Run a small SaaS feedback loop with Claude: a weekly PostHog survey digest, weekly release summaries and browser-based product testing
A course-creation software company asks new free users to rate their first result, has a scheduled Claude job turn the answers into a weekly issue list, has Claude summarise each week's releases for the team, and has Claude test the product in a browser like a new customer.
Evidence: The author reports this. We have not checked it beyond reading the source.
The business problem
A small team gets scattered, unfiltered feedback from new users, loses track of what shipped, and has little time for manual testing.
What was tried
Show a short survey to new free users after they create their first course, asking for a rating from one to five and a reason. Each week, a scheduled automation uses the PostHog MCP with Claude to read the responses, drop non-constructive ones, strip names and identifying data, and post a short list of the top product issues to the team chat. Each week, ask Claude to review the past week's product releases and post a summary for the whole team, replacing a manual process that depended on a few people and often missed updates. Separately, have Claude open a browser, sign up and build a course as a customer would, reviewing the experience as a user and a tester, and report serious bugs and usability issues.
What was reported (not reported)
The company says the weekly digest and release summaries replaced patchy manual routines, that output roughly doubled for non-technical staff with larger gains for technical staff, and estimates savings of several thousand hours a year, 'say 3000 plus', from all its AI use together. No counts of issues fixed or changes in ratings are given.
Limitations
A company feature in an interview series, with no data shown. The 3,000-hour figure is the co-founder's informal estimate across all AI use, not a measurement of these tasks. The survey filter discards non-constructive responses, so some signal could be lost, and the article gives no error rates, bug counts, or detail on who checks the AI summaries and QA reports against the real product. The interviewee is describing his own company, and prefers some tools for personal reasons.
What you need
PostHog with surveys enabled and its MCP server, Claude with scheduled tasks and a browser-capable setup, access to your release notes, and a team chat to post to.
Sources
- AI Center on Substack (interview with Alex Hey, co-founder of Coursebox AI) ↗ Firsthand write-up, published June 30, 2026
Source published: June 30, 2026. Last reviewed here: October 14, 2026. Spot a mistake? Tell us.
Tools in this workflow
- Claude: Filters, anonymises and summarises the responses each weekTry Claude
- PostHog: Runs the new-user survey and exposes responses through its MCP serverTry PostHog
Related workflows
Monitor your brand and competitors on X by sending mentions to Grok for sorting, then review a daily digest
A managed-IT firm describes pulling mentions of a brand and competitors from the X API, having Grok sort each one into complaint, praise, comparison or noise with a rough sentiment score, and posting a digest for a person to review.
Draft a training course outline with Claude Projects: your own voice documents, 16 design questions, then edit every step
A learning and development professional loads her voice, expertise and story documents into a Claude Project, gives it her outline structure and about 16 design questions, and edits the draft outline and workbook text it returns.
Use an AI reviewer to check a draft against your written voice rules instead of having AI write it
An author with forty years in influence and sales training writes each section himself, locks it, and then runs the draft past an AI reviewer whose only job is to flag where it drifts from his voice.
Turn a blog post into a Pinterest pin, alt text and LinkedIn post in about four minutes with a free Claude skill
A creator replaced an eight-step manual routine of separate prompts with a single Claude skill that produces the pin title, description, alt text and a LinkedIn post from a blog title and full text.