The Problem
Anna Moore runs a yoga studio across the street from our office. A yoga instructor, marathon runner, and former military officer. She did not start Sweat so she could do paperwork at 10 PM.
By the time we sat down, she was. Cancellation emails came in at all hours. Membership changes piled up every Monday. Instagram DMs ate her evenings on the couch. Stillwater is a college town, so the admin load is not steady. It comes in waves.
Anna was not thinking about growth. She was thinking about getting out.
I kind of feel like we were just programmed to feel like we always should be stressed out.
Why this was hard
Anna's booking platform is WellnessLiving. It runs her schedule, her memberships, her billing. It is good at what it is built for. It is not built for what she needed.
- No API for membership actions. WellnessLiving exposes client lookups and purchase history. It does not expose endpoints for membership cancellation or freeze. A developer cannot build a clean integration against functions that are not published.
- An API agreement priced for enterprises, not studios. $19 a month for access, plus a $150 per hour change fee with no scope cap. For a small business, that is a non-starter.
- Zapier and n8n cannot help. They connect APIs that exist. They cannot navigate a web UI and perform actions the way a person does.
This is exactly the kind of problem AI-driven browser automation is built for. An agent that logs in, reads context, fills in forms, validates against business rules, and flags anything unusual for a human. That is what we built.
What we built
Membership cancellation and freeze
Anna forwards a client email or a Modify Membership form submission to the agent. The agent reads context, picks the right PDF (cancellation with a 30 day notice, or freeze with a 30 to 90 day window), and sends it to the client.
When the completed PDF comes back, the agent validates against Sweat's rules:
- Cancellations require at least 3 months of payment history
- Freezes must fall between 30 and 90 days
- Start dates in the past default to today
- End dates past 90 days get capped at the max
If everything checks out, the agent updates WellnessLiving and sends a confirmation email. If anything is off, say a member only has 2 months paid or dates fall outside the window, the agent stops and flags both Anna and Logan for review.
Anna is always in the driver's seat. The agent does not do anything consequential without her eyes on it first.
The agent doing what a human used to do, in real time.
If I were to sit and do the cancellations myself, it would eat up an entire day. Anna Moore
Instagram DM responder
About 95% of Sweat's Instagram DMs are the same five questions. Pricing. Schedule. How to book. What to bring the first time. Late-cancel sob stories.
The agent reads the DM, classifies the intent, answers in Sweat's voice and Sweat's policies, and pushes the app download when it makes sense. Anything it cannot handle routes to a human inbox with context attached.
Built on ManyChat Pro with the AI Add-on. Meta-approved, running on the official Instagram Graph API. No password sharing, no account suspension risk.
During the supervised rollout, Anna reviews suggested responses before they go out. Once trust is built, the agent sends directly. That trust is earned, not assumed.
Monthly employee scheduling
Anna's team is 5 to 10 interchangeable hourly employees, many of them OSU students with shifting availability. Building next month's schedule is tedious, constraint-heavy, and the kind of thing humans get wrong and AI gets right.
The agent runs once a month, 7 days before the 1st of the following month. It pulls availability from When I Work, respects a hard 29 hour weekly cap per employee, and distributes shifts fairly based on the prior month's actual hours. Anything it cannot fully cover, it flags.
The agent drafts shifts as unpublished in When I Work. Anna reviews and publishes. A coverage gap email listing unfilled shifts is prepped as a draft for Anna to approve before it sends.
The AI enjoys this kind of problem. Anna did not.
Also shipped along the way
Infrastructure that quietly supports the automations:
- WellnessLiving duplicate profile cleanup so every downstream automation operates on clean data.
- Website updates to support the new workflows.
- API research across WellnessLiving and When I Work, so every decision on what to automate was informed by what was actually possible.
Under one month of the cancellation and freeze automation being live. Before finals week, which is the annual peak.
Sweat runs on a college-town calendar. Membership changes come in waves, not a steady drip. The cancellation wave hits at finals. Students leave. Students graduate. New members arrive. The automation does not just save Anna a few minutes here and there. It absorbs the biggest spikes in her admin load at exactly the moments she is most likely to be buried in them.
It feels weird that I don't have to be monitoring my email constantly.
What changed for Anna
Anna now spends her evenings off her phone and with her daughter. She is refocusing Sweat on being an active presence in the Stillwater community and a rejuvenating place for her clients. That is the work she started the business to do.
The automation is not the win. The win is that Anna gets to go back to being who she is.
You can hire less people, but you just have peace of mind that things are getting done.
Shimmer Labs does not sell you a tool and walk away. Anna knows her business. We know how to build sidecars. Together, we figured out what to automate first, what to automate next, and what should always stay in her hands. You drive. We build the sidecar.
Hear it from Anna
Tech Stack
- Perplexity Computer
- ManyChat Pro + AI
- Claude Desktop
- Claude in Chrome
- Gmail
- WellnessLiving
- When I Work
- WordPress
Let's build your business a sidecar.
Curious what it costs? Learn about the service →