01 the business
Heating and cooling installer
35 staff, four sales estimators
The estimators spend their days in customers' homes, and the office coordinator also runs scheduling. Enquiries arrive from the website, a quote-comparison portal and calls nobody could pick up. The owner suspects good jobs are being lost before anyone even speaks to the customer.
02 before
The problem
Web forms, missed calls and portal enquiries land while the team is on site or with other clients. Replies go out hours or days later, by which point the buyer has called someone else. When someone does call back, they often spend the conversation on people who were never going to buy.
- Weekend and evening enquiries wait until Monday, by which time many people have booked a competitor.
- Estimators drive to site visits for jobs that were never going to fit the budget or the service area.
- Lead details live in an inbox and on sticky notes, so the CRM shows only half the pipeline.
03 the build
What the AI does
- Step 1: Picks up each new enquiry from your website form, missed-call log or listing portal the moment it arrives.
- Step 2: Replies by SMS or email in your tone, thanks the person by name and asks the questions you use to qualify, such as budget, timing and location.
- Step 3: Answers simple questions about services, areas covered and availability from the notes you have approved.
- Step 4: Scores the lead against your rules and writes the answers and score into HubSpot or Pipedrive.
- Step 5: Offers a slot from your Calendly for good-fit leads and hands warm but unclear ones to a salesperson with a short summary.
paused: waiting for a person to approve
How it was put together
- Connected the website form, the portal's email alerts and the phone system's missed-call log to one intake queue.
- Wrote the qualifying questions with the owner: property type, system age, budget band, postcode and urgency.
- Loaded an approved notes file with services, areas covered and typical lead times, so answers come only from what the owner signed off.
- Scored each lead with the owner's own rules and wrote every answer into the CRM with the conversation attached.
04 human in the loop
Where a person stays in charge
You approve the qualifying questions, the reply templates and the scoring rules before launch. Any lead that asks about price exceptions, complaints or anything outside the approved notes is passed straight to a named person instead of answered.
- The assistant says it is automated in its first message.
- Questions about price exceptions, complaints or anything outside the approved notes go straight to a named person.
- It never commits to a price or a date. It offers visit slots from the estimators' real calendars.
05 rollout
From kickoff to live in 4–5 weeks
- Week 1Map lead sources, agree qualifying questions and scoring rules, write the approved notes.
- Weeks 2–3Build and test on the last three months of real enquiries, replayed with names removed.
- Weeks 4–5Go live on evenings and weekends first, review every conversation daily, then extend to all hours.
06 after
What changes
Enquiries get an answer while the buyer is still interested, even at night and on weekends. Salespeople open their day to a list of booked calls and qualified leads with the context already written down, instead of a pile of unread forms.
Assume 300 enquiries a month, 40% arriving outside office hours, and 10 minutes of coordinator time to read, reply to and log each one.
300 × 10 minutes = 3,000 minutes, or 50 hours a month of first-response admin. 300 × 40% = 120 enquiries that currently wait overnight or over a weekend.
What to measure
- Median time from enquiry to first reply.
- Share of enquiries that get a qualified or not-a-fit decision within a day.
- Booked site visits per 100 enquiries.
- Site visits that end in a quote, as a check on lead quality.
07 evidence
What the research says
A survey of 4,050 sales professionals found the average seller spends about 40% of their time actually selling, and 51% of sales leaders using AI say disconnected systems are slowing their AI work.
Salesforce, State of Sales report, 2026 edition (opens in a new tab) · 2026
Calls using AI-generated voices count as artificial voice calls under US telephone consumer law, so they need the person's prior express consent.
From August 2026, AI systems that talk directly to people in the EU must make clear they are AI, unless that is obvious.
EU AI Act, Article 50: transparency obligations (opens in a new tab) · 2024
08 lessons
Mistakes to avoid
- Letting the assistant improvise answers about price. Keep it to the approved notes.
- Asking too many questions up front. Three or four short ones get more replies than a long form.
- Forgetting the handover. A booked visit is only useful if the estimator sees the notes before arriving.
09 questions
Common questions
- Will customers know they are talking to AI?
- Yes. The first message says so, and anyone can ask for a person at any point. In the EU this disclosure becomes a legal requirement from August 2026.
- Can it call people back by phone?
- It can, but outbound AI voice calls need the person's prior consent in the US. Most businesses start with SMS and email replies instead.