Salesforce just admitted the per-seat model has a ceiling
On its Q2 fiscal 2027 earnings call, reported on August 27 by CX Today, Salesforce did something more interesting than beat a revenue number. It told the market it intends to sell software by the outcome rather than by the human.
Marc Benioff put it plainly on the call: "We're still trapped in some ways in old per user pricing models." The company said customers now want to buy AI five different ways — by user, by agent, by consumption, by transaction outcome, and by business outcome. That is not a footnote in a pricing page. That is the vendor that invented modern SaaS procurement telling you that the unit it has billed you on for two decades is running out of road.
Outcome-based pricing matters to anyone running a sales team for a reason that has almost nothing to do with Salesforce's stock price. When your GTM tools stop charging per rep and start charging per action, per resolution, or per booked meeting, the marginal cost of outbound stops being zero. And the moment volume has a price tag, the spray-and-pray motion that survived 2024 and 2025 on flat-rate seats stops penciling out.
This post covers what changed, what the underlying buyer data actually says, what to renegotiate before your next renewal, and how to run a prospecting motion that gets cheaper rather than more expensive under metered pricing.
What "AI seat compression" actually means
The phrase showing up across vendor earnings calls this quarter is AI seat compression, and it is worth being precise about it because the panic version and the real version are different.
The old commercial logic
Software priced per seat assumed a straightforward chain: more work meant more people, more people meant more logins, more logins meant more recurring revenue. Every SDR you hired was a line item for your sales engagement tool, your data provider, your dialler, your enrichment vendor, and your CRM. Vendors grew when you grew headcount.
That model had a hidden gift for buyers. Once you had paid for the seat, using it more cost you nothing. A rep who sent 200 messages a week and a rep who sent 2,000 cost the same. Which is exactly why the industry drifted toward volume: the incremental send was free.
The new commercial logic
If an agent qualifies leads, drafts messages, researches accounts, and updates records without a human sitting in the CRM all day, the vendor cannot capture that value through logins. So it moves the meter.
Salesforce's numbers show how far this has already gone. The company reported Agentforce ARR of $1.5 billion, with customers driving 3.2 billion Agentforce Work Units in the quarter, up 97% quarter over quarter, and 2,000 paying Agentforce customers moved into production, up 70% quarter over quarter. Half of Agentforce bookings, the company said, came from customers refilling credits after usage. Refill behaviour is the tell. Credits get refilled when something is running in production, not when a pilot is being demoed to a steering committee.
Benioff also noted that Agentforce use of apps through Model Context Protocol and command-line calls surged sixfold. Software is increasingly being consumed by agents and systems, not by people clicking screens. You cannot bill a login that never happens.
The buyer data says this is not a Salesforce quirk
It would be easy to read one earnings call as vendor positioning. The buyer-side research says otherwise.
According to Futurum Research's 1H 2026 Enterprise Software Decision Makers survey, fewer than one in five enterprise software buyers still prefer classic per-user pricing. Of those surveyed, 43% prefer consumption-based models and 27% favour outcome-based structures. Futurum's blunt conclusion: vendors restricted to seat-only pricing risk immediate disqualification.
The same research documents who has already moved. Zendesk charges for successful AI-driven issue resolutions rather than AI seat licences. Intercom applies a similar model to its Fin agent. Decagon prices around resolved customer interactions. On the hybrid side, Adobe, Salesforce, ServiceNow, UiPath, Automation Anywhere and Workhuman are all blending subscription, consumption and outcome-linked components.
| Pricing model | What you are billed on | Where the risk sits | Typical in GTM tools |
|---|---|---|---|
| Per seat | Named users | Buyer (you pay for unused licences) | CRM, sales engagement, legacy data vendors |
| Consumption / credits | Actions, enrichments, tokens, work units | Buyer (usage can spike) | Enrichment, AI research, agent platforms |
| Outcome-based | Resolved cases, qualified leads, booked meetings | Shared, in theory | Support AI, emerging AI SDR contracts |
| Hybrid | Base subscription plus variable component | Split, negotiated | Where most enterprise vendors are landing in 2026 |
Analyst projections point the same direction. Gartner's widely reported forecast is that at least 40% of enterprise SaaS spend shifts toward usage-, agent- or outcome-based models by 2030, with seat-based revenue share falling from 21% to 15%. Treat the precise numbers as directional. Treat the direction as settled.
Why this changes outbound, not just procurement
Most commentary on outcome-based pricing is written for CFOs and procurement leads. The more interesting consequence is operational, and it lands squarely on the sales floor.
When every action has a price, volume stops being free
Under flat seat pricing, the constraint on outbound volume was reputational: burned domains, LinkedIn restrictions, dead-list decay. Those constraints were real but abstract. They showed up in next quarter's reply rate, not this month's invoice.
Under metered pricing, the constraint becomes financial and immediate. Every enrichment, every AI-written message, every research pass on an account that was never going to buy shows up as a line on a usage report that someone in finance reads.
This is a good thing, and teams that plan for it will come out ahead. It forces a question most outbound orgs have avoided: what is our cost per qualified conversation, and how much of our spend is being consumed by prospects who never had a reason to reply?
The reply-rate arithmetic gets unforgiving
The volume era already had a maths problem. Autobound's State of AI Sales Prospecting 2026 found per-rep monthly outbound rising from a human baseline of roughly 1,150 messages to an AI-augmented mean of about 7,400, while raw reply rates fell from 4.7% to 2.9%. A 6x increase in output bought a roughly 40% decline in response quality.
While seats were flat-rate, you could argue that trade was fine in absolute terms: more replies overall, even at a worse rate. Under consumption pricing, that argument collapses. You are now paying 6x the platform cost to generate output that converts at 60% of the previous rate, and you are paying it in cash rather than in reputation.
The same dynamic explains the AI SDR churn problem the category has been living with. Industry analysis this year has repeatedly put first-year cancellation for AI SDR tools in the 50 to 70% range — roughly double the turnover of the human reps they were meant to replace. Volume-maximising systems on metered infrastructure produce a cost curve that outruns the pipeline curve, and the renewal conversation ends badly.
Fewer, better actions become the economically rational motion
Here is the part that should reframe your 2027 planning. Outcome-based and consumption-based pricing quietly rewards exactly the motion that signal-based warm outbound has argued for on quality grounds.
If you pay per research pass, you want to research people who might actually buy. If you pay per message, you want to send to people with a live reason to hear from you. If you pay per booked meeting, you want meetings that hold and convert, not meetings booked off a misleading subject line.
Volume outbound and metered pricing are structurally incompatible. Signal-based outbound and metered pricing are structurally aligned. That is not a marketing claim; it falls straight out of the unit economics.
| Motion | Actions per qualified conversation | Behaviour under metered pricing |
|---|---|---|
| Broad list-based outbound | High — most spend lands on non-buyers | Cost per meeting rises with volume |
| Intent-data-led outbound | Medium — better targeting, noisy signals | Cost improves, attribution stays murky |
| Signal-based warm outbound | Low — actions concentrated on live triggers | Cost per meeting falls as targeting tightens |
Platforms built around capturing warm intent — profile views, post engagers, competitor mentions, hiring signals, pain-point posts on Reddit, LinkedIn and X — start from a small qualified queue rather than a large cold list. Updately was designed around that constraint: enrich and score against ICP first, research deeply on the survivors, then send a small number of genuinely personalised messages inside platform limits. Under seat pricing that was a quality argument. Under metered pricing it is a budget argument too.
What to renegotiate before your next renewal
The move to outcome and consumption pricing is not automatically good for buyers. As CX Today's analysis noted, outcome contracts can get expensive if the vendor captures too much of the upside, or if the buyer signs up to metrics that ignore quality. The definition of the outcome becomes the entire contract.
Before you sign anything with a variable component, get these written down:
- The outcome definition, in operational language. "Qualified lead" is not a definition. "Contact matching the agreed ICP filter who books and attends a meeting of at least 15 minutes" is closer. Ambiguity always resolves in the vendor's favour at invoicing time.
- The baseline. What was the metric before the tool was deployed? Without a baseline you cannot separate the vendor's contribution from your own demand gen, seasonality, or a good quarter.
- Attribution rules and exclusions. Inbound leads that the agent happened to touch. Existing opportunities. Reactivated closed-lost. Decide now who owns those, because the vendor has an opinion.
- Quality thresholds and clawbacks. If a booked meeting no-shows or is disqualified within 48 hours, is it billable? Get this in writing.
- A usage ceiling. Consumption pricing without a cap is an open cheque. Ask for a hard cap, an alerting threshold at 70 to 80% of the commit, and the right to pause.
- Rate protection on the variable unit. Per-unit prices set in a land-grab phase have a habit of rising at renewal once the workflow is embedded. Lock the unit rate for the term, with a defined cap on increases.
- Exit mechanics. Where does the data live, what is exportable, and what notice is required? Annual auto-renewal clauses in this category commonly require written notice 60 to 90 days ahead.
Watch for the hybrid trap
Most vendors are landing on hybrid: a base subscription plus a variable usage or outcome fee. Sensible in principle. The trap is that the base does not shrink when the variable is introduced. You end up paying roughly the old seat cost plus a new meter.
When a vendor moves you to hybrid mid-contract, the correct question is not "what is the new rate?" It is "what did you take out of the base to make room for the meter?" If the answer is nothing, you have absorbed a price rise dressed as a modernisation.
How to run outbound when actions are metered
Four practical changes, none of which require you to rip anything out this quarter.
1. Shift budget from seats to signals
Audit your stack for tools you pay for by headcount that deliver value by targeting quality. Broad contact databases are the usual candidate: you pay per user for access to everyone, then use a fraction of a percent of it. Signal capture — who viewed your profile, who engaged with a relevant post, who just posted about the problem you solve, who just started hiring for the role that implies your use case — produces a smaller, warmer, cheaper-to-work queue.
2. Score before you enrich, not after
Under seat pricing, teams enrich everything and filter later because enrichment felt free. Under consumption pricing that order is backwards and expensive. Apply cheap ICP filters — firmographics, headcount band, tech signals you already have — before you spend a research credit. Deep research on 60-plus data points is worth paying for on a prospect that scored well; it is money set on fire on one that did not.
3. Change the metric you manage to
Activity metrics were built for the flat-rate era. Messages sent, connections requested, sequences launched. None of them mean anything when each unit has a marginal cost.
Move your weekly reporting to cost per qualified conversation and cost per held meeting, blended across tooling spend and rep time. It is a harder number to produce and an uncomfortable one to look at the first time. It is also the only metric that survives contact with a usage-based invoice.
4. Keep a human on the send
Fully autonomous, volume-maximising agents are the worst possible fit for metered pricing: they consume units fastest, produce the lowest-quality output, and carry the highest platform risk. The deployments that survive their first renewal consistently pair AI research and personalisation with a human owning the actual outreach. That model got cheaper to justify the moment volume acquired a price.
The counter-argument, taken seriously
Not everyone thinks per-seat pricing is dying, and the sceptics have a point worth hearing.
Per-seat pricing is simple, predictable, and easy to budget. Finance teams like it precisely because it does not surprise them. Outcome-based pricing shifts variance onto the buyer's P&L unless the contract is negotiated carefully, and most buyers negotiate these terms for the first time with a vendor that has done it a hundred times.
There is also a measurement problem. Outcome pricing works cleanly in support, where a resolution is observable and binary. It works far less cleanly in sales, where a "qualified lead" is a judgement call, attribution is contested, and the gap between a booked meeting and closed revenue contains most of the actual difficulty. Expect sales tooling to move to consumption pricing well before it credibly moves to outcome pricing, whatever the marketing says.
And the realistic destination is hybrid rather than pure outcome. Vendors need revenue predictability as much as you need cost predictability. The negotiation is about where the line sits, not about whether the meter exists.
Takeaways for the next 90 days
The specific news — one vendor's earnings call — matters less than what it confirms. The commercial unit of GTM software is shifting from the human seat to the action and the outcome, and it is shifting because AI agents broke the link between headcount and value.
What to do about it:
- Inventory your variable exposure. List every GTM tool with a usage, credit or outcome component today. Most teams have more than they think, buried in AI add-ons bought in 2025.
- Model your stack cost at 3x current AI usage. If that number is frightening, you have a targeting problem, not a pricing problem.
- Put outcome definitions in writing before the renewal conversation, not during it. Bring your definition to the table first.
- Rebalance from list volume to signal quality. Every action you do not waste on a non-buyer is now money as well as goodwill.
- Instrument cost per qualified conversation this quarter. You will need the baseline before your first metered renewal, not after it.
The teams that struggle with this shift will be the ones running high-volume outbound on the assumption that sending is free. It is not free any more, and the invoice will say so. The teams that do well will be the ones already sending fewer, better-targeted messages to people with a live reason to reply — because that motion was always more effective, and is now also cheaper to run.
Sources: CX Today on Salesforce's Q2 FY27 call, Futurum Research on outcome-based and hybrid AI pricing, Autobound's State of AI Sales Prospecting 2026.