FoundersOperators, strategy & lessons
SaaStr's Jason Lemkin Tells B2B Companies Rebuilding for AI to Cut Per-Action Prices Sharply
SaaStr founder Jason Lemkin argues that B2B companies rebuilding for AI agents should sharply lower per-action prices and rely on agent volume to offset the cut, after SaaStr's own agents slashed calls to a vendor whose bill rose 40%.
From the SaaStr article of November 16, 2026 and cited sources: SaaStr's agents were calling one unnamed vendor 35,000 to 45,000 times a day. [1]
AI-assisted, not human-edited. Methodology Corrections Who owns this site
- What happened
- SaaStr's agents were calling one unnamed vendor 35,000 to 45,000 times a day. [1]
- Original source
- SaaStr article, independent perspective
- Brief by
- AT Pulse research desk AI-assisted; publisher and contact on the desk page
- When
- November 16, 2026 article published
- Covered
- Nov 16 edition
- Evidence
- 11 sources across 10 sites
- 1SaaStr: Make Your Product Cheapersaastr.com
- 2Linked video: If You’re Refounding Your B2B Company for…youtube.com
- 3TNW: Databricks $5B at $190Bthenextweb.com
- 4Pulse 2.0: Databricks run-ratepulse2.com
- 5MarTech: HubSpot rebuilds around agentsmartech.org
- 6SaaStr: Gabrisko 5 learningssaastr.com
RisingHow it’s measured
YouTube video · 32.9K views · 96 likes
Reach blends public engagement (views, likes, comments, shares, stars) from the platforms where the story or its original source lives: YouTube, Hacker News, GitHub, Substack and similar. A video and the article written from its transcript share one score, taken from the video. Platforms with no data are left out rather than counted as zero, and the age of a story never lowers its score.

SaaStr's agents were calling one unnamed vendor 35,000 to 45,000 times a day. After the bill rose 40%, they cut those calls by about 90%. These figures are self-reported, and no outside source confirms them. Lemkin says agents react to high per-call prices by batching calls, caching results and switching to other vendors. [1]
Lemkin's advice: price agent and API use separately from human seats, drop punitive overage fees, test 50–80% cuts on one tier, and track calls per customer per day. The math is demanding. An 80% cut needs five times the usage just to keep revenue flat. The post's 20x scenario is an illustration, not measured data. [1]
Lemkin's main precedent is Databricks. Its CRO, Ron Gabrisko, says customers named a $1,000 per-user fee as what held back wider rollout, and consumption rose sharply once users were free. But Gabrisko describes pricing as matching price to value: a consumption unit (DBUs) with add-ons priced as multipliers. He calls failing to capture value the biggest mistake, which is not an argument for deep discounts. [2] [6]
The wider market is moving away from pure seat pricing, though not toward blanket price cuts. BCG recommends hybrid pricing and event-level metering. HubSpot Credits bill by the volume of agent work. Reported survey data shows pure per-seat pricing falling from 21% to 15% of SaaS companies. [7] [8] [9]
one usage collapse at SaaStr, one anecdote about Databricks dropping a seat fee, and survey data showing seat pricing in decline.
Read the full assessment
Implication: when customers use agents, pricing pain may show up as falling call volume before churn, which puts pressure on seat-based revenue, NRR and sales-comp models.
Executive brief
SaaStr founder Jason Lemkin says one of SaaStr's vendors lost about 90% of its own usage. SaaStr's agents had been calling the vendor's system 35,000–45,000 times a day. When the bill rose 40%, the agents cut their calls by about 90% (SaaStr). His advice to B2B companies rebuilding around AI is to cut the price per action, often by a lot, so that agent volume makes up the revenue. His main precedent is Databricks dropping a $1,000 per-user fee. In the Databricks CRO's own telling, though, the lesson is matching price to value, not discounting.
What changed and event timeline
Databricks raises $5B at $190B
Databricks reported a run-rate above $7B, growing 80%+ year over year, and pitched products for enterprise agents (;).
HubSpot rebuilds around agents
At Unbound26, HubSpot repositioned its CRM around Breeze-coordinated agents and "Growth Context" ().
Gabrisko podcast
On SaaStr's AI CRO Confidential, Databricks CRO Ron Gabrisko said free user seats unlocked consumption (). The host mentions their previous recording, October 24, 2024, as "exactly 2 years ago" ().
BCG says SaaS must be "pivot-ready."
BCG recommends hybrid pricing, event-level metering, and preparing for agents that compare and choose vendors on their own ().
Lemkin publishes "make it cheaper."
He calls for separate low agent/API prices, no punitive overage fees, and testing 50–80% cuts on one tier ().
Capabilities and access
No model or release is involved. This is commentary on pricing.
Read the full section
No model or release is involved. This is commentary on pricing. Concrete mechanisms cited in the reviewed sources:
- Databricks charges in DBUs (Databricks Units), its consumption unit, with add-ons priced as multipliers on that unit (18:09).
- HubSpot Credits bill by the volume of agent work, not by human logins (Observer-Reporter/MarketMinute).
- SaaStr moved its marketing email from Marketo to an email API it describes only as cheaper, without naming it (SaaStr).
Technical analysis for researchers and developers
- Agents optimize against cost. Lemkin describes agents batching calls, caching results and routing to other vendors when per-call prices are high (SaaStr).
- An 80% price cut needs 5× usage just to keep revenue flat.
- BCG says vendors need event-level metering because token costs vary per customer (BCG). Lemkin's proposed health metric is calls per customer per day.
Read the full section
- Agents optimize against cost. Lemkin describes agents batching calls, caching results and routing to other vendors when per-call prices are high (SaaStr). For developers, that means publishing per-call prices an agent can model and setting rate limits that fit a real workflow.
- The elasticity math. An 80% price cut needs 5× usage just to keep revenue flat. The 20× (and so 4× revenue) in the post is an illustration, not measured data.
- Metering. BCG says vendors need event-level metering because token costs vary per customer (BCG). Lemkin's proposed health metric is calls per customer per day.
Claims and evidence
- SaaStr's own numbers (35k–45k daily calls, 40% bill increase, 90% cut) are self-reported. The vendor is not named, and no outside source confirms them.
- The Databricks $1,000 per-user fee: Gabrisko says customers named the fee as the reason they weren't rolling out wider, and consumption "went out the roof" once users were free (32:20).
- "$200B company": the last priced round was $190B, announced by Databricks (TNW). The 80% growth figure is also Databricks' own.
Read the full section
- SaaStr's own numbers (35k–45k daily calls, 40% bill increase, 90% cut) are self-reported. The vendor is not named, and no outside source confirms them.
- The Databricks $1,000 per-user fee: Gabrisko says customers named the fee as the reason they weren't rolling out wider, and consumption "went out the roof" once users were free (32:20). This is an anecdote with no figures attached.
- "$200B company": the last priced round was $190B, announced by Databricks (TNW). The 80% growth figure is also Databricks' own.
- Seat pricing is declining: reported survey data says pure per-seat pricing fell from 21% to 15% of SaaS companies (PYMNTS).
Context and prior work
- Databricks has priced on consumption since early on.
- He sees "millions of agents" querying data as a tailwind for Databricks (30:57).
- HubSpot folded Dharmesh Shah's Agent.ai into its core product (StartupHub).
Read the full section
- Databricks has priced on consumption since early on. Gabrisko calls it one of the company's smartest decisions and points to LLM vendors pricing in tokens (29:10).
- He sees "millions of agents" querying data as a tailwind for Databricks (30:57).
- HubSpot folded Dharmesh Shah's Agent.ai into its core product (StartupHub).
- Other analysts have made the same case against seat pricing (Channel Insider).
Limitations, safety and contested findings
- Databricks removed a seat fee on top of consumption pricing. It did not cut its per-unit price.
- BCG recommends hybrid models, not across-the-board cuts (BCG).
- Lemkin himself acknowledges inference costs and lower early margins.
Read the full section
- The precedent doesn't match the advice. Databricks removed a seat fee on top of consumption pricing. It did not cut its per-unit price. Gabrisko calls not capturing value "the biggest mistake" and frames pricing as matching price to value (32:52).
- BCG recommends hybrid models, not across-the-board cuts (BCG).
- Margins. Lemkin himself acknowledges inference costs and lower early margins.
- Transcript quality. The video transcript is automatic captions and may contain errors.
Business and practitioner implications
- Vendors: separate agent tiers from human seats, drop overage penalties, and track agent call volume, since Lemkin says pricing pain shows up in usage before churn.
- Buyers: agent cost audits can show where routing to a cheaper vendor pays off. SaaStr's Marketo exit is one example.
- Finance: NRR models (net revenue retention) and sales comp built on seat uplifts would need reworking.
Read the full section
- Vendors: separate agent tiers from human seats, drop overage penalties, and track agent call volume, since Lemkin says pricing pain shows up in usage before churn.
- Buyers: agent cost audits can show where routing to a cheaper vendor pays off. SaaStr's Marketo exit is one example.
- Finance: NRR models (net revenue retention) and sales comp built on seat uplifts would need reworking.
- Go-to-market: Gabrisko recommends low entry points and pay-as-you-go over large upfront commitments (33:00).
The source trail.
Sources (11)
SaaStr: Make Your Product Cheaper
saastr.comLinked video: If You’re Refounding Your B2B Company for AI, Make Your Product Cheaper. Probably Much Cheaper.
linked video auto captions
www.youtube.com