AI-generated QBR decks (L3)

The steps

1. 01

   Map data sources and create schema

   To begin, the CS Ops leader and Product team must define the 'Source of Truth' for data extraction. You need to map out where Adoption, Support, and Risk data live. • Identify your Product Analytics tool (e.g., Mixpanel, Pendo, or a Snowflake warehouse) to extract seat utilization and feature-specific 'stickiness' metrics. • Locate your Support tickets (Zendesk or Salesforce Service Cloud) to pull volume and resolution time trends. • Open your CRM (Salesforce or HubSpot) to find the 'Account Notes' or 'Success Plan' fields where CSMs log qualitative updates. • Time Estimate: 4-6 hours. • Owner: CS Ops or Data Analyst. • Common Pitfall: Trying to pull everything. Start with only 5 core KPIs (e.g., DAU/MAU, open bugs, and last renewal date) to prevent the LLM from hallucinating on noisy data. • Definition of Done: A CSV or JSON export script (using a query like `SELECT account_id, feature_usage, ticket_volume FROM data_warehouse WHERE date > current_date - 90`) that produces a clean data row per customer.

2. 02

   Build the LLM generation prompt

   Now, translate your standard QBR slide titles into a structured prompt template for the LLM. You are moving from a blank page to a fixed schema: Adoption, Value Delivered, Support Health, and Future Roadmap. • Use an LLM orchestration tool or an API (OpenAI GPT-4o or Claude 3.5 Sonnet). • Define the 'System Prompt' to act as a 'Strategic Customer Success Executive.' • Example Prompt Fragment: 'Given the following data: {{usage_data}} and {{support_logs}}, write three bullet points for a slide titled "Business Impact." Focus on cost savings and efficiency gains. Use a professional, data-driven tone. Do not use corporate jargon like "synergy".' • Owner: CS Ops. • Time Estimate: 3 hours. • Common Pitfall: Not providing 'Few-Shot' examples. Give the LLM 1 or 2 examples of what a 'Perfect' slide looks like so it learns the desired brevity and tone. • Definition of Done: A prompt template that consistently generates a text-heavy draft reflecting the underlying data.

3. 03

   Automate the draft assembly pipeline

   Connect your data extraction script to the LLM via an automation platform like Zapier, Make.com, or a custom Python script. • In Zapier, set your 'Trigger' as a new row in a 'QBR Request' Google Sheet or a CRM checkbox (e.g., 'Trigger QBR Draft'). • The 'Action' step should be the 'Send Prompt to OpenAI' module using the schema from Step 2. • Map the output to a presentation tool. Use the Google Slides API or a tool like 'Plus AI' or 'Gamma' to automatically populate slide text boxes. • Owner: Sales/CS Ops or RevOps Developer. • Time Estimate: 4-8 hours. • Common Pitfall: API rate limits. If you process 100 QBRs at once, the LLM might timeout. Build in 'Sleep' steps or use a queueing system. • Definition of Done: A 'Push-Button' workflow where clicking a button in the CRM generates a 5-8 slide draft directly in a Google Drive folder.

4. 04

   Human review and narrative layer

   The CSM's role transitions from 'Author' to 'Editor.' The AI provides the data-heavy slides, but the CSM must add the 'Narrative' and the 'Asks.' • The CSM opens the generated deck and looks for the 'Context' slide. • Action: Record a 2-minute Loom for the customer summarizing the AI's findings or add 2 personalized bullets regarding the customer's specific Q1 goals that weren't in the database. • Setting: In Google Slides, use 'Comments' to tag the Account Executive for feedback on the 'Expansion' slide. • Owner: Customer Success Manager. • Time Estimate: 30 minutes per deck. • Common Pitfall: Blindly trusting the AI's "Adoption" summary. AI cannot see if a low-usage account is actually a "seasonal" business. The CSM must verify the 'Why' behind the numbers. • Definition of Done: A completed deck with a human-verified executive summary and a clear 'Call to Action' for the customer.

5. 05

   Measure outcome and iterate schema

   Finally, close the loop by measuring the impact of these AI-assisted meetings on bottom-line results. • Tag each QBR in your CRM with a field: 'Drafting Method: AI-Generated' vs. 'Manual.' • Create a Salesforce report comparing the 'Time to Prep' (tracked via time-tracking or simple surveys) and the 'Renewal Rate' for AI-supported vs. manual accounts. • Use a SQL query like: `SELECT drafting_method, avg(retention_rate), count(qbr_id) FROM qbr_table GROUP BY 1` to see if the increased frequency of QBRs (due to less prep time) leads to higher retention. • Owner: CS Leadership / RevOps. • Time Estimate: 1 hour (monthly review). • Common Pitfall: Ignoring qualitative feedback. Ask CSMs if they feel "stuck" by the AI layout. If so, iterate on the prompt in Step 2. • Definition of Done: A dashboard showing a 50%+ reduction in QBR prep time and a benchmark for executive attendance rates.

Next playbooks

• CSM call summaries via Gainsight + AI (L2)
• Intercom Fin AI for support deflection (L3)
• Support deflection with AI search (L3)
• Customer health scoring with AI signals (L4)