robomotionio Sales Coaching
Three deterministic analyzers compute the rep's mechanical metrics; you, the agent, do the pattern-finding, scorecard, playbook, surprise finding, and 30-day plan. Every grade must cite specific metrics/quotes from the rep's own data.
v2.0.0 — depth upgrade.
analyze_calls.pynow computes a discovery-quality score (question volume + open-vs-closed ratio + MEDDIC/BANT topic coverage), objection-handling detection (objection cues + whether the rep acknowledged/clarified vs. steamrolled), and the won-vs-lost split now includes discovery score and open ratio. This SKILL.md carries the discovery & objection rubrics, the role-specific skill tables (SDR/AE/Founder), and the recommendation logic.
When to use
- "Coach me", "how can I improve my sales", "what am I doing wrong", "review my selling style".
- "Why am I losing deals?", "analyze my calls"; periodic self-improvement.
How to run
Run whichever analyzers match the data the rep has (degrades by availability — any one source works; emails+calls+pipeline is best).
Email patterns (deterministic, keyless CSV)
python3 ${SKILL_DIR}/scripts/analyze_emails.py --input ${WORKSPACE}/emails.csv \
--top-n 3 --output ${WORKSPACE}/email_metrics.json
CSV: template,sends,replies,positive_replies[,subject,body]. Returns per-template reply
rates + a top/bottom split (ranking templates with ≥100 sends).
Call patterns (deterministic, keyless transcripts)
python3 ${SKILL_DIR}/scripts/analyze_calls.py --input ${WORKSPACE}/calls.json \
--rep-name "Alex" --output ${WORKSPACE}/call_metrics.json
calls.json = array of {call_id, outcome, turns:[{speaker,text}]}. Returns talk:listen
ratio, rep talk %, longest monologue, question count + open/closed split, filler rate,
a MEDDIC/BANT coverage flag-set, a discovery_score (0–100), objection-handling
detection, and a won-vs-lost split (now incl. discovery score & open ratio). If only audio
exists, transcribe first (out of scope here) — normally transcripts are given.
Discovery score (0–100) = question volume (40 pts, full at ~12 Qs) + open-question ratio (25 pts) + qualification-topic coverage (35 pts across 8 topics). MEDDIC/BANT topics detected: pain · budget · authority · timeline · metrics · champion · decision_process · competition. Objection handling = for each prospect objection cue, did the rep's next turn acknowledge/clarify (or ask a question) vs. plow ahead?
Deal patterns (deterministic, keyless CSV)
python3 ${SKILL_DIR}/scripts/analyze_deals.py --input ${WORKSPACE}/deals.csv \
--output ${WORKSPACE}/deal_metrics.json
CSV: deal,stage,amount,outcome,created_at,closed_at,industry,size. Returns win rate,
win rate by industry/size (sweet-spot/kill-zone), avg cycle, and loss-stage distribution.
Scorecard rubric (per-skill grade anchors)
Grade each skill A–F against the rep's own metrics — never generic. Anchors:
| Skill | A (strong) | C (developing) | F (problem) | Primary metric |
|---|---|---|---|---|
| Discovery | score ≥70, open ratio ≥0.6, ≥6 topics | 40–69, ratio ~0.4 | <40, mostly closed Qs | discovery_score, open_question_ratio, topics_covered |
| Talk/listen | 35–45% talk on discovery | 46–60% | >65% talk (monologuing) | rep_talk_pct, longest_rep_monologue_words |
| Qualification | 6–8 MEDDIC/BANT topics | 3–5 | ≤2 (no budget/authority/timeline) | meddic_bant_coverage |
| Objection handling | handles ≥80% of objections | ~50% | steamrolls (<30%) | objection_handling_rate |
| Conciseness | filler <2/100w | 2–4/100w | >5/100w | rep_filler_per_100w |
| reply rate top-quartile of own templates | mid | bottom templates dominate sends | reply_rate, positive_rate |
|
| Deal execution | win rate ≥ peer, fast cycle | average | loses late-stage repeatedly | win_rate, loss_stage_distribution, avg_cycle_days_won |
Role-specific emphasis (weight the scorecard by the rep's role)
| Skill dimension | SDR | AE | Founder-led sales |
|---|---|---|---|
| Pipeline generation / prospecting volume | ●●● core | ●● | ●● |
| Email copy & reply quality | ●●● core | ● | ●● |
| Discovery depth (open Qs, pain) | ●● | ●●● core | ●●● core |
| MEDDIC/BANT qualification | ●● (budget/authority/timeline) | ●●● core | ●● (don't over-qualify early) |
| Objection handling | ●● | ●●● core | ●●● core (vision objections) |
| Talk/listen discipline | ● | ●●● core | ●●● core (founders over-pitch) |
| Multi-threading / decision process | ○ | ●●● core | ●● |
| Closing / next-step discipline | ● | ●●● core | ●● |
Read it as: an SDR's scorecard leans on email + prospecting + booking-the-meeting; an AE's leans on discovery + qualification + objection handling + closing; a Founder selling their own product almost always over-talks and under-discovers — check talk% and discovery_score first, that's the usual surprise.
Recommendation logic (metric → diagnosis → exercise)
| Metric pattern | Diagnosis | Coaching exercise |
|---|---|---|
open_question_ratio < 0.4 |
Interrogating, not exploring | Replace closed Qs with "what/how/why"; 5 open Qs per discovery call. |
rep_talk_pct > 65% |
Monologuing / pitching too early | 40/60 talk target; pause after each value statement and ask. |
topics_covered ≤ 3 (missing budget/authority/timeline) |
Deals stall late from no qualification | Run a MEDDIC checklist; never advance a stage without budget+authority+timeline. |
objection_handling_rate low |
Steamrolling objections | Acknowledge→clarify→reframe→proof; practice the top 3 objections. |
won_avg discovery_score ≫ lost |
Discovery is the differentiator for THIS rep | Make great discovery non-negotiable; reverse-engineer their best call. |
rep_filler_per_100w > 5 |
Verbal tics undercut authority | Record + count; pause instead of filling. |
Synthesize the coaching report (you, the agent)
Read the metrics JSONs and the underlying copy/transcripts. Produce: a sales scorecard (per-skill grades A-F + trend, weighted by the rep's role above), "what you're great at" (strengths first), prioritized "where to improve" (evidence + root cause + exercise + own-data proof + measurable goal), a personalized playbook reconstructed from the rep's top performers, the surprise finding (validate their self-assessed weakness against the data — the won-vs-lost discovery/talk split is usually where the surprise lives; self-assessments are wrong ~60% of the time), and a 30-day plan with check-back metrics.
Outputs
sales-coaching-[YYYY-MM-DD].md— scorecard, strengths, improvement areas, playbook, surprise finding, 30-day plan. Workspace + Agent Teams channel attachment.
Credentials / env
- Required: none for the CSV/transcript path — all three analyzers are keyless. No LLM key (the coaching synthesis is your job as the agent).
- Optional: if a source key is set → pull the rep's data live; if not → keyless
CSV/transcript exports (default).
LEMLIST_API_KEY/INSTANTLY_API_KEY(email), a call-tool API key (Gong/Fireflies/etc., transcripts), a CRM key (HUBSPOT_API_KEY/PIPEDRIVE_API_TOKEN/SALESFORCE_*, pipeline) each gate their own live source; without them, paste/export the corresponding data.
Notes & edge cases
- Degrades by data availability: pipeline-only = win/loss patterns only; minimum viable = any one source. Prioritize call coaching when transcripts exist (higher ROI per fix).
- Keep grading evidence-anchored — every grade cites a metric or quote from the rep's own data; no generic advice.
- Pure analysis — no scraping; cost is the rep's own tool data.