Skip to main content

Updated Sep 30, 2026Verified against the product Sep 30, 2026

Open in ClaudeOpen in ChatGPT

Turn Leadership Reporting into Agents

Every week someone in Ops or CS leadership assembles the same report: which renewals are at risk, what changed, which accounts need an executive's attention. It takes hours, it's stale by the time it's read, and it waits whenever that person is out. A scheduled agent can do it instead — query FunnelStory at the same time every week, write the summary, and post it where leadership reads it. This guide shows how to choose what to automate, build the first report, and measure the time it gives back.

Step 1: Inventory the recurring reports​

List every report your team produces on a schedule, with who reads it and how long it takes:

ReportAudienceCadenceHours per runSources today
Weekly renewal risk roll-upVP of CS, CROWeekly3CRM export, spreadsheet, Slack threads

Start with the report that costs the most hours and has the most stable format. A report whose structure changes every week needs a person; one that answers the same questions every week is ready for an agent.

Step 2: Write down the questions it answers​

A good leadership report answers a handful of fixed questions. For a renewal risk roll-up, for example:

  1. Which renewals in the next 90 days are predicted to churn, and how much ARR is that?
  2. What new risks appeared this week — pricing, competitors, champions leaving?
  3. Which large accounts are single-threaded?
  4. What changed since last week?

Each question becomes a query against FunnelStory's semantic database, which holds everything the report needs in one place:

QuestionWhere the answer lives
Renewals, ARR, predictions, healthaccounts (expires_at, amount, prediction, health_score)
New risks and opportunitiesneedle_movers
Relationship coveragerelationship_map_history
Account context for the narrativeaccount_profiles
Ownersaccounts.assignees, joined to workspace_users

Step 3: Build the agent​

With the MCP server connected, describe the report to Claude and let it build the agent:

"Read the FunnelStory flow guide, then build a draft agent that runs every Monday at 8am Pacific. Answer these questions: [your list from Step 2]. Summarize the answers in under 200 words with the three accounts that most need an executive's attention, and post it to Slack channel #cs-leadership. Test it before I publish it."

The agent Claude builds typically has:

  • a Schedule trigger with a cron expression and timezone;
  • one semantic.query step per question, so each number is computed, not guessed;
  • an AGENT step that writes the summary from those results and cites the accounts it names;
  • a slack.send_message or email.send step to deliver it.

Keep the numbers in query steps and the prose in the AGENT step. Leaders will compare this week's figure with last week's, so the figure must come from the same query every time.

Prefer the canvas? The same agent can be built step by step — see Getting started and the agent examples, which include a scheduled health alert to Slack.

Step 4: Run it next to the manual report​

For the first two or three weeks, run the agent alongside the report you're replacing. Compare the numbers line by line. Differences usually come from definitions — which accounts count as "renewing", which ARR field is used — so fix the query, not the summary. When the two match, publish the agent and retire the manual version.

Going further​

  • Board or QBR decks. Agents can build a PowerPoint from your own template with the pptx.create function. Ask Claude to read the file://flow/pptx.md resource over MCP before building one.
  • A live view instead of a message. For numbers leadership wants to explore rather than read, a custom dashboard or a Data App may fit better than a weekly post.
  • Exception alerts. Some questions don't need a weekly cadence. "Tell me when a top-20 account is predicted to churn" is better as an agent with a Signal trigger that fires the moment it happens.

Measure the time it gives back​

Track two numbers for each report you automate: the hours of manual work removed per month, and whether the report now arrives on time every time. Add them up across reports to show what automation is worth to your team, and use the same inventory from Step 1 to pick the next one.

Last updated on