHaqMEAENn7GqdyFO4-2Fuc-2FopaW-2B6Up3Pf40fJlMmAFWD82LUw-3D-3D), investor **Erik "YWR" [Analyst]** conceded the tell: "the earnings estimates are going up on them... that kind of semi-refutes this software disruption... the derating is higher than the actual risk is warranted." Outcome pricing works where engineered for it, Intercom's **Brian Donohue [Operator]** ([The Way of Product, 7/2](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOj7a1Dbx-2BdcmHsrdNRYcSbnshmp3GdGfX61k0pDQE22I3B-2FaXgNxu1ROTAdIPA0Bf8U9AdGnvNRLVg-2BUcaJkIX4-2BoHW1gw-2FEKpjdmDRMfRmZw-3D-3DGj19_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbWHk4RGVVGQEpZR890xCB2Oy5ywNr8pe2wbRiZVmG6cnUDOtCcU9akvNFNN-2BvDdvCJt3fM6NwaVTUVr38gt-2BtfmoZhNV9i-2FN-2B8ZHBUdHumjiXgDclhMexmzufPj9r80KNewdQhrDwjfj8I6UJ7DXEcGug7tM6NgTwwqY-2FsdIh-2FrrA-3D-3D)): "if you get a resolution, it's 99 cents... no base cost." Owners of the data and workflow can graft consumption onto a sticky seat base.

**Bear, structural, and improving with time.** Erik gave the mirror: "all this seat-based pricing feels like that's going to have to change... these might be semi-dead money," and "the bear case might actually be improving with time." The channel agrees: on [Business of Tech, 7/2](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOjlEjJ5Y-2FRs-2FqpZcf5otj1bd-2F9eLJK3BnM6-2BRoiG0wtJKKM8YsCxpUpC2ya4GLT3cNnpg6jH6yULw3j2Hrvdug5HPBVEmYIloGLdO-2Bn35NVAg-3D-3DJKAG_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbWHk4RGVVGQEpZR890xCB2Oy5ywNr8pe2wbRiZVmG6cnQvYzWpwLR7IElBuypYBZqlSim2TsleCnjcLE3S93f6BFmYW5biINYyZB0bny0yWWv2zLCGrRQcNyRbRilAuJotXW0ZTEeTxQpNDpKP49Zo-2FSEs5eId11kByA4kcBQQK7g-3D-3D), **Rich Freeman [Analyst]** said "the one pricing model everyone in this industry now agrees is dying is the per-seat model," citing "healthy growing businesses with declining headcounts." Fewer seats, capped tokens. The swing factor: does consumption revenue outrun seat bleed before multiples de-rate further?

## Stocks in play

- **ADBE** *(direct):* *Bull:* cheap, buying back, "$25 billion stock authorization," ~8–12x cash flow off a ~$190 low ([The Synopsis, 7/3](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOhNxy5PLyEJDi8PlQjthXX6S4g55Co1sbKqtEzDSnkI7Vo19-2BkSWJ3f0-2FiIrUrpkR-2BmN5m0erd6uxehAH5Zcq8kqqzcKK2zaTPw6Q4s3U-2BcQg-3D-3Dv4Hb_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbWHk4RGVVGQEpZR890xCB2Oy5ywNr8pe2wbRiZVmG6cnWQ7TS7zrMjhNSwI-2F5Ghf0dBpZJO5dFfRX8ntF-2BRmujJcwsBAm0kkBE86W2DsfpFPTlteHE3omUgrqdCrI-2FOcbQSj6J1sMVEFpNSazbmNrzxNzum6RVWfgKCeMQrHcu1WA-3D-3D) [Analyst]). *Bear:* "10th consecutive quarter" of decelerating ARR, price hikes delayed until after the CEO exits, no permanent CEO/CFO, "melting ice cube." *Catalyst:* who takes the price increase.
- **CRM**, *Bull:* the data layer, "going headless" via MCP. *Bear:* seat-priced into a headcount-flat world. *Catalyst:* Agentforce consumption-ARR.
- **DDOG** *(no direct):* *Bull:* already consumption-priced, more agent activity = more volume. *Bear:* if customers cap tokens, metered revenue slows too. *Catalyst:* NRR.
- **TEAM** *(operator):* *Bull:* 154B-connection context graph as a moat. *Bear:* opening the graph to ChatGPT/Cowork dilutes Rovo's seat capture; still seat-priced. *Catalyst:* Rovo consumption metrics.
- **HUBS** *(no direct):* *Bull:* SMB "source of truth"; SMBs resist leaving per-seat. *Bear:* most exposed to the sub-10-person AI-native squeeze Asana flagged. *Catalyst:* Breeze attach + NRR.
- **ASAN** *(operator):* *Bull:* work graph as the enterprise "self-learning brain." *Bear:* CPO concedes sub-10-person accounts can now "hack around" with agents; smallest, most seat-dependent. *Catalyst:* AI Studio monetization.
- **MNDY** *(dark):* *Bull:* furthest along on hybrid seat + AI-credit pricing. *Bear:* SMB seat base, de-rated cohort. *Catalyst:* first hard AI-credit consumption print.

## Read-throughs

- **Seat-heavy SMB SaaS (HUBS, ASAN, MNDY):** Help Scout is the tell, buyers cling to per-seat, so the book erodes slower than headlines imply, but "declining headcounts" cap the ceiling. Watch NRR, not logos.
- **Model/inference vendors:** OpenAI's ~50% cost cut and Amazon's forced token repricing mean labs protect margin, not pass it on, resellers get no automatic COGS tailwind, and Coinbase-style cuts show demand is now budget-elastic.
- **Multiple de-rating:** proof-based split, data-moat, consumption-durable names (CRM, DDOG) get the benefit of the doubt; bolt-on seat-counters stay "semi-dead money" until they print, even as estimates rise.

## What changed vs last week

The consumption narrative flipped from triumphant to contested: last week Confluent's CFO called usage pricing a "one-way door"; this week Help Scout showed customers slam it shut when given the choice. The token-cap theme widened (Uber/Meta/Microsoft → now Coinbase -50%, Ford, Priceline). Last week's Anthropic 38%→70% margin claim went quiet, replaced by OpenAI's ~50% inference-cost cut and Amazon's hourly→token repricing. **Airtime rotated hard:** Asana, Atlassian and Adobe got direct coverage, while **Monday.com, last week's only direct name, went completely silent**; DDOG and HUBS stayed read-through only. Still missing, four weeks running: an explicit AI-feature gross margin % and a hard NRR print from any of the seven.