Rebuild coverage around agents, not headcount (L6)

The steps

1. 01

   Map coverage to outcomes instead of accounts per rep

   Tool: Salesforce

   For each segment write the outcomes coverage must produce, new pipeline, expansion revenue, retention, time-to-first-value. Then measure what each current role actually contributes to each outcome, with real numbers from the warehouse, not estimates. Some roles will not map cleanly to any outcome. That finding is the point of the exercise, and it is uncomfortable enough that most teams stop here. Do not stop here. • Owner: CRO plus RevOps • Tool options: your territory model plus the semantic layer from playbook 4 • Pitfall: stopping the exercise once you find roles that do not map cleanly to any outcome • Definition of done: an outcome-by-role matrix exists with a real number in every cell, sourced from the warehouse

2. 02

   Decompose each role into tasks and sort them three ways, with the team in the room

   List every task in the role, then sort each into one of three buckets: machine does it better, human does it better, or human must own it for trust reasons even if a machine could do it. Ask about tasks, never about jobs. "Which parts of your week would you hand off" gets honest answers. "Could AI do your job" gets defensive ones. Do it with the team present. A model of their job built without them will be wrong, and they will know it before you do. • Owner: Enablement plus front-line managers • Tool options: two-hour workshops with the people who actually do the job • Pitfall: building the task model without the team in the room • Definition of done: every task in the top three roles is classified with the team's agreement on record

3. 03

   Redraw the operating model on one page

   Redraw the motion with agents owning the machine-side tasks and humans concentrated where judgment, relationships and negotiation live. Name an owner for every step, human or agent. Be honest about the consequence: if a seven-step process now has agents covering four steps, the human has to be genuinely excellent at the remaining three. Mediocre does not survive this design, and that has hiring and development implications you should say out loud rather than discover in month six. • Owner: CRO plus enablement • Tool options: a one-page operating model document • Pitfall: leaving hiring and development implications unstated until month six • Definition of done: the model fits on one page, agent-owned and human-owned steps are visually separated, and every step has a named owner

4. 04

   Pilot on your least strategic segment for one full sales cycle

   Tool: LiteLLM

   Pick the least strategic segment, not the most enthusiastic one. Run for one complete sales-cycle length rather than one quarter, so you actually see closed-won rather than early-funnel optimism. Keep a comparable segment on the old model as a control. Instrument four things from day one: pipeline created, conversion by stage, cycle time, and customer satisfaction. The fourth one is the one that catches the failure mode nobody wants to name. • Owner: one segment leader • Tool options: the new model, running on the control plane from playbook 9 • Pitfall: piloting on the most strategic segment instead of the least strategic one • Definition of done: one full cycle is complete with all four metrics compared against the control segment

5. 05

   Rewrite comp and the career path before segment two

   Tool: Workday

   Answer the questions the pilot team is already asking privately. If agents generate pipeline, who gets credited. If a CSM covers three times the accounts with agent support, what is quota and what does promotion look like. If a role's task list shrank by half, what is the new role and what is it called. Unanswered comp questions sink this faster than any technical failure. The answers must exist in writing before segment two goes live, and they should be reviewed with the pilot team before they are announced. • Owner: CRO plus HR • Tool options: comp modeling in Workday, SuccessFactors, HiBob or Rippling • Pitfall: launching segment two before comp and career-path questions are answered in writing • Definition of done: comp and career-path documents are approved by HR and have been walked through with the pilot team

6. 06

   Scale one segment per quarter with a public side-by-side scoreboard

   Tool: Looker

   One segment per quarter, maximum. Publish the same five metrics for old-model and new-model segments side by side, including the customer-experience metric, and publish them even in the months where the new model looks worse. Transformation feels like going backwards right before it accelerates. The scoreboard is what keeps the room calm during the backwards part, and hiding a bad month is how you lose the mandate. • Owner: CRO • Tool options: a monthly review plus one dashboard in Looker, Tableau or Power BI • Pitfall: hiding a bad month on the scoreboard instead of publishing it • Definition of done: two segments are running the new model with side-by-side metrics published for both

7. 07

   Instrument the result

   Instrument: revenue per fully loaded GTM dollar, alongside customer satisfaction. Both, always. Either one alone gets gamed inside a single quarter. • Where it breaks: it gets run as a headcount reduction with an AI story stapled on. Talent leaves, institutional knowledge leaves with them, and the agents inherit a broken system. AI scales whatever system you have, good or bad. Second failure: piloting on the strategic segment, where one bad quarter is unrecoverable and the whole program dies with it. Third: comp stays untouched, so the org quietly reverts to the old behavior while the new org chart stays on the slide. • Visual guidance: a from/to diagram. Left: today's coverage as roles with accounts stacked under them. Right: coverage as outcomes, with agent-owned and human-owned steps in different weights and the trust-critical steps drawn heaviest.

Tools in this playbook

• Salesforce
• LiteLLM
• Workday
• Looker

Next playbooks

• AI literacy bootcamp for GTM (L2)
• Centralized AI prompt library (L2)
• Claude for Work shared projects (L2)
• Notion AI + Claude for company knowledge ops (L2)