Consumption Comp Won't Kill ACV, It'll Absorb It

The pitch: pay reps on what customers actually use

Databricks pays its sales reps on realized consumption, not signed commitment. CJ Gustafson’s breakdown of the model in Mostly Metrics traces it from the first $1 million in revenue to $5 billion, and the logic is clean: if the business grows when customers run more workloads, then reps should get paid when workloads run, not when a contract gets inked. The commission follows the meter, not the signature.

Read the coverage and you would conclude bookings comp is finished for infrastructure companies. The argument writes itself. Annual contract value comp rewards the wrong things. Consumption comp fixes them. Any infra business still paying on bookings is behind.

That reading is wrong. Consumption comp is a good idea, but the coverage mistakes the finish line for the starting blocks. Databricks did not switch to consumption comp instead of building an ACV motion. It switched because it had already run an ACV motion for years and won. The model everyone is calling the replacement is the payoff of the thing it supposedly replaces.

The case against bookings comp is real, and it’s the strongest part of the story

The argument against bookings comp deserves a plain statement, because it carries the whole inevitability narrative.

Bookings comp rewards the signature over the outcome. A rep paid on committed ACV has every reason to push a customer toward a bigger number than they will ever consume. That produces shelfware: seats bought and never used, capacity committed and never burned. It produces sandbagging, where reps hold deals to hit next quarter’s accelerator instead of this quarter’s quota. It produces quarter-end theater, the discounting sprint that trains customers to wait until the last week of March for a better price.

Consumption comp cuts all of it. You cannot sandbag a meter. There is no incentive to oversell capacity nobody will touch, because the rep gets nothing until the customer touches it. The rep and the customer want the same thing, which is more real usage. On paper it is the cleaner design, and the alignment argument is correct.

So the inevitability story has a solid foundation. The problem is the conclusion built on top of it.

The inevitability story ignores what had to exist first

Databricks could make this switch because it had already solved the three hardest problems in usage-based selling. Every one of them was solved on ACV-era money.

The first is telemetry. Paying a rep on consumption requires knowing, precisely and in near real time, what each account consumed and which rep to attribute it to. That is a data infrastructure problem, and it is not free. A company at early revenue does not have clean, trusted, per-account consumption data wired into its commission engine. Databricks does, because it spent years and a large engineering budget building it. The comp model rides on telemetry that a smaller company cannot produce yet.

The second is forecasting. When you pay on committed ACV, the forecast is the contract. When you pay on consumption, the forecast is a prediction about behavior you do not fully control, across thousands of accounts with different usage curves. Building a forecasting apparatus that a CFO will stake a comp plan on takes scale, history, and a data science function. You need years of consumption patterns before the model is trustworthy. Databricks has those years. A Series B company has guesses.

The third is the land motion itself. Consumption comp drives expansion. It does almost nothing to drive the initial land, because a brand-new account consumes nearly zero in month one. Pay a rep purely on early consumption and you have told them not to bother chasing new logos, which is the opposite of what a growing company needs. Databricks can afford to weight comp toward consumption because its land motion is a solved, mature engine. Its reps are not fighting to get in the door; they are growing accounts that ACV-era selling already opened.

Strip those three things away and the model collapses. The Databricks plan is not a template a smaller infra company can copy. It describes what comp looks like after you win.

The real pattern is fund the land, then drive the expand

The sequence matters more than the endpoint, and the sequence runs in one direction.

Early-stage infrastructure companies need bookings comp because they need to fund the land. A signed commitment gives the business predictable revenue to plan against and gives the rep a clear, immediate reason to open new accounts. In the phase where the whole game is getting into enough accounts to matter, paying on commitment is the correct incentive. It buys you the installed base that everything else depends on.

Then, as accounts mature and usage data gets trustworthy, you layer consumption comp on top to drive the expand. Now the rep has a reason to go deep in accounts they already landed, to find new workloads, to grow the meter. This is the phase Databricks is in, and it is the phase where consumption comp works best.

Skip stage one and you starve the growth you are trying to design for. A young company that pays only on consumption tells its reps to ignore new logos and nurse a base too small to nurse. It gets a beautifully aligned comp plan attached to a revenue line that never reaches the scale where alignment pays off. The consumption model is not a shortcut past the land motion. It is the reward for having completed it.

There is a fair objection here: some infra products are so bottoms-up, so self-serve, that reps never really run a traditional land motion at all. Usage lands itself, and sales enters late to expand. For those companies, consumption comp from early days can make sense. But that is a distribution model, not a comp philosophy, and it is the exception. Most infrastructure businesses that carry a real sales force earned their base through committed deals first. For them the sequence holds.

What operators should actually build

The mistake is treating this as a switch to flip company-wide. It is not a switch. It is a dial, and the dial is set per account, not per company.

Design a hybrid plan that moves commission weight from commitment toward consumption as accounts mature. A new logo pays the rep primarily on the initial commitment, because the job in a new account is to land it. As the account ages and usage telemetry becomes reliable, the plan shifts weight toward realized consumption, because now the job is to grow the meter. The same rep, working two accounts at different stages, is paid on different things in each. That matches the incentive to the work the account actually needs.

This also means the infrastructure has to come first. You cannot pay on consumption you cannot measure per account and attribute per rep. Before any of the comp weight shifts, the telemetry and the forecasting have to exist and be trusted. Data trust is the precondition, not a nice-to-have. If the finance team and the reps do not believe the consumption numbers, no comp plan built on them survives the first disputed check. Whether the buyer is a procurement team or, increasingly, an automated system provisioning its own usage, the meter only works when everyone trusts what it reads.

Databricks did not kill ACV comp. It ran ACV comp until it built the base, the telemetry, and the forecasting that let consumption comp work, and then it let consumption comp absorb the weight that commitment comp had carried. This is the same shift from seat-based to outcome-based revenue playing out in the comp plan, and it only lands once usage compounds into data the business can trust. The lesson for everyone earlier in the arc is not to copy the endpoint. It is to build the stage that makes the endpoint reachable.