The problem
Every commission plan you have run rests on an assumption so basic it is rarely written down: the number on the contract at signature is the revenue. Quota credit, accelerators, payout dates — all of it hangs on a figure known and fixed the day the deal closes.
Consumption pricing deletes that figure. The contract now carries an estimate, and the revenue arrives in twelve monthly instalments that may total more, less, or almost nothing. AI products sharpen this further: usage depends on whether the customer’s own team adopts a workflow that did not exist when they signed.
The failure state: a rep closes a $200k estimated deal in January and is paid in February. The account ramps slowly, narrows its scope, and finishes the year at $70k. You have paid real cash against revenue that never arrived. The money is the smaller loss; the larger one is that you have told the rep, in the only language a comp plan speaks, that the estimate is the deliverable.
Both obvious fixes miss. A process step — deal-desk review of the estimate — gates a number nobody can verify for twelve months; the estimate gets slower to produce, not more honest. A commission tool computes the same wrong quantity faster, because the defect is in the plan’s definition of revenue, not in the arithmetic.
The estimate being wrong is not the problem. Estimates are supposed to be wrong. The problem is a plan with no place to put that error.
How it works
In Usage-Based Compensation Model: Aligning Compensation with Consumption (September 2024), Graham Collins sets out how to structure a plan when revenue is consumed rather than contracted. Three disclosures. This is vendor content: QuotaPath sells commission-tracking software, and the article closes by inviting the reader to schedule time with their team. It is a taxonomy, not a study, and reports no outcome data on any model. And it states no role for Collins; he introduces himself as QuotaPath’s Head of Partnerships in a companion piece two months later. For scale, the article cites OpenView’s 2023 usage-based pricing study, which found that three out of five SaaS companies then used some form of usage-based pricing.
Collins presents two as the purest ways to pay on usage-based plans.
1. Estimated revenue. Attainment is tied to projected spend over the contract period. The article’s example: if you project a customer will spend $100,000 over the next year, the rep earns $100K of quota attainment. Collins calls this very similar to a regular compensation plan, and says you should only do it if you are confident in your revenue model’s estimates.
2. Actual revenue. Quota credit accrues monthly against real consumption.
In this process, if I'm the rep, I close a deal, and every month for the first 12 months, whatever the customer spends is what I earn toward quota credit and commissions.
Collins names the cost directly: it would take a full year for the rep to earn credit, and the delay may hurt motivation or put new reps in a tough spot.
The remaining five he groups as middle-ground options, introduced as just a few of the countless ways companies pay reps who sell on consumption pricing.
3. Blended. The two are combined.
In this model, you reward partial quota credit at the front end based on estimated revenue with the opportunity to earn the rest, more, or less, throughout the year.
The worked example: on a $120k estimate, to pay 10% of deal value ($12k) you first pay 5% of the estimate ($6k), then 5% of usage for the next 12 months.
4. Commission triggers. The payout is released in instalments tied to events rather than to a proportion of revenue. In the example, a rep paid 10% of estimated revenue closes an $80k deal and earns $8k. Of that, 25% ($2k) is paid immediately, the next 25% once the customer is onboarded, and the final 50% once the customer hits 50% of their estimated spend for the year — or $40k. The article does not specify what counts as “onboarded” or who declares it.
5. True-ups. You pay a commission based on the estimate, then at the end of the 12 months you add to or subtract from that commission based on actual usage.
An example would be if someone closes an estimated $40k deal, they get paid 10% of that up front—$4k. Then, at the end of the year, if the customer actually only spent $20k, you would claw back $2k.
The article also notes a tolerance band below which no reconciliation happens:
Some organizations only do a true-up if the estimate is wrong by a certain amount, say +/- 20% off the estimate.
6. Guaranteed minimums. The customer commits to a minimum spend over the year, and reps are compensated on that amount. Collins notes this requires the company to buy into the strategy, and cautions that comp plans should not dictate company strategy.
7. Shorter durations. Most organisations pay on the first 12 months of a contract. Collins points out you could instead pay 20% on any revenue in the first six months, or 40% in the first three — and says most avoid it because spend varies through the year and takes time to ramp.
Collins does not recommend one, though he does frame estimated and actual revenue as the two purest forms and the rest as middle ground between them. His closing advice is that a clear and transparent usage-based comp plan is essential for team alignment.
Why it works — our read
Read as a list, this is seven ways to do one thing. It is not. It is seven different answers to a single question that consumption pricing forces and a subscription plan never had to ask: who carries the risk that this revenue does not show up?
What the taxonomy changes at the input is small and decisive. It converts an inherited default into a deliberate allocation. Estimated revenue puts the entire risk on the company. Actual revenue moves all of it to the rep. Blended splits it proportionally. Triggers stage it against milestones. A true-up defers the settlement of it to a single annual event. A guaranteed minimum pushes part of it onto the customer. A shorter duration shrinks the window in which it can materialise at all.
What that does to the mechanism is that each allocation prices a different behaviour, because a rep optimises the thing the plan measures and nothing else. Pay on the estimate and the estimate is what gets produced — optimistically, and with no downside. Pay on actuals and the rep starts avoiding accounts whose usage is genuinely variable — which, by construction, are the accounts with the most expansion headroom — while absorbing onboarding delays owned by someone else. Stage the payout against onboarding and implementation becomes the rep’s problem: excellent alignment if they can influence it, an arbitrary tax if they cannot.
Which moves the outcome from section one, because the failure there was never the bad estimate. It was a plan that paid the same regardless. Once some portion of the payout is contingent on what actually happens, the estimate stops being a free variable.
The transferable principle: when revenue arrives after the deal does, your payout schedule is a forecast you are making in cash — so decide whose forecast it is.
The counterfactual is the reason this matters. Keep last year’s subscription plan and you have not avoided the choice. You have made it silently: 100% of the risk on the company, and a standing incentive for your reps to be optimistic.
Where it breaks
The article names its own challenges — calculation complexity, forecasting fluctuating usage, data quality, and the year or more a customer can take to reach full consumption. These are the ones we would add.
- Clawbacks land on money that is gone. This is intrinsic to true-ups: the model works only if you can recover cash from someone who already spent and paid tax on it. Enforceability varies by jurisdiction, and we are not qualified to advise — get counsel before writing the clause. Symptom: the first clawback is quietly waived and the model becomes decorative.
- Triggers only align what the rep controls. If onboarding runs through a queue no rep can jump, a milestone payout is a lottery wearing an incentive’s clothes. Symptom: reps escalating about staffing rather than selling.
- Deferral costs you the reps. The immediate payout is a large part of why anyone takes a sales job. A plan that is perfectly risk-aligned on paper can lose your top performer to a competitor still paying at close. Symptom: your best rep asks what next year’s plan looks like.
- The meter is assumed, not given. Every model presupposes usage measured per account and attributed to an owner. For AI products with shared workspaces, pooled credits and silent expansion into adjacent teams, that is often untrue. Symptom: arguments about whose usage it was.
Don’t do this if your consumption variance is low and your median contract is small. The reconciliation overhead will exceed the misallocation you are fixing.
What has to be true at your company
Four preconditions. Someone owns comp design — RevOps or finance — with the authority to change a plan mid-year rather than only at the annual rewrite. Usage is metered per account, attributed to an owner, and visible at a latency shorter than your payout cycle; a monthly model needs monthly data, not a quarterly billing extract. Legal has signed off on the clawback or true-up language in every jurisdiction where you employ reps. And finance tolerates a commission accrual that moves during the year instead of settling at close.
One cultural precondition matters more than the four: reps must be able to watch their own accrual continuously, in a place they trust. A deferred plan a rep cannot see is a deferred plan a rep does not believe, and disbelieved compensation changes no behaviour at all.
Translating across company shapes: with three reps and a founder writing the plan, skip the taxonomy as a decision framework and use it as a menu — pick triggers or a true-up with a wide band, write it in one paragraph, and revisit in two quarters. At several hundred people the design is the easy part and the binding constraint is systems: the model you can actually run is the one your billing data supports this quarter. In a PLG motion where usage grows without a rep touching it, decide first which consumption is even commissionable — otherwise you are paying for product-led growth twice.
Try it this week
Pull the last 10 to 20 closed deals where you have both the estimate at signature and at least twelve months of actual consumption. For each one, compute actual divided by estimate. That is a spreadsheet, under an hour, one person, no approval needed and nothing to undo.
What you should see if this is worth pursuing: a spread. If more than about a third of deals land outside Collins’s +/- 20% figure — a third is our threshold, not his — you are already misallocating commission at a scale worth a design session, and the taxonomy becomes a real choice. Check the direction too: if the misses are mostly overestimates, you are overpaying; if mostly underestimates, you are quietly underpaying your best reps and the fix is more urgent than it looks.
And the stopping condition, which matters just as much. If nearly every deal lands inside +/- 20%, your estimates are good enough that your current plan is approximately correct. Conclude that, write it down, and do not run this project.