Subscription cancellation-reason analysis and save-offer drafts
Group cancellation reasons into themes each week and draft tailored save-offer messages for a person to approve before any are sent.
The use case
Subscription brands collect cancellation reasons through surveys and support tickets, but the free-text answers pile up unread. Owners suspect patterns, such as too much product, shipping problems, or flavor fatigue, without clear evidence. This agent reads the week's cancellation reasons, groups them into themes with example quotes, and notes which themes are fixable with an offer like skipping a delivery or changing frequency. For recent cancellations where policy allows outreach, it drafts a short, respectful save message matched to the stated reason. A retention owner approves the offers and messages, and nothing goes to customers without that review.
Example workflow
- A weekly job exports cancellation survey answers and related support tickets from the subscription app and helpdesk.
- The agent removes payment details and other sensitive fields, then groups reasons into themes with representative anonymized quotes.
- It writes a short summary of themes, counts per theme, and product or process issues worth investigating.
- For eligible recent cancellations, it drafts a save message matched to the reason using only offers on your approved list.
- A retention owner reviews the summary and drafts, removes anyone who asked not to be contacted, and approves messages to send.
- Approved messages are sent through the email or helpdesk tool and outcomes are logged against each theme.
Tools
- Subscription app export
- Klaviyo or Gmail
- Claude
- Google Sheets
- Notion
- Zendesk
Risks and guardrails
- Contacting customers who opted out could breach consent rules — guardrail: the agent filters out opted-out contacts and the reviewer confirms the list before sending.
- The agent might invent offers you do not run — guardrail: drafts may only use offers from an approved list maintained by the retention owner.
- Theme counts on small samples can mislead decisions — guardrail: the summary shows raw counts and example quotes rather than percentages.
- Pushy save messages can harm the brand — guardrail: templates make it easy to confirm the cancellation and drafts are reviewed for tone.
Related recipe
Support Ticket and Review Triage with Policy-Cited Draft Replies: An agent tags every helpdesk ticket and public review by intent, urgency, and sentiment, routes it, and drafts a reply that cites your help center. People approve refunds and all public replies.
Want it set up for you? Request a quote.
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- Klaviyo MCP server (official page): developers.klaviyo.com/en/docs/klaviyo_mcp_server (checked 2026-11-09)
- Gmail (official page): developers.google.com/workspace/gmail (checked 2026-11-19)
- Claude (official page): claude.ai (checked 2026-11-09)
- Notion MCP server (official page): developers.notion.com/guides/mcp/overview (checked 2026-11-09)
- Zendesk AI (official page): www.zendesk.com/service/ai (checked 2026-11-09)