If AI Did Half the Work, Who Gets Paid for the Deal?

Sales compensation encodes a theory of who created the value. For thirty years that theory was stable: a rep sourced or received an opportunity, worked it, and closed it, so the rep got paid. Every plan mechanic, from SDR splits to territory design, descends from that assumption.

Agents now perform recognizable portions of sourcing, qualification, research, and follow-up. The assumption is no longer clean, and comp is where that shows up first and most painfully, because comp is the one document the field reads carefully.

The three mistakes companies are making right now

Mistake one: raising quota because a tool was purchased

The most common and most damaging move. Leadership buys a productivity tool, assumes a lift, and raises quota in the next planning cycle to capture it. The lift has not been measured, the tool has not been adopted, and the field concludes that any productivity gain they generate will be taxed.

That conclusion is rational and permanent. Once a team believes efficiency gains get converted into quota, they stop surfacing gains. You will have destroyed your own measurement system in exchange for one planning cycle of optimism.

Mistake two: paying for AI-sourced pipeline the same way as rep-sourced

If an agent books a meeting and an AE closes it, paying full sourcing credit to the AE is generous but survivable. Paying an SDR team full credit for agent-generated volume is not, because it disconnects effort from earnings in a way the rest of the team can see.

Mistake three: changing everything at once

Comp changes are trust events. A plan that changes quota, territory, and credit rules in the same cycle, citing AI, will be read as a pay cut regardless of the math. The sequence matters as much as the design.

Comp is the only document the field reads word for word. Treat every change as a message about whether their effort still counts.

The principle to design around

Pay for judgment and relationship. Automate and do not pay for volume and mechanics.

That single line resolves most edge cases. An agent that drafts an email has not created value worth a commission split. A rep who read a committee correctly and changed the strategy has. Where the human contribution is genuinely reduced, the honest answer is not a smaller commission rate. It is a different role definition, handled through role design rather than through quiet plan mechanics.

A sequence that does not detonate trust

1. Measure the lift before you price it. One full planning cycle of measurement with no comp consequence. Publish that you are doing this and why. The credibility you buy is worth more than the quarter you delay.
2. Change capacity assumptions before quota numbers. If AI genuinely expands coverage, adjust territory size or account load first. That is a visible, explainable change tied to work, not an unexplained target increase.
3. Adjust the mix, not just the number. If sourcing is increasingly automated, shift weight toward the outcomes humans still control: multithreading, expansion, win rate against a named competitor, forecast accuracy.
4. Grandfather one cycle. Whatever changes, protect the current cohort for a period. The cost is small. The signal is enormous.
5. Publish the reasoning. Not the spreadsheet, the logic. Teams accept difficult plans they understand and reject generous plans they suspect.

Territory and capacity, the part nobody models

The real structural question is not commission rate. It is how many accounts a rep can meaningfully cover when research and follow-up are partly automated.

If the honest answer is more, you have a choice: same headcount with more coverage, or fewer people. Most companies make that choice implicitly through attrition and backfill decisions, which is the worst version, because the field watches it happen and draws conclusions nobody stated.

Make the choice explicitly and say it out loud. Coverage expansion with stable headcount is a defensible, motivating story. Silent non-backfill is a resignation trigger for exactly the people you most want to keep.

What to protect at all costs

Two things. First, the belief that additional effort produces additional earnings. Every mechanic that weakens that link costs you more than it saves. Second, the willingness of the field to tell you what is actually working with AI. That information flow is worth more than any single planning cycle's quota capture, and it is the first casualty of a plan the team reads as a tax on their own productivity.