The problem
He writes that e-commerce brands have run abandoned-cart flows for years. Service businesses have booking flows that “fail every day,” but “Nobody’s running recovery on those carts.”
He gives three reasons:
- “The software vendors haven’t built it.” The salon-scheduling platforms focus on calendar management and stylist-side operations.
- “The build was too expensive before AI.”
- “Most owners don’t know the cart exists in the first place.”
He thinks salon carts are worth chasing because most people who abandon them “got pulled away” rather than lost interest.
There was also a problem at home. His wife Crystal runs the operation, and he writes: “My wife’s honest read, for months, was that we were just bugging people.”
The blueprint (as he describes it)
- Rebuild the booking flow you control. He started rebuilding the site in Lovable in early 2025. When he hit the platform’s ceiling, he moved the code to GitHub, kept Supabase as the backend, deployed on Vercel and installed the Claude CLI. That move happened at the end of June 2025. Between late September 2025 and the end of January 2026 he rebuilt the public website and booking flow from scratch with Claude. He sums up that rebuild as “Four months. One person.”
- Capture the partial booking. When a prospective customer picks a service, a stylist, or a date or time, “the system captures that partial state.”
- Wait, then enroll. If they don’t finish “within a reasonable window, they join a recovery flow.” The length of the window is not stated.
- Send a specific email first. He says the email is not a generic “come back” message. It is “A real reminder of the exact service they were considering, the stylist they chose, and a nudge on timing.”
- Follow with a text, then an incentive. “A text follows after a gap. An incentive appears on a later touch if the previous ones didn’t work.” The gap and the type of incentive are not stated.
- Write it like a person. He says the copy “reads like a real human remembering you, not a marketing automation blast.” He does not say who writes the copy or how it is generated.
- Keep staff out of it. “Our stylists don’t chase anyone. Our front desk doesn’t do outreach.” Staff see only the resulting booking.
- Let the data decide. He showed Crystal the dashboard in May. He writes: “If the numbers had come back the other way, the system would be gone”.
Stated results
- Sixteen percent of abandoned bookings came back through the email flow.
- 17.3 percent came back once texts were layered in.
- Tracked recovered bookings crossed ten thousand dollars.
- These figures come from the dashboard he showed his wife in May, published July 27, 2026.
- In the June 29, 2026 post he wrote, “I won’t share specific numbers here.”
- In the April post he says overall revenue conversion through the booking flow “has lifted measurably since the new website went live.” No figure is given, and that metric covers the whole new site, not only recovery.
- The number of abandoned bookings and the time window behind the percentages are not stated.
Caveats
- All results are self-reported on his own Medium. He runs the salon, and no product, course or kit is sold in these posts. The source text was captured through a Freedium mirror of the Medium posts.
- The posts do not say the recovery emails or texts are written by AI, or how the copy is generated. This is an automated workflow he built with Claude. Claude is the build tool.
- The figures are a single snapshot. There is no breakdown by location or time window and no count of abandoned bookings.
- “Ten thousand dollars” is tracked recovered bookings, not profit. How a booking gets attributed to the flow is not stated.
- The “5–15% range” for e-commerce recovery and the “30% of visitors who bail halfway through” are his general claims, not his salon’s data.
- The timing is loose. The July and June posts say he built the system the year before, while the April post lists it among things built after the new site launched.
- The custom CRM and the “AI-drafted SMS responses that handle incoming customer texts using our real client history as context” are separate systems from the recovery flow.
- Handles:
- LinkedIn /in/mike-frehner-80117544 (verified 2026-11-08: linked from his Medium About page; posts about Hottie Hair).
- Medium @mikefrehner (verified 2026-11-08: author).
- Company accounts linked from hottiehair.com (verified 2026-11-08): Instagram @hottieextensions, TikTok @hottieextensions, YouTube @hottiehair, Facebook hottiehairsalon, X @hottiehair.
- No personal X, Bluesky, Threads or Mastodon found.
FAQ
Does AI write the recovery messages?
The source doesn’t say. He built the flow with Claude as part of a website rebuild, but the posts never explain how the email or text copy is generated.
What tools is it built on?
He rebuilt the site and booking flow with Claude, with code on GitHub, a Supabase backend, Vercel hosting and the Claude CLI, after starting in Lovable. The email and SMS providers are not named.
How long does the system wait before following up?
The source says only that people join the flow if they don’t finish “within a reasonable window” and that a text follows “after a gap.” No specific timings are given.
Is the ten thousand dollars profit?
No. It is tracked recovered bookings, and the source does not state profit or how recoveries are attributed.
Sources
- https://medium.com/@mikefrehner/my-wife-thought-the-ai-i-built-was-just-bugging-people-then-i-showed-her-the-numbers-0cd6e2b8e82d (primary: first-person results post, July 27, 2026)
- https://medium.com/@mikefrehner/we-built-abandoned-cart-recovery-for-a-salon-most-others-still-havent-60d08cb8ecc7 (companion: mechanics, June 29, 2026)
- https://medium.com/@mikefrehner/im-not-a-consultant-i-m-a-salon-owner-who-writes-his-own-software-59f2e755e8ba (companion: background and stack, April 23, 2026)
Want agents like these working your distribution? Get It Built. Get new teardowns weekly: subscribe (double opt-in).
Get one teardown a week.
Same format, a new operator every issue. Double opt-in.