RelayOps AI Rebooking Agent

RelayOps is a Slack-native AI agent for rebooking recovery with grounded insights and human-approved outreach.

Comment

Inspiration

Small businesses spend real money earning customer trust, but many lose repeat revenue simply because no one follows up after the last appointment. Salons, facial boutiques, dental clinics, massage clinics, physio clinics, cleaning companies, and contractors often rely on manual calls, memory, or scattered CRM notes.

RelayOps AI Rebooking Agent was inspired by a simple question: what if every small business had an AI employee in Slack that checked the books every morning and told the team exactly who to contact, why they matter, and what to say?

What it does

RelayOps scans mock CRM and appointment data, identifies customers who are overdue for rebooking, scores each opportunity, and generates a Slack-ready daily revenue recovery report.

For each customer, RelayOps can:

Explain why the customer was selected Show last visit date and typical return cycle Calculate a priority score Estimate recoverable revenue Recommend the best communication channel Draft personalized SMS, email, or phone outreach Let staff mark the customer as contacted from Slack, which suppresses them from future reports for a 14-day cooldown so no one gets double-contacted Staff can ask natural-language questions such as:

Which customers should we contact today? Who has not returned in 90 days? Show overdue VIP customers Draft a follow-up message Summarize today's opportunities Who have we already contacted?

How we built it

We built RelayOps as a Slack-first AI operations agent with a TypeScript and Node.js backend.

The MVP includes:

A Slack app using Bolt for JavaScript (v4) over Socket Mode A 100-customer mock dataset with appointment history, spend, service types, and last visit dates SQLite storage with a clear path to PostgreSQL A scoring engine that ranks customers by overdue gap, spend, loyalty, VIP status, and consent OpenAI GPT tool calling for grounded AI responses Deterministic fallback responses so the demo works even without an OpenAI API key Slack Block Kit reports with action buttons and an App Home dashboard A 14-day contact cooldown that closes the loop so contacted customers drop out of tomorrow's report Daily scan scheduling for morning reports An MCP server exposing the same rebooking intelligence to any AI client Documentation, demo script, pitch deck outline, and video script For the Slack Builder Challenge, RelayOps satisfies the required technology by also running as an MCP (Model Context Protocol) server: the exact same service layer that powers Slack is exposed as five stdio tools, callable from Claude Desktop or any MCP client. One domain core, two front doors — Slack and MCP. The core workflow is a repeatable business process: scan records, score opportunities, generate recommendations, route actions to staff, log outcomes, and repeat.

Challenges we ran into

The main challenge was avoiding a generic chatbot. We wanted the agent to behave like a practical revenue recovery employee, not just answer vague questions.

That meant grounding every AI response in structured customer data, exposing clear business logic, and making the Slack experience actionable. The agent needed to say not only "contact Sarah," but also why Sarah was selected, how much revenue may be recovered, what channel to use, and what message to send.

Another challenge was keeping the architecture simple enough for small businesses while still being production-minded. We started with SQLite and mock data, but designed the service layer so it can later connect to booking platforms, CRMs, and PostgreSQL — and so the same layer could be surfaced through a second interface like MCP without duplicating logic.

Accomplishments that we're proud of

We are proud that RelayOps feels like a real startup product, not just a hackathon prototype.

The MVP already demonstrates a full workflow:

Generate realistic customer and appointment data Run a daily rebooking scan Score overdue customers Estimate recoverable revenue Ask natural-language questions Draft personalized outreach Log follow-up actions from Slack and suppress contacted customers from the next report Expose the same intelligence to any AI client through an MCP server The local demo produced 81 overdue customers, 32 high-priority opportunities, and an estimated $7,235 in recoverable revenue from the mock dataset.

What we learned

We learned that the highest-value AI use cases for small businesses are often not flashy. They are practical workflows that help teams make better decisions every day.

We also learned that AI agents are much more useful when they are connected to structured tools. RelayOps uses GPT for reasoning and language, but customer facts come from function calls against the CRM and booking data. That keeps the agent grounded, auditable, and business-focused — and once that domain core exists, exposing it through a second surface like MCP is nearly free.

What's next

Next, we would connect RelayOps to real booking and CRM platforms such as Square, Fresha, Mindbody, Jane, Jobber, ServiceTitan, and dental practice systems.

We would also add:

UiPath Maestro BPMN orchestration for governed process flow Human approval tasks for sensitive outreach Multi-location reporting PostgreSQL multi-tenant storage Consent and opt-out enforcement RAG over customer notes, campaigns, and staff playbooks Analytics showing booked revenue recovered by RelayOps The long-term vision is for RelayOps to become the AI operations employee for small businesses, starting with rebooking recovery and expanding into retention, reminders, win-back campaigns, and revenue operations.

Built With

Share this project:

Updates