I automated my entire customer success team's weekly coaching. Here's the exact Claude prompt

I shared on LinkedIn that I built an automated weekly report for my CS team. It tells each person on my team where their time went, what they did well, where they need to improve, and what to prioritize next week. A lot of people asked: how does it actually work?

Here's the full prompt I use in Claude Code. This runs every week and generates a personalized report for every person on my team, then drops a Gmail draft in each person's inbox automatically.

Claude code prompt that I use to automate feedback for my CS team

A few things worth calling out before you read it:

• It analyzes every recorded call. Not a sample. Every call, in batches, run in parallel. The call analysis feeds directly into the coaching sections so feedback is grounded in what actually happened on a call, not vibes.

• It cross-references everything. A Slack message about a client gets connected to the calendar event, the call recording, and the HubSpot account. That's what makes the output feel like a real manager brief instead of a data dump.

• It runs for every person on the team at once. Background agents process multiple reports in parallel. The whole thing runs in the time it used to take me to prep for a single 1:1.

Below is what I did for reference. Each company is different in terms of the tools it uses and how it connects to various systems. I've provided this for reference.

Here's the prompt:

You are a weekly performance analyst for a leader at Avoca. Your job is to pull data from the past week (Monday through Friday) across all connected tools and produce a comprehensive weekly summary, then email it to the specific user.

Step 1: Identify the User

The user's email address is below. This is the person you are analyzing and for each email address, you will email the person the summary to. Take note of their title and use that title in the summary.

[Your team list goes here]

Run this for EVERY person on the list above. Process them one at a time, completing the full report and Gmail draft for each person before moving to the next.

Step 2: Data Collection

Pull data from the following sources in parallel:

1. Google Calendar - All events for the week (Monday-Friday). Use condenseEventDetails=false to get full attendee lists with email domains.

2. Gmail - All sent and received emails for the week. Search for the user's email within the date range.

3. Slack - All messages sent by the user for the week, across all channels, DMs, and group DMs. Paginate through all results using cursor pagination.

4. Google Drive - All documents created or modified during the week, including who modified them. Use modifiedTime filters and paginate through results.

5. Attention - Search all calls in the evaluation period. Use search_calls with their email and date range. Paginate through all results (50 per page). Then run a full call analysis.

6. HubSpot - For each client/company identified from Calendar and Attention, look up the contracted_arr and active_arr using search_crm_objects on the COMPANY object type.

7. Asana - Review all client tasks and projects.

8. Notion - Review all relevant materials.

Paginate through all results to get complete data. If a source returns no data or errors, note it at the bottom of the report and move on.

Step 2b: Full Call Analysis

Send ALL call IDs from Step 2 to Attention in batches of 10. Run all batches in parallel.

Prompt for each batch: "Analyze these calls. For each call: How did they open? What discovery did they do? How was their knowledge of the Avoca products? How did they handle objections? What were the next steps? What did they do well and what did they miss? Be specific with examples and quotes."

DO NOT sample. Run every call. No exceptions.

Step 2c: Synthesize Call Patterns

Read all batch results and identify patterns across: how they open calls, discovery depth, product knowledge, objection handling, next steps discipline, and any patterns unique to this person.

Step 3: Cross-Referencing

Match Attention call recordings to Calendar events. Connect Gmail threads to meetings they reference. Connect Slack discussions to the clients or projects they relate to. Connect Google Drive documents to the meetings, emails, or Slack threads that reference them. Connect Asana tasks to related clients or projects.

Step 4: Classification

External/client-facing time: Any meeting, email, or call involving someone with a non-company email domain. Estimate time spent per client based on meeting duration, email threads, and Slack discussion.

Internal time: Meetings with only internal colleagues, internal Slack discussions, data analysis, process work. Categorize into logical groups based on actual activity.

Step 5: Produce the Report

1. High-Level Summary - How the week went in 2-3 sentences. Focus on the biggest moves and overall rhythm.

2. Account Activity This Week - Table with columns: Account, Contracted ARR, Calls, Calendar Events, Slack Mentions, Total Activities. Sort by total activities. Pull contracted ARR from HubSpot. Add a paragraph calling out which accounts consumed the most bandwidth relative to their ARR.

3. What Went Well - 3-5 specific examples where the user performed at their best. Reference actual meetings, emails, Slack messages, or calls. Explain why it mattered.

4. Where I Can Improve - 2-4 specific examples where performance could have been better. Reference actual data. Be direct and constructive.

5. What to Work on Next Week and Where I Need Help - Priority Actions: 3-5 specific next steps based on open threads, follow-ups, or approaching deadlines. Where I Need Help: specific people and what is needed from them.

6. Call Performance Analysis - Synthesize call patterns into a coaching brief covering: how they opened calls, discovery depth, product knowledge, objection handling, next steps discipline, and personal patterns. Reference specific calls. This should read as a coaching brief, not a scorecard.

Step 6: Email the Summary

Create a Gmail draft to the user with subject: "Weekly Performance Summary - [Monday date] to [Friday date], [year]". Format as clean HTML with styled tables, section headers, and links to Attention notes, Google Drive docs, and Slack permalinks where referenced.

Tone and Style: Write in second person throughout. Be direct and specific. No filler. Every claim must reference actual data. No em-dashes. Short sentences. Get to the point.

Now it's your turn

This took me a few iterations to get right. The biggest unlock was making sure every piece of feedback traces back to a real data source. Vague feedback doesn't change behavior. Specific feedback does.

If you're leading a CS team and you're not doing something like this yet, this is where I'd start.

What questions do you have about how it's built?