THE PROBLEM
Most replies say the same few things
Picture a 12-person contractor's office that gets 60 emails and messages a day: new enquiries, questions about a job already booked, "did you get my payment?", supplier notices, and spam. Someone reads every one, works out what it is, looks the answer up in another system, and types a reply.
Most of those replies say the same few things. The work is not writing them; it is the sorting and the looking up, and that is the part software does well.
HOW IT SORTS
Four places a message can go
- New enquirySorted by service, added to the CRM, and a reply drafted with a booking link.
- About a jobMatched to the job, its status looked up, and an update drafted.
- BillingMatched to the invoice, and "yes, it's paid" answered from the payment record.
- Needs judgmentA complaint, a refund, or anything unclear goes to a person, with the history attached.
HOW IT FLOWS
From arriving to answered
The amber step is the person, and at the start every reply passes through it.
- ArrivesBy email, the website form or a text.
- ReadAI pulls out who it is, which job, and what they want.
- SortedOne of four kinds, with how sure it is.
- Looked upThe job, the invoice or the customer, from the business's own systems.
- DraftedA reply in the business's own wording, from its own facts only.
- ApprovedA person approves it with one tap, or edits it first.
- Sent and loggedThe reply goes out and the record shows what was said.
- TightenedWhat people change in a draft becomes a rule for the next one.
Runs on its ownA person decides
THE RULES
The rules that come with it
- It answers only from the business's own records and wording. A fact it cannot find goes to a person.
- Nothing about price, refunds or complaints goes out without a person.
- Every message and reply is logged, so anyone can see what it did and why.
- It starts with every reply approved by a person, and sends routine ones on its own only once the business is satisfied.
It is the same pattern already running in our content engine, where nothing the AI writes goes out without a yes.
WHERE ELSE IT FITS
Where else it fits
Maintenance requests sorted by property and urgency, with the tenant told what happens next.
Job updates and scheduling questions answered from the job itself.
Order status and quote requests matched to the order or the quote.
What it would use
- An AI model (Claude)
- Your email
- Your website form
- Your CRM or job system
- Your invoicing
- A log of every reply