A cleaning company's website chat has two jobs: give a price people trust, and get them to book. Most chatbots fail one of two ways. They invent an answer (a discount, an insurance policy, a free slot today), or they collect everything and lose the client at the last step.
What I built
A customer-facing agent on Agentplace for Maple & Mop, the same demo company as my QuoteDesk app.
One question at a time: home type, rooms, the right kind of clean, frequency, extras.
An itemized estimate from fixed price rules: base $90, $25 per bedroom, $30 per bathroom, deep clean ×1.5, move-out ×1.8, frequency discount applied last. Every line is shown, not just a total.
Client notes ("two cats, fourth-floor walk-up") go into the quote recap for the crew.
Guardrails: it never confirms a booking or a time slot. When it doesn't know (insurance, same-day jobs), it says the team will confirm by email instead of guessing.
Handoff: the recap ends with "Get my exact quote", which opens the live quote app where the real booking happens.
How I tested it
The same scenarios before and after publishing: 2 bedrooms and 1 bath, standard ($170.00); deep clean every 2 weeks with inside oven ($246.50); a 3-bedroom move-out with inside fridge ($435.00); notes with pets and stairs; and questions it must not answer itself. One rendering error on the estimate card showed up in testing and was fixed before publishing.