When revenue is recognised per outcome rather than per seat, backlog and remaining performance obligations stop describing the future. That is a revenue operations problem long before it becomes an accounting one.
Somewhere in a renewal conversation this quarter, a customer is going to ask what they are actually committing to, and the honest answer will be: nobody knows. Not the seller, not the buyer, not the controller who has to book it. The contract will be denominated in resolved conversations, or qualified leads, or completed tasks — and the number of those the customer will consume next year is a forecast dressed as a commitment.
This is not a hypothetical. It is what several of the largest vendors in this category shipped between April and July.
What actually changed, with dates
On 13 April 2026, effective the following day, HubSpot repriced two of its agents away from activity and onto results. Its Customer Agent moved from one dollar per conversation to fifty cents per resolved conversation; the Prospecting Agent moved from a monthly charge per enrolled contact to one dollar per lead recommended for outreach. The company's own figures — a 65 per cent resolution rate and a 39 per cent cut in resolution time — are vendor-supplied and unaudited, and should be read as such. The pricing change is the fact.
On 27 May 2026, Salesforce reported Agentforce annual recurring revenue of $1.2bn, up 205 per cent, and did something more consequential in passing: it reported 3.8 billion Agentic Work Units delivered to date, up 111 per cent quarter on quarter, alongside 28.6 trillion tokens processed. The company has invented a unit of account — one discrete task completed by an agent — and started reporting it to the market next to ARR. Constellation Research's scepticism is worth repeating: a token generated is not a return earned.
On 1 July 2026, Gartner put a number on the structural version of this: $234bn of enterprise application software spend exposed to what it calls agentic arbitrage by 2030, roughly a fifth of enterprise software spend. The mechanism it describes is the one that matters here — agents complete work across systems, software becomes invisible to the end user, and "the link between user growth and revenue growth" breaks. Gartner's own framing is that this is a metamorphosis rather than an apocalypse. Either way, seats stop being the meter.
Per-seat pricing had one underrated virtue: it was knowable in advance. You could count the seats. Nobody has to forecast how many employees a customer will have next quarter — they tell you, and then they sign.
The accounting problem is now on the record
On 4 June 2026, Deloitte published a technology spotlight on the ASC 606 treatment of outcome-based agentic-AI pricing, and it is the most useful document in this whole conversation because it is written for people who have to book the revenue rather than sell the model.
The issues it identifies are structural, not edge cases. Is the vendor's obligation to stand ready, or to deliver each outcome? Does the variable-consideration constraint apply, and if so how much revenue can be recognised at all before the outcomes occur? Do the invoice practical expedient or the variable-consideration allocation exception apply? And — the one that should stop a revenue leader mid-sentence — contracts must define a successful outcome against objective criteria, or the revenue cannot properly be allocated.
Deloitte flags two consequences directly. Quarterly revenue becomes lumpy. And remaining performance obligations and backlog become hard to quantify when outcome volumes are undefined.
That second one is the story, and almost nobody in sales has read it.
Why this lands on RevOps first
RPO and backlog are not accounting trivia. They are the sanity check underneath nearly every forecast conversation that matters. Coverage ratios are argued against them. Board numbers are triangulated to them. When a forecast and a backlog disagree, the backlog usually wins, because it is the contracted number and the forecast is a story.
Remove the contracted quantity and that anchor goes. What is left is a pipeline forecast validated against a revenue base that is itself a forecast — of your customer's consumption, which you do not control, cannot see in real time, and which will be shaped by decisions inside their business that nobody tells you about.
Visual 1 — What changes when the meter moves
Dimension | Per-seat | Per-outcome |
|---|---|---|
Quantity at signature | Known and contracted | Estimated; the contract may specify a rate and no volume |
RPO / backlog | Reliable forward number | Hard to quantify — Deloitte flags this explicitly |
Forecast risk | Sits with the customer, who committed | Migrates to the vendor, who now forecasts someone else's usage |
Expansion signal | Headcount growth; visible, slow, legible | Consumption growth; fast, volatile, and indistinguishable from a one-off spike |
Churn warning | Non-renewal, months of notice | Usage decay, no notice, no event to trigger a play |
Comp timing | Booked at signature | Contested — commission on what, recognised when? |
Deal desk workload | Discount approval | Defining a testable "successful outcome" per contract |
How to read it: Every row moves risk or work from the customer to the vendor. That is the trade being made, and it is rarely priced into the decision to reprice.
The contrarian bit: this is usually sold as customer-friendly
Outcome pricing is presented as fairness — pay for value delivered, not for licences gathering dust. On the buyer's side that is broadly true, and the shelfware problem it solves was real.
The under-discussed half is that fairness has a direction. Moving payment to outcomes moves forecasting risk to the seller. Your revenue now depends on your customer's operational volume, which you have no visibility into and no right to. You have effectively taken a position on their business, and taken it without pricing it.
There is a corroborating silence worth noting. HubSpot reported Q2 on 5 August 2026: revenue $911.7m, up 20 per cent; 306,446 customers, up 14 per cent; average subscription revenue per customer $11,800, up only 4 per cent. Growth held — the repricing did not break the business. But the company disclosed no agent-revenue or AI-credit breakout. A vendor that repriced to outcomes in April was not yet publishing outcome revenue in August. Read that as early-days rather than evasion, but read it.
Someone has to define "successful"
The practical work lands in the deal desk, and it is work nobody currently owns. If revenue recognition requires objective criteria for a successful outcome, then every contract needs a definition that is testable, measurable in a system both parties can see, and robust when the customer disputes it nine months later.
A resolved conversation — resolved by whose judgement? A qualified lead — qualified against which criteria, evaluated when? These are not legal boilerplate questions. They are product, finance and commercial questions that currently fall between three functions, which is exactly how they end up decided by whoever drafts the order form under time pressure.
What follows
Write the outcome definition before the pricing page, not after. If finance cannot recognise revenue against your definition, you have built a discount, not a pricing model.
Stop treating RPO as the truth the forecast is measured against. Where contracts carry rates without volumes, say so in the forecast package rather than letting a shrinking backlog read as deterioration when it is a definitional artefact.
Build a consumption leading indicator now. With no renewal event to warn you, weekly usage trend by account is the only churn signal you will get, and it needs to exist before you need it.
Settle comp before the first outcome-priced quarter closes. Commission on estimated contract value rewards optimism; commission on recognised revenue delays payment past the point where it changes behaviour. Both are defensible; leaving it unresolved is not.
Salesforce reports again on 26 August. It will be the clearest read yet on whether agentic work units keep compounding, and whether a unit of account invented eighteen months ago has become a number the market prices. Either way, the seat is no longer the meter, and the forecast has not caught up.
Sources and method. A SalesHubMedia original. Deloitte, Technology Spotlight on outcome-based pricing for agentic AI, 4 June 2026 (ASC 606 treatment, RPO/backlog quantification). HubSpot company news, 13 April 2026 (agent repricing; resolution figures are vendor-supplied and unaudited) and HubSpot Q2 2026 results, 5 August 2026. Salesforce Q1 FY27 results, 27 May 2026 (Agentforce ARR, Agentic Work Units, tokens); Constellation Research on the Agentic Work Unit. Gartner, 1 July 2026 ($234bn figure is a Gartner projection to 2030, not a measurement). Salesforce Q2 FY27 reporting date per company announcement, 5 August 2026. No public 2026 dataset exists on outcome-pricing adoption rates or on AI-driven changes to sales compensation; we have not implied one. Journalism, not procurement advice. Corrections will be made openly on this article.
