Running a Churn Study From Your AI Agent: Brief to Insights in 48 Hours

Traditional churn studies take 4-8 weeks. Brief the vendor. Wait for recruitment. Schedule 20 moderated calls. Get transcripts. Analyze themes. Receive the report. By the time you understand why customers left last quarter, they have already left this quarter.

Agent-driven churn studies on your own customer list take 24-48 hours. The User Intuition agentic research platform gives your AI agent the tools to design the study, recruit from your CRM, run AI-moderated interviews, analyze transcripts, and retrieve a final report — all without a human in the loop.

This guide walks through the full workflow with the actual tool calls.

The Workflow at a Glance

Six steps, end to end:

1. Load the planning prompt — built-in guidance for structuring a churn research brief
2. create_assistant — configure the AI-moderated interview with discussion guide and study parameters
3. bulk_create_invites_from_segment — recruit from your churned-customer segment in Shopify or HubSpot
4. list_calls — monitor interview completion in real time
5. analyze_transcripts — run structured theme extraction across all completed calls
6. get_assistant_report — retrieve the synthesized findings document

Then, ongoing: query_intelligence — query across this study and all past studies to surface patterns.

Step-by-Step: The Actual Tool Calls

Step 1: Load the Planning Prompt

User Intuition’s MCP server includes built-in prompts for common research workflows. The churn research prompt gives your agent a structured brief template covering: study objectives, target segment definition, key hypotheses, discussion guide outline, and output requirements. Ask your agent: “Load the churn research planning prompt.”

This takes about 5 minutes. You are drafting the brief, not writing a discussion guide from scratch.

Step 2: Configure the Interview with create_assistant

The agent creates the interview assistant based on your brief:

create_assistant({
  name: "Q2 2026 Churn Study",
  description: "Understand the primary drivers of churn among customers who cancelled in Q1 2026",
  study_type: "churn_analysis",
  discussion_guide: "...",
  target_sample_size: 25,
  interview_language: "en",
  interview_mode: "audio"
})

The response includes an assistant_id and invite_link you will use in the next step. The assistant is now configured: discussion guide loaded, moderation logic set, analysis pipeline ready.

Step 3: Recruit With bulk_create_invites_from_segment

This is where BYO customers matter. Instead of recruiting from the general panel, recruit the people who actually churned:

bulk_create_invites_from_segment({
  assistant_id: "asst_abc123",
  segment_id: "shopify_churned_q1_2026",
  invite_type: "email",
  max_invites: 50
})

The tool generates individual invites for every customer in the segment and queues them for delivery. No manual CSV. No third-party panel. The people you are studying are your actual customers — which means the data is directly attributable to your specific product, pricing, and experience.

If you are using HubSpot, the same call works with a HubSpot list ID. You can also pass a BYO email list directly for companies not yet connected to a supported integration.

Step 4: Monitor With list_calls

Interviews complete asynchronously over 24-48 hours as customers respond to invites. Monitor progress:

list_calls({
  assistant_id: "asst_abc123",
  status: "completed"
})

The response shows completed call count, average duration, and engagement flags. Aim for 20 completions before running analysis — 25 is better if you have the response rate.

Step 5: Analyze With analyze_transcripts

Once you have sufficient completions, trigger structured analysis:

analyze_transcripts({
  assistant_id: "asst_abc123"
})

This runs the full NLP pipeline across all completed calls. Returns:

{
  "themes": [
    {
      "theme": "Pricing vs. perceived value gap",
      "prevalence": 0.64,
      "sentiment": "negative",
      "summary": "Most churned customers described the price as fair in absolute terms but felt the specific features they used most did not justify the plan cost.",
      "evidence": [
        { "quote": "I wasn't using the video interviews at all — only the chat ones. But I was paying for the tier that bundled both." }
      ]
    },
    {
      "theme": "Onboarding friction for the Intelligence Hub",
      "prevalence": 0.40,
      "summary": "Customers who churned early frequently mentioned not fully activating the Intelligence Hub before deciding to cancel.",
      "evidence": [
        { "quote": "I ran a couple of studies but I never figured out the search and query piece. I was still copying findings into a doc manually." }
      ]
    }
  ],
  "minority_objections": [
    {
      "theme": "Alternative found with lower minimums",
      "prevalence": 0.16,
      "summary": "A subset of churned customers moved to a competitor specifically because of lower per-study minimums for small sample sizes."
    }
  ]
}

Step 6: Retrieve the Full Report With get_assistant_report

The analysis JSON is useful for the agent to parse and reason about. The report is useful for sharing with humans:

get_assistant_report({
  assistant_id: "asst_abc123"
})

Returns a structured document with executive summary, ranked themes with evidence, minority objections, strategic recommendations, and appendix of verbatim quotes. Format is structured text that can be rendered as markdown, passed to generate_powerpoint for a deck, or read directly in your agent interface.

What Do You Get Out the Other End?

By the end of the 48-hour window, you have:

• Theme analysis — ranked by prevalence, each with supporting verbatim evidence. Not a word cloud. Actual structured insight.
• Minority objections — the 15-20% of churn signals that don’t fit the main themes but may be early indicators of a larger trend.
• Citable quotes — real participant language, traceable to individual conversations. Useful for product discussions, board slides, and prioritization arguments.
• Recommendations — generated from the full pattern of 20-25 conversations, not a single outlier.

The output is structured data the agent can act on. It can summarize findings in a Slack message, create a Linear issue from the top theme, or ask a follow-up question using query_intelligence.

Following Up Via query_intelligence

The churn study does not expire. It feeds the Intelligence Hub — the same knowledge base that stores every other study you have ever run. An agent calling query_intelligence three months from now can ask: “What churn drivers have appeared consistently across our quarterly churn studies?” and get cross-study synthesized findings.

query_intelligence({
  query: "What are the most common churn drivers across all completed studies?",
  session_id: "research_session_abc"
})

The response pulls from every completed study in your account — not just this one. The Q3 churn study is contextualized against Q1 and Q2. Pattern recognition surfaces across waves. A 2% increase in “pricing vs. value” mentions looks like noise in one study. The same increase across three consecutive quarters is a trend line.

This is the compounding advantage of running research on a platform with institutional memory rather than one-off interview tools.

Why Does “On Your Own Customers” Matter?

Panel-based churn research answers the question: “Why do people in this demographic leave products like yours?” That is useful directional research. But it is not the same as understanding why your customers — the ones who actually used your specific product, at your specific price point, in your specific competitive context — decided to cancel.

bulk_create_invites_from_segment closes that gap. The participants are the people who cancelled. The discussion guide is calibrated to your product. The findings are directly attributable to your specific experience, not a panel proxy.

For attributable, actionable churn research, the signal from 20 of your own customers consistently outperforms 100 panel proxies.

How Does User Intuition Handle Agent-Driven Churn Studies?

User Intuition’s agentic research platform was designed for this exact use case: programmatic, end-to-end qualitative research that requires no human in the loop between brief and findings. Three capabilities make it particularly well-suited for churn work.

First, CRM integration via bulk_create_invites_from_segment means your churned customers are the participants — not panel proxies. Shopify and HubSpot integrations are available on all paid plans. The invite management tools (create_invite, mark_invite_sent, send_reward) handle the full participant lifecycle programmatically. At $20 per audio interview on the Pro plan with 98% participant satisfaction, a 25-person churn study costs $500 and delivers genuine, multi-layer conversations — not checkbox completions.

Second, the analysis pipeline is built for depth. analyze_transcripts runs on AI-moderated conversations that average 5-7 levels of laddering depth and 30+ minutes per participant. The themes it extracts reflect genuine explanatory chains — why people feel the way they do — not just surface-level topic tags from keyword analysis. This matters for churn research because the presenting reason (pricing) often obscures the actual driver (onboarding friction that prevented value realization).

Third, the Intelligence Hub makes each quarterly churn wave smarter. Studies from 24 months of quarterly churn waves, all queryable via query_intelligence, give your agent the ability to contextualize new findings against institutional memory that would otherwise live in a series of disconnected research decks. The platform stores 98% satisfaction-rated conversations from a 4M+ panel across 50+ languages, plus your own BYO customer interviews — and surfaces pattern-level insight across all of them on demand.

Connect your Shopify or HubSpot customer list and run your first study this week at app.userintuition.ai/sign-up.