AGT-701 — Rep Performance & Coaching
Layer 7: Measurement & Feedback · Role-parameterized coaching across AE, AM, CSM, SDR
L7 · Agent 01
Built · v31
Owns: RepSkillAssessmentLog, RepCoachingLog
v31 — promoted from Specced to Built. AGT-701 is the entry point to L7 measurement layer. With L8 fully Built (recognition, billing, payment data trustworthy) and AGT-407 v23 ripple delivering call_owner_role for role-filtered ConvIntelligence reads, AGT-701's role-parameterized coaching architecture has all the deterministic source data it needs. Manager approval gate on coaching delivery is unchanged from v21 spec.
Purpose
AGT-701 is the individual-level performance and coaching agent for the GTM OS. It operates across four roles — AE, AM, CSM, SDR — using a single shared architecture with role-specific metric inputs. The agent detects performance patterns, recommends corrective actions for human approval, and escalates persistent gaps to management. It does not replace manager judgment; it surfaces signal systematically so managers are never operating blind.
Role parameterization
One agent, four role configs. The output skeleton is identical across roles: performance summary → pattern diagnosis → recommended actions → escalation flags. What populates each section is role-specific. Role is determined by RepActivity.role at runtime — no separate agent instances required.
| Section | AE | AM | CSM | SDR |
|---|
| Performance summary | Quota attainment vs pace, forecast accuracy | NRR by cohort, renewal rate | TTV achievement, health trajectory | Meeting acceptance, opp conversion |
| Pattern diagnosis | Stage conversion gaps, deal velocity vs benchmark | Health score trend, expansion conversion | Onboarding adherence, escalation rate | Response rate by step, talk time ratio |
| Skill gaps | Objection handling, product knowledge, pricing/negotiation | Objection handling, product knowledge, pricing/negotiation | Product knowledge, sales methodology | Objection handling, discovery question quality |
| Recommended actions | Role-specific coaching actions — human approval required before delivery |
| Escalation flags | Fires when defined thresholds are breached or patterns persist |
Metric inputs by role
AE
| Metric | Source table | Notes |
|---|
| Quota attainment vs pace | QuotaRetirementLedger, QuotaStore | Compares closed + committed to time-weighted quota pace |
| Stage conversion rates | Opportunities | By stage pair vs segment benchmark |
| Deal velocity vs segment benchmark | Opportunities | Days per stage vs segment median |
| ConvIntelligence: objection handling + next-step quality | ConvIntelligence | Trailing 30-day pattern from AGT-407 scoring |
| AGT-405 deal advance scoring | OpportunityBriefLog | Did rep execute stated stage objectives? |
| Forecast accuracy (rep commit vs closed) | QuotaRetirementLedger | Rep commit accuracy fed to AGT-703 as distinct output |
| Product knowledge gaps | ConvIntelligence, RepSkillAssessmentLog | Detected from call patterns; written to RepSkillAssessmentLog |
| Pricing/term negotiation gaps vs segment benchmark | QuoteLog, RepSkillAssessmentLog | Discount depth, term exceptions vs benchmark |
AM
| Metric | Source table | Notes |
|---|
| Health score trend across book | CustomerHealthLog | Directional trend, not single-point score |
| Expansion conversion rate | ExpansionLog | Plays opened vs closed vs suppressed |
| Renewal rate + renewal opp open timing | Opportunities, ChurnRiskLog | Flags AM who opens renewal opp late relative to proximity window |
| QBR cadence adherence | QBRLog | QBRs scheduled and completed vs accounts requiring one |
| AGT-603 QBR outcome quality | QBRLog | Post-QBR outcome field logged by CSM |
| NRR by account cohort | ChurnRiskLog, ExpansionLog | Segment × vertical cohort comparison |
| Objection handling | ConvIntelligence | AM-owned calls; trailing 30-day pattern |
| Product knowledge gaps | ConvIntelligence, RepSkillAssessmentLog | Detected from call patterns |
| Pricing/term negotiation gaps vs segment benchmark | QuoteLog, RepSkillAssessmentLog | Expansion deal discount depth vs benchmark |
CSM
| Metric | Source table | Notes |
|---|
| TTV achievement vs target | OnboardingLog, Accounts | ttv_achieved_days vs ttv_target_days from AGT-601 |
| NPS/CSAT trend | CustomerHealthLog | NPS/CSAT dimension trend across CSM's book |
| Onboarding milestone adherence | OnboardingLog | Milestone completion on time vs target per AGT-601 schedule |
| Health score trajectory across book | CustomerHealthLog | Directional trend; flags CSMs with deteriorating books |
| ConvIntelligence: CSM call sentiment + talk ratio | ConvIntelligence | CSM-owned calls; sentiment trend + talk ratio vs benchmark |
| Escalation rate | CustomerHealthLog | Frequency of Critical/High tier accounts in CSM's book |
| Product knowledge gaps | ConvIntelligence, RepSkillAssessmentLog | Detected from call patterns |
| Sales methodology gaps | RepSkillAssessmentLog | Expansion motion execution quality |
SDR
| Metric | Source table | Notes |
|---|
| Meeting acceptance rate | CadenceEventLog | Meetings booked vs touchpoints sent |
| Sequence performance by cadence type | CadenceEventLog | Response and acceptance rates by cadence type (cold, warm, ABM, re-engage) |
| Response rate by step | CadenceEventLog | Step-level response rates to identify drop-off points |
| Conversion to qualified opp | Leads, Opportunities | Meetings booked → AE-qualified opp rate |
| Outreach volume vs activity targets | CadenceEventLog | Volume compliance vs targets in CompPlans/activity expectations |
| AGT-303 cadence intelligence flags | CadenceEventLog | Risk flags and recommendations generated by AGT-303 |
| Objection handling | ConvIntelligence | Discovery call objection patterns vs team benchmark |
| Talk time ratio vs team benchmark | ConvIntelligence | SDR talk ratio on discovery calls vs team median |
| Discovery question quality | ConvIntelligence, RepSkillAssessmentLog | Question depth and qualification coverage detected from transcripts |
Cadence
| Trigger | Output | Recipients |
|---|
| Monthly digest — runs on first business day of each month | Full performance summary + pattern diagnosis + recommended actions for prior month | Rep (coaching artifact) + Manager (pattern summary) |
| Daily flag — critical threshold breach detected | Point alert with specific metric, current value, threshold breached, recommended action | Manager only — rep not alerted on daily flags |
| Manager on-demand — explicit request | Full rep performance view, same format as monthly digest | Manager only |
Monthly digest is the primary coaching artifact. Daily flags are exception-only — they should not be frequent. If daily flags are firing repeatedly for the same rep, the escalation path activates.
Differentiated recipient model
Rep receives
Monthly digest only. Contains: performance summary vs benchmarks, pattern diagnosis (what the agent observed), recommended actions (presented as suggestions pending manager approval — rep sees the recommendation after manager confirms, not before). Rep does not receive daily flags or escalation notices.
Manager receives
Monthly pattern summary (roll-up across their book, not individual rep detail). Daily flags on threshold breaches. Escalation notices when patterns persist. Manager on-demand view available at any time. Manager approves recommended actions before they are surfaced to reps.
Coaching posture
AGT-701 is both diagnostic and prescriptive. Pattern detection is always active. Recommended actions are always generated. The distinction is in delivery: recommendations are held for manager approval before surfacing to reps. The escalation layer activates when approved recommendations go unresolved.
Standard flow
1
Pattern detected — agent identifies a performance gap from metric inputs (e.g., stage 2→3 conversion 18pts below segment benchmark for 4 consecutive weeks)
2
Recommended action generated — specific, actionable coaching recommendation written to RepCoachingLog with status pending_approval
3
Manager notified — pattern summary delivered with recommendation. Manager approves, modifies, or dismisses.
4
Rep notified — approved recommendation surfaced in next monthly digest. Dismissed recommendations are logged but not surfaced.
Escalation flow (pattern persistence)
1
Week 1 — pattern flagged, recommendation generated, manager notified
2
Week 3 (same pattern, no improvement) — automatic SLM escalation. No manager discretion. RepCoachingLog.escalated_to_slm = TRUE. SLM receives pattern summary + full 3-week history.
3
Post-escalation — SLM owns resolution. Agent continues monitoring and flags if pattern persists beyond escalation.
Pattern persistence clock resets on measurable improvement — not on a recommended action being approved. Improvement is defined as the flagged metric moving in the correct direction by ≥10% vs the prior measurement period.
Escalation thresholds
| Condition | Roles | Escalation target | Notes |
|---|
| Attainment <50% with <6 weeks remaining in quarter | AE | Manager + SLM concurrently | Severity warrants skipping standard manager-first routing |
| Same pattern flagged 3+ consecutive weeks, no improvement | All | SLM | Pattern persistence clock; resets on ≥10% directional improvement |
| ConvIntelligence quality score below threshold for 30+ days | All | Manager | Threshold = bottom quartile of role cohort on trailing 30-day score |
| Critical health drop across majority of AM/CSM book | AM CSM | Manager + SLM | Majority = >50% of accounts in rep's book drop one tier in same period |
| Manager on-demand escalation request | All | SLM | Always available; manager provides context note |
Skill gap detection model
Skill gaps (product knowledge, pricing/negotiation, objection handling, sales methodology, discovery question quality, talk time ratio) are not self-reported — they are detected from patterns in ConvIntelligence and QuoteLog data and written to RepSkillAssessmentLog. This table persists skill gap signals over time, enabling trend detection (improving vs deteriorating) and cohort comparison (rep vs role benchmark).
Schema dependency: RepSkillAssessmentLog is a net new table introduced in v21. Fields: rep_id, role, assessment_date, skill_dimension, gap_detected (bool), severity (low/medium/high), evidence_source (ConvIntelligence / QuoteLog / CadenceEventLog), trend (improving/stable/deteriorating), cohort_percentile.
Schema — tables written by AGT-701
| Table | Type | One row per | Key fields |
|---|
RepSkillAssessmentLog |
Net new (v21) |
Rep × skill dimension × assessment date |
rep_id, role, skill_dimension, gap_detected, severity, evidence_source, trend, cohort_percentile |
RepCoachingLog |
Net new (v21) |
Rep × coaching event |
rep_id, role, coaching_date, pattern_detected, recommended_action, approval_status, approved_by, escalated_to_slm, escalation_date, resolution_date, improvement_detected |
Schema — tables read by AGT-701
QuotaRetirementLedger · QuotaStore · CompPlans · Opportunities · CadenceEventLog · ConvIntelligence · OpportunityBriefLog · DiscoveryBriefLog · CustomerHealthLog · ChurnRiskLog · ExpansionLog · OnboardingLog · QBRLog · QuoteLog · Leads · Accounts · TerritoryDefinitions
RepActivity table assumed as a cross-cutting activity log. If not yet in schema, AGT-701 requires it in v21 — fields: rep_id, role, activity_date, activity_type, account_id, contact_id.
Downstream consumers
| Consumer | What they receive |
|---|
| AGT-703 Win-Loss & Forecast Accuracy | Rep-level forecast accuracy (commit vs closed) as a distinct coaching input |
| AGT-704 Business Review Orchestrator | RepCoachingLog patterns surfaced in WBR (team-level) and MBR (persistent gaps + escalations) |
| Manager | Monthly pattern summary across book + daily flags + escalation notices |
| SLM | Escalation notices with full pattern history |
Design guardrails
Rep never receives a coaching recommendation before manager approves it. Human-in-the-loop is non-negotiable at the delivery gate.
Escalation to SLM at week 3 is automatic — manager cannot suppress it. This is intentional. If the pattern has persisted three weeks without resolution, it has become an organizational issue, not a single-manager issue.
AGT-701 does not write to CompPlans, QuotaStore, or any comp-bearing table. Coaching signal is advisory only. Any comp or quota adjustment triggered by coaching insight routes through AGT-101/102/104 with full approval chains.