You get the behaviour you pay for, not the behaviour you asked for

The two comfortable answers, and why both fail

Paying on closed won sounds rigorous. It ties the SDR to revenue, which feels like alignment. In practice it drags their attention downstream, away from the one thing only they do: creating net-new conversations. AEs quickly learn that an SDR paid on closed won can be recruited into deal admin, chasing paperwork and rescheduling stakeholders, and the SDR agrees because that is where their money is.

Then there is the lag. With a long cycle, work done in February pays out the following January. Nobody stays motivated by a number that far away. Worse is what happens when promotion is tied to the same figure. Career progression then depends on the AE's skill, the market they were handed, and how deal size fell. A mediocre rep on a warm campaign can carry more ARR against their name than a strong one opening a difficult new segment. You end up promoting the luckier person and knowing it.

The opposite answer, paying on meetings set or meetings held, fails faster and more visibly. It rewards volume, so you get volume: booked slots that fit no profile, prospects who never intended to attend, and AEs who stop trusting anything on the calendar. Pay per booking and you have bought bookings, not pipeline.

Pay on the middle milestone

The milestone worth paying on sits between the two: far enough down the funnel that quality is proven, near enough that the SDR is paid this quarter and still controls the outcome.

Two structures work in the field. The first pays entirely on the SQO or SAO: the meeting has to convert from stage 1 to stage 2 after a genuine discovery call. The second splits it, roughly half on the meeting being booked against a written ICP definition and half on the SQO. The split reduces the lag without giving up the gate, which matters most for teams hiring reps who cannot wait ninety days for their first variable payment.

Closed won is not always wrong, it just needs a small role. One approach ties around 30% of OTE to closed won and makes attainment against it a formal part of the career framework. Another deliberately keeps closed won out of promotion decisions entirely and instead adds roughly 1% of the deal as a bonus at close. Both make the SDR care about quality. The first makes revenue a career gate and accepts the fairness problem that comes with it. The second keeps the incentive without letting the AE's quarter decide the SDR's. Pick the consequence you are willing to own.

Define the qualified meeting before you price it

None of the above works if the definition is soft. A four-part test holds up well because every part is checkable by someone who was not on the call.

Some teams skip the bespoke test and simply use their stage 2 entry criteria: ICP fit, a qualification framework applied, next steps recorded. That is cleaner if your stages are already honest, because the comp plan and the forecast then describe the same thing.

Who gets to say no

At effectively every company, the AE validates. That is right, because the AE is the one who has to work the deal. It only functions on two conditions. The criteria have to be specific enough that two people reading the same notes reach the same verdict, and the reps have to have genuinely bought into them rather than been sent them. The SDR must be able to escalate a rejection to management and get a hearing.

The inversion is worth checking. Teams assume the risk is AEs waving through weak meetings to keep the peace. Often the bigger problem is the reverse: AE hesitancy to move anything to SQO, because their own stage discipline is being watched. Audit both directions or you will fix the wrong one.

Clawbacks are rarer than the internet suggests, and they cost more in atmosphere than they recover in cash. The stronger safeguard is reviewing at QBRs how many of each rep's accepted meetings actually progressed to later stages or closed. That is a coaching conversation rather than a payroll adjustment, and it catches the same problem earlier.

SPIFs are for strategy, not for the core plan

Keep the core plan boring and use SPIFs to buy specific behaviour for a short window. Extra percentage on a newly launched product. A team reward for the most meetings booked before a conference, judged on the prospect actually turning up. Cold-calling days with small prizes: an early finish, an extra day working from home. A bonus percentage for a multi-year commitment or a certain attach rate. A quarterly pipeline-generation week that front-loads meetings into the first fortnight, with SPIFs of roughly $250 to $1,000 a person rewarding both volume and AE-plus-SDR joint sourcing into named target accounts.

The pattern across all of those: none has to be expensive, and none has to touch commission. A day off and a leaderboard moves a Tuesday more than another point of variable.

Where AI actually helps here

The hardest part of a quality-based plan is not designing it. It is auditing it consistently, every week, when everyone is busy and the disputed meetings are the ones nobody wants to reopen. That is the job worth handing to a model: check the recorded criteria against the call transcripts and emails, and flag the inconsistencies for a human to rule on. Some teams now run exactly this as a standing weekly job. It does not decide anything. It finds the cases where the record and the stage disagree.

Here is our current definition of a qualified meeting:
{paste the wording as it exists today, however loose}

Here is our ICP and our accepted-title list:
{paste}

Rewrite it as a pass/fail rubric with four to six
criteria. For each criterion give me:
- The test, written so two people reading the same
  call notes reach the same verdict
- One example that clearly passes
- One example that clearly fails
- Two edge cases my team will argue about, with
  your recommended ruling and the reason

Do not soften a criterion to make it easier to pass.
If a criterion in our current wording cannot be
tested from a call record, say so and tell me what
evidence would be needed instead.

Rubric: {paste the rubric from above}

Here are 30 meetings accepted as SQO last month, with
the call notes or transcript and the CRM record for
each:
{paste or connect}

For each meeting return one of: AGREE, DISAGREE, or
CANNOT JUDGE.

For every DISAGREE, name the criterion that fails and
quote the exact line from the record that made you
decide. No paraphrasing.

For every CANNOT JUDGE, say what is missing. Put these
in a separate list at the end. Do not guess to avoid
an empty answer, and do not mark something AGREE
because the rest of the record looks strong.

Finish with the count in each bucket and whether the
disagreements cluster on one rep, one AE, or one
criterion.

An SDR has escalated a rejected meeting. I have to
rule on it.

Rubric: {paste}
The record: {call notes or transcript, email thread,
CRM fields, and the AE's stated reason for rejecting}

Write the case both ways, using only what is in the
record:
1. The strongest argument that this meets the rubric,
   quoting the evidence
2. The strongest argument that it does not, quoting
   the evidence
3. The single fact that would settle it, and where
   that fact would live if it exists
4. Whether the rubric itself is ambiguous here

Do not recommend a verdict. I am deciding.

The honest limit: a rubric applied by a model is only as good as the rubric. If the AE and the SDR do not actually agree on the criteria, automating the judgement does not resolve anything. It just makes the argument happen faster and with better citations. Fix the agreement first, then automate the checking.

The practical constraint is access. Pasting thirty transcripts once proves the idea and nobody does it twice. Running this weekly means the model reads the CRM and the call records directly, which is the setup covered in Salesgear's writeup on connecting Claude to CRM data for sales workflows. If you are deciding what to trust it with first, their checklist for putting AI into a sales process is a reasonable order of operations.

If you are rewriting the plan this quarter

1. Write the qualified-meeting definition first. Price the milestone second. A plan built on a definition nobody can apply is a dispute schedule.
2. Choose your middle milestone deliberately: SQO alone, or the split with booking. Say out loud which lag you are accepting and why.
3. Decide what closed won does. Career gate, small bonus, or nothing. All three are defensible. Drifting into one is not.
4. Set the escalation path before the first dispute, and audit AE hesitancy as seriously as you audit SDR optimism.
5. Run the rubric audit by hand on one month of accepted meetings before automating anything. If the output surprises you, the plan was never measuring what you thought.

Comp decides what your SDRs aim at. The work of actually hitting it, building a list worth calling and earning the reply, is covered in prospecting with Claude. What happens to those meetings once the AE owns them is in managing pipeline.

Background reading: how sales teams are using Claude day to day