AI Coaching with Multi-Rep Prompt Customization: A Framework

TL;DR

A single team scorecard does not coach a team. It coaches an average rep who does not exist. Multi-rep prompt customization means each rep gets a scorecard weighted to their actual gaps — discovery for one, objection handling for another, multi-threading for the third — while the team rubric stays consistent. Demodesk's AI Coach supports custom scorecards per rep, per segment, and per call type, so coaching adapts to the individual development plan instead of the other way around.

The default take is wrong: one scorecard per team

Most AI coaching rollouts start with one scorecard. MEDDIC, BANT, or a custom 8-criterion rubric the VP wrote in a Notion doc. Every call gets the same score. Every rep gets the same feedback.

This produces a strange outcome. The top rep, who already nails discovery, keeps getting told “great discovery questions.” The struggling rep, who is fine on discovery but loses every deal at procurement, gets a 7/10 on the same call where the deal just died — because the rubric never asked about pricing pressure tactics. The scorecard is doing its job. It's measuring the wrong thing for that rep.

The assumption — one team, one rubric, one coaching standard — comes from a pre-AI world where consistency was the hardest problem. When a human manager was the bottleneck, standardizing the rubric was the only way to make scoring fair across reps.

That constraint is gone. The bottleneck used to be the manager's time. Now it's the rubric's fit.

Why uniform scorecards fail at the rep level

Three things break when you run one scorecard across a team of reps with different skills.

Coaching points stop being actionable.A rep who scores 9/10 on discovery and 4/10 on closing reads the team average — 6.5 — and shrugs. The actionable detail is buried in two sub-criteria. Most reps don't dig.

Top reps stop trusting the system.When the rubric flags things they already do well, they tune out. Demodesk is built rep-first — for the user, not the dashboard. A coaching tool that doesn't earn the rep's attention every call has already failed. Reps with 20 calls a week, like those at OMR Reviews, won't open a dashboard that doesn't tell them something they didn't know.

Development plans become abstract.“Improve discovery” is not a plan. “Ask 3 quantifying questions in the first 10 minutes — measured on every call this week” is a plan. Uniform scorecards can't produce the second kind of feedback because they're not asking the second kind of question.

This is the counter-case to the consensus: AI coaching does not work by making one rubric scale. It works by making many rubrics cheap.

The framework: three layers of customization

A working multi-rep coaching system has three layers, each customizable independently.

Layer 1: Team rubric

The team rubric is the baseline. MEDDIC, BANT, Challenger, SPIN, or your own framework. It scores every call from every rep against the same core criteria — 5 to 8 items.

This layer answers “are we running our methodology?” It's what the VP of Sales looks at in the aggregate dashboard. It's what makes pipeline reviews comparable across the team. Without it, individual customization becomes anarchy.

Layer 2: Segment overlays

Segment overlays add 2 to 4 criteria specific to the deal context. An enterprise scorecard asks about multi-threading and procurement readiness. An SMB scorecard asks about close-on-the-call rate and objection volume. A renewal scorecard asks about expansion signals.

Same rep, different segments, different overlays. The team rubric still runs underneath, so the VP sees consistent MEDDIC scoring across all deals.

Layer 3: Individual development criteria

This is where the framework departs from the default. Each rep gets 2 or 3 extra criteria tied to their current development plan, refreshed monthly.

• Rep A's plan: improve discovery depth. Custom criterion: “Did the rep ask at least 3 quantifying questions (revenue impact, headcount, current cost)?”
• Rep B's plan: handle pricing objections. Custom criterion: “When pricing came up, did the rep anchor on value or defend on price?”
• Rep C's plan: multi-threading. Custom criterion: “Did the rep ask who else is involved in the decision, and propose a next step that includes them?”

The team rubric runs. The segment overlay runs. The individual criteria run on top. The rep sees one report after every call, with the section relevant to their development plan highlighted.

This is the AI Coach's unlock. Not “AI that scores calls” — every competitor does that. AI that scores the call the rep needed scored, not the call the rubric was written for.

How to build individual development plans from AI Coach data

The development plan is the bridge between the scorecard and the rep's behavior. Without it, the custom criteria are just more numbers.

A working monthly cycle:

Week 1 — diagnose.Pull the rep's last 20 calls from AI Coach. Look at the team rubric scores by criterion, not the average. Find the lowest two. Confirm against win/loss outcomes — is the low score correlated with deals that stalled or lost?

Week 2 — write the plan.Two development goals per rep. Each goal needs a behavioral target (what the rep will do differently), a measurement criterion (what AI Coach will score), and a baseline number (current rate). Example: “Increase quantifying discovery questions from 1.2 per call to 3+ per call by end of month.”

Weeks 3–4 — coach to the criterion.Every call the rep runs, AI Coach scores the new criterion alongside the team rubric. The rep sees the score within minutes of hanging up. The manager reviews the rep's trend line in the weekly 1:1 — not the call, the trend. The 1:1 stops being “let me listen to a call and tell you what I think” and becomes “here is the pattern across 18 calls, here is what we adjust.”

End of month — re-diagnose.Did the criterion move? If yes, the plan worked — pick the next gap. If no, the plan was wrong. Either the behavior was wrong, the criterion didn't measure what you thought, or there's a deeper issue — territory, ICP fit, motivation — that coaching can't fix.

This is the 1:50 manager-to-rep coaching ratio in practice. The manager isn't reviewing 100 calls a week. They're reviewing 8 trend lines a week, with the scoring already done.

How Demodesk's AI Coach handles multi-rep customization

AI Coach is built around custom scorecards from the start. Four things make multi-rep customization workable instead of theoretical.

Custom scorecards per rep, segment, and call type. Build a MEDDIC scorecard for enterprise discovery, a Challenger scorecard for objection-heavy mid-market, and a custom 4-criterion scorecard for an SDR's qualification calls. Assign them per rep or per call type. Every call gets scored against the right rubric automatically.

Instant post-call scoring. The rep sees their score within minutes of hanging up. Not the next morning, not in the weekly 1:1. Coaching that lands while the call is still in working memory is coaching that sticks.

Per-rep dashboards reps actually open.The AE sees their own trend lines across the criteria that matter to their development plan. The manager sees the same data in aggregate. Same source, two views — without a separate enablement tool.

Fine-tuned on 10M+ real sales conversations.The scoring engine was trained on sales calls, not generic transcripts. Criteria like “did the rep handle the pricing objection by anchoring on value” work because the model knows what value-anchoring sounds like in B2B sales.

Tanso Technologies described the impact as “performance improvements driven by closer and more effective coaching, enabled through call recordings and transcripts, resulting in faster response times and more detailed follow-ups.” That's the multi-rep coaching loop running — call recorded, scored against the rep's criteria, manager reviews the pattern, next call shows the adjustment.

What this looks like for a team of 8

A sales manager with 8 reps:

• 1 team rubric (MEDDIC, 6 criteria)
• 3 segment overlays (enterprise, SMB, renewal)
• 8 individual development plans, 2 criteria each — 16 individual criteria

Total criteria AI Coach tracks: 25. The manager doesn't track 25 criteria. They review 8 trend lines, one per rep, each filtered to that rep's current development plan.

Time spent listening to calls: zero, except for deals where AI Coach flagged a risk that needed a human ear. Time spent coaching: weekly 1:1s, 30 minutes each, focused on patterns instead of one-off calls.

That's how the manager-to-rep ratio shifts from 1:10 to 1:50. Not because coaching got worse — because scoring stopped being the bottleneck.