AI Observability Platform Selling to the VP of AI Engineering — 60-Min Training

Direct Answer

AI Observability Platform Selling to the VP of AI Engineering is a 60-minute training for AEs running $80K–$650K ACV cycles against LangSmith, Langfuse, Arize, Braintrust, Helicone, Datadog. Qualify against the three-buyer reality (VP AI Engineering, Director of ML Platform, CISO), run discovery on trace volume + eval-in-production + drift + cost, demo against the customer's actual LLM traffic, trap-set the multi-year renewal at month 12.

Built on MEDDPICC + Force Management.

Section 1 — Why AI Observability Selling Is Different (5 min)

Customers buying AI observability are already running production LLM workloads. Three buyers, technical bar.

• VP AI Engineering funds.
• ML Platform Director picks platform.
• CISO governs trace data privacy.

End with Mark Roberge's rule: *"Sell production telemetry depth, not generic APM."*

Section 2 — The 60-Minute Discovery (15 min)

1. Opening (3 min): "Walk me through your LLM production stack — providers, traces, eval setup."
2. LLM spend coverage (10 min): "What % of LLM API spend flows traces into your observability platform?"
3. Eval-in-production adoption (10 min): "Running LLM-as-judge on production traffic? 50%+ best-in-class."
4. Drift detection (10 min): "Monitoring embedding drift, refusal rate, tool-call patterns?"
5. Integration breadth (8 min): "OpenAI, Anthropic, Google, LangChain, LlamaIndex — all native?"
6. Cost discipline (7 min): "What % of LLM spend is observability infrastructure?"
7. Renewal posture (5 min): "Existing contracts and renewal dates?"

Section 3 — The POC That Wins (15 min)

Failure modes to ban. Sandbox-only POCs. No eval-in-production. No drift detection demo.

Wins to coach. Customer's real production traces ingested. Eval-in-production scoring on sample traffic. Drift signal delivered mid-pilot.

End with Andy Paul's rule: *"Show the customer their LLM issues caught earlier."*

Section 4 — Handling the Incumbent (10 min)

Counter-move 1 — Eval-in-production wedge. *"Does your incumbent run LLM-as-judge on production sample?"*

Counter-move 2 — Drift detection wedge. *"Embedding + refusal-rate drift signals?"*

Counter-move 3 — Integration breadth wedge. *"Native OpenAI + Anthropic + Google + LangChain + LlamaIndex?"*

Section 5 — Pricing Conversation (10 min)

Landmine 1 — Per-trace vs. Per-customer pricing.

Landmine 2 — Multi-year discount. 12–18%.

Landmine 3 — No procurement-only.

Section 6 — The Trap-Set for Renewal at Month 12 (5 min)

Trap-set 1 — LLM spend coverage 80%+ within 6 months.

Trap-set 2 — Eval-in-production adoption above 50%.

Trap-set 3 — Drift detection live across 5+ signal types.

Trap-set 4 — Joint VP AI dashboard in QBR.

Close with Jeb Blount's rule.

FAQ

LangSmith or Braintrust? LangSmith for trace + LangChain-native; Braintrust for eval-in-production.

Datadog competitive? For existing Datadog customers, yes.

Open-source Langfuse? Yes for cost-sensitive.

Eval-in-production target? 50%+ customer adoption.

LLM spend coverage target? 80%+.

Sources

• LangChain — LangSmith Customer Outcomes
• Langfuse — Open-Source Reference
• Arize AI — Phoenix Reference
• Braintrust — Eval-in-Production Reference
• Helicone — Proxy-Based Reference
• Datadog — LLM Observability Reference
• Force Management — MEDDPICC
• Mark Roberge — Sales Acceleration Formula
• Andy Paul — Sell Without Selling Out
• Jeb Blount — Fanatical Prospecting