The Compensation Plan Has a Denominator Problem
Attribution, not headcount, is the load-bearing assumption in every variable compensation plan, and agentic go-to-market breaks attribution first. It…
A variable compensation plan is arithmetic with an assumption buried in it. It pays a rate on an outcome and assigns that outcome to a person through an attribution record: an owner field, a closed-won date, a name. The record held because the name and the labor were the same fact recorded twice. Agentic go-to-market severs that correspondence without touching the record. The name stays. The work moves. The rate was never priced against the signature. The rate was priced against everything a person did to reach it: the list, the research, the sequence, the meeting, the negotiation, the close. That bundle is the denominator. Agents now perform the first four. The opportunity still requires an owner, so a rep is named. The rate is still paid against all six. No row fails validation. The breakage has no error state, and it runs for quarters. A headcount reduction needs a budget line and a signature, so it waits for a planning cycle. Attribution changes the first time an agent runs. The plan is wrong before the org chart is. The same assumption was priced outward to the customer as the seat. The seat is the unit you charge. This is the unit you pay. The first answer is to change nothing. The rep named on the opportunity is paid for a cycle the system ran, and the plan becomes a territory annuity. Money does not follow advancement. It follows assignment. Enterprise sellers have been paid on inherited accounts since named accounts existed. That annuity was visible, and it was priced in at planning. Agents extend it across the whole funnel in one fiscal year, and the record looks identical on every account. A distortion that reconciles gets paid. The second answer is a reduced rate on machine-sourced pipeline, which pulls the incentive out of the last place a person is load-bearing. The stack sources, researches, and books. It does not decide which meeting is worth a concession, and that call moves more revenue than the meeting did. The third answer is multi-touch weighting, and every weight is a claim somebody can dispute. A scheme that costs more to administer than it redistributes is a tax collected in arguments. All three optimize inside a structure nobody designed. The structural answer is surgery on the denominator. Pay for what a person still uniquely does, and name it precisely. Which accounts to decline. Which concession to grant and at what price. Which exception to escalate. Everything the machine performs with no person in the path moves into cost of revenue. It arrives with an invoice. Each decision is priced against the result it produced, not the fact that it was logged. The rate has to move up. A smaller base at the old rate is a pay cut with new vocabulary. The team reads it that way within an hour. Round numbers make the shape visible. These are illustrative, not observed. A rep carrying $4M in bookings against $200K of variable pay is at five points. Move sourcing and sequencing to the machine, and the decisions a person still makes touch $2.5M. Hold the $200K and the rate is eight points. The money does not leave the team. It moves inside it. The failure arrives on a schedule, not as a surprise. No canonical example has finished running yet. This is inference from earlier corrections. A team automates the top of the funnel, and the plan stays as written because it was signed in January. Two quarters of commission are paid on machine output. The correction arrives mid-year as a rate cut, and it reads as punishment for the automation working. Then the sorting starts. The reps whose accounts the machine carried have the least reason to leave. The ones making the calls it cannot make have somewhere to go. Most companies treat the plan as policy, not as commercial architecture. We rewrite it when the unit it pays for and the work a person does no longer describe each other: naming the decisions the system does not make, instrumenting them against results, raising the rate rather than trimming the payout. The cheapest version happens in planning season, before anybody has modeled a number against a mortgage.