This is The Agentic Commercial Model, Season 2: the build. Issue 02 set out the operating model as a loop. Issue 2.5 built the org to hold it. This is the instrument both of them run on.
In January 2026 Stripe acquired Metronome for around a billion dollars. In June, Salesforce signed to acquire m3ter, the London metering and rating platform, and folded it into Agentforce Revenue Management. Neither company bought a competitor or a customer base. Both bought the ability to measure what software actually delivers.
Patrick Collison called metered pricing the native business model for the AI era. Salesforce was blunter still, stating the acquisition brings high volume mediation, metering and rating capabilities natively to Agentforce, so enterprises can bill with the usage and outcome-based models the AI era requires.
The reasoning matters more than the price. Stripe Billing tops out at roughly a thousand events per second while modern AI companies need over a hundred thousand. That is not a feature gap, it is an architectural one, and Stripe chose to buy the architecture rather than rebuild it. Salesforce reached the same conclusion about its own quote to cash stack six months later.
Read both as a signal about your business. Two of the companies that record and process the world's commercial transactions concluded, within six months of each other, that the measurement layer was worth buying rather than building. You are the one who has to prove what your product delivered. The layer is worth more to you than it was to them.
However measurement is not metering, and almost every company confuses the two. Metering counts what was consumed. Measurement establishes what was delivered. Metronome's own research found 85% of software companies have adopted usage-based pricing in some form, which means 85% are counting consumption. Almost none are measuring value.
Value-Realised Retention is the instrument that closes that gap. This issue is the specification.
The unglamorous companies who just became strategic
Worth pausing on who is being bought, because it says something about where advantage now sits.
DigitalRoute has spent two decades doing usage data mediation, metering and intelligence for over 400 clients, much of it rooted in telco, an industry that solved consumption billing before SaaS existed. LogiSense has built usage rating and billing in the same tradition. m3ter was founded specifically to solve usage-based pricing at scale without forcing companies to replace the CRM and ERP they already run, and CB Insights named it a leader among fifteen platforms including Stripe, Zuora and Chargebee before Salesforce moved.
None of these were fashionable companies. Mediation and rating is plumbing, and plumbing does not get keynotes. However the value proposition of every one of them has exploded, because the capability they quietly built for telcos and infrastructure businesses is now the capability every software company needs and almost none possess.
The analyst read on the m3ter deal put it precisely. Most enterprises selling software are still running legacy CPQ and billing systems designed for seat-based pricing, while their own products have become AI-driven and usage-heavy. They are building custom metering layers, stitching Stripe together with homegrown tools, or leaving money on the table because they cannot accurately track consumption.
That is the market. Telco worked this out in the 1990s, SaaS spent twenty years not needing to, and agentic AI has made it urgent overnight.
Component one: the verifiable outcome
Everything starts with a definition, and the definition has to survive an argument with your customer twelve months from now.
An outcome is verifiable when three things are true. It is defined before the contract starts, agreed by both sides in writing. It is detected by a system event rather than asserted by a person. And it is falsifiable, meaning there is a clear condition under which the outcome did not occur.
Most companies fail the second test. They define an outcome, then measure it through a health score a CSM updates quarterly from memory. That is not detection, it is opinion with a number attached.
The practical work is smaller than it sounds. Take the workflow your product replaces. Identify the completion event, the moment the customer's job is done. Define the exception, the condition that invalidates it. Instrument both. That is the outcome.
One hard rule from the metering world that applies directly. Stripe notes that configured meters cannot be changed once live, except for the display name, so production meter names should be treated as durable design decisions. The same discipline applies to outcome definitions. Change the definition mid-contract and every historical number becomes uncomparable, which means the instrument stops working precisely when you need the trend.
Component two: margin weighting
An outcome you cannot cost is a number you cannot trust.
The arithmetic is simple and brutal. If your support agent costs fourteen cents per successful resolution, a ninety-nine cent outcome price works. If your research agent costs four dollars twenty per accepted brief and you charge two dollars, then every unit of growth makes the business worse. Usage growth becomes a margin problem.
This is the component that separates VRR from every retention metric that came before it. NRR counts revenue and says nothing about whether the revenue was profitable. In a seat world that was tolerable because marginal cost was near zero. In an agentic world the marginal cost of delivery is real, variable, and rising with model complexity.
So every outcome in the instrument carries its delivered cost. Tokens, tool calls, retries, cache misses, model mix, the lot. You do not put that on the customer's invoice. You track it internally and roll it into the billable unit, because the customer wants a price and you need a margin.
The output is a contribution-weighted view of retention. Which accounts are growing profitably, which are growing expensively, and which are quietly becoming your most demanding loss-makers while your dashboard congratulates you on the expansion.
Component three: usage depth as a live signal
The first two components tell you what happened. The third tells you what is about to.
Usage depth is the breadth and frequency of engagement across workflows, teams and use cases. It is the only signal that distinguishes the customer who is optimising, using you more efficiently and getting healthier, from the customer who is disengaging and about to leave. Both show up as falling consumption. They are opposite trajectories.
It is also the moat, measured. Every additional workflow a customer runs through your product deepens the calibration that makes leaving expensive. Tracking usage depth is not a health exercise, it is a defensibility metric.
The cadence matters more than the sophistication. This signal is worthless quarterly and transformative daily. Instrument it to flag stalls automatically, feed it into the weekly value standup, and act on it in weeks rather than discovering it in a renewal post-mortem.
The product ledger
One architectural decision underpins all three components, and getting it wrong is expensive.
Your source of truth lives inside your product, not inside your billing system. The product records the events, the outcomes, the costs and the depth. Clean events then flow out to Stripe, Orb, Metronome or whatever you bill with. Skip the product ledger and every dispute becomes archaeology, every reconciliation is a negotiation, and you cannot answer the only question that matters, which is what did we actually deliver.
Orb approaches this with SQL-defined usage metrics and pricing simulation, which is the right shape. Paid AI goes further and ties agent cost, delivered value and customer facing ROI evidence together, which is closer still to what VRR needs. However none of these tools will define your outcome for you. They meter what you tell them to meter. The definitional work is yours and it cannot be outsourced.
What this buys you
Build the instrument and four things change immediately.
You can price on outcomes, because you can prove them. You can pay your people on realised value, which is Issue 06. You can see decay a year before NRR does. And you can put a number in front of a customer that they cannot argue with, which is the strongest negotiating position in commercial software.
Skip it and everything else in this season is theatre. The loop cannot turn without a signal. The org cannot hold a metric that does not exist. Outcome pricing without verified outcomes is just consumption pricing with better marketing.
Deloitte has already flagged that tying fees to agent results creates new measurement and revenue recognition questions. That is the price of admission, not a reason to avoid it. The companies that answer those questions first will define what value means in their category, and everyone else will be selling against their scoreboard.
Stripe and Salesforce both bought the ability to count. The ability to prove is worth considerably more, and nobody is going to sell it to you.
Build it yourself, or keep guessing.
Next issue: Winning the Machine. The GTM rebuild for a buyer who forms the shortlist inside a model before your CRM ever fires.
First published in the Agentic Commercial Model newsletter on LinkedIn, August 18, 2026. Read the original on LinkedIn.