01 the business
Plumbing and drainage company
28 staff, 24/7 emergency service
Calls come in at all hours, and the on-call plumber cannot answer while working. Many callers hang up at voicemail and ring the next company. The messages that are left are hard to follow, and the first hour of every morning goes on call-backs.
02 before
The problem
Calls that come in after hours, at lunch or while everyone is on a job go to voicemail. Many callers hang up and ring the next business on the list. The messages that are left are often unclear, and someone loses the first hour of the morning calling people back.
- After-hours calls go to voicemail.
- Real emergencies are mixed in with routine requests.
- Morning call-backs delay the day's first jobs.
03 the build
What the AI does
- Step 1: Answers the phone in a natural voice through Twilio when your team is busy or closed, and says clearly that it is an automated assistant.
- Step 2: Answers common questions about services, opening hours, service areas and prices you have approved.
- Step 3: Takes the caller's name, number, address and reason for calling, and checks the details back with them.
- Step 4: Books a slot straight into ServiceTitan, Jobber or your calendar when the request fits your booking rules.
- Step 5: Texts or calls the on-call person for genuine emergencies, using the escalation rules you set.
- Step 6: Logs a written summary and recording of every call in your CRM so the morning team can see what happened.
paused: waiting for a person to approve
How it was put together
- Answered calls through Twilio with an approved script, stating it is an automated assistant.
- Took name, number, address and the problem, and read them back to confirm.
- Booked routine jobs into the scheduling system within the owner's rules.
- Escalated emergencies such as active leaks to the on-call plumber by call and text.
04 human in the loop
Where a person stays in charge
You approve the script, prices and booking rules before go-live. Anything outside them, such as a quote for unusual work or a clinical question, is taken as a message for a person to call back.
- The assistant identifies itself as automated at the start of every call.
- Quotes for unusual work are taken as messages, never given.
- Every call is recorded and summarised in the CRM, with callers told about the recording.
05 rollout
From kickoff to live in 4 weeks
- Week 1Write the script, approved prices and emergency rules with the owner.
- Weeks 2–3Test with staff playing callers, including difficult and emergency calls.
- Week 4Go live after hours only; review every call for two weeks before extending to busy daytime overflow.
06 after
What changes
Every call gets answered, including at night and on weekends. Callers get a booking or a clear next step, and your team starts the day with tidy call notes instead of a full voicemail box.
Assume 250 after-hours calls a month, 30% of which currently hang up at voicemail, and an average job value of $350.
250 × 30% = 75 lost callers a month. If one in three would have booked, that is 25 jobs × $350 = $8,750 of work a month.
What to measure
- Missed calls and abandoned calls.
- Jobs booked outside office hours.
- Time from emergency call to on-call contact.
- Callers who ask for a person, and why.
07 evidence
What the research says
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
Across 5,179 customer support agents, an AI assistant raised issues resolved per hour by 14% on average and by 34% for novice and lower-skilled agents.
NBER, Generative AI at Work (Brynjolfsson, Li and Raymond) (opens in a new tab) · 2023
08 lessons
Mistakes to avoid
- Pretending to be human. Say it is automated; it builds trust and is required in more places every year.
- No clear emergency rule. Decide exactly what wakes someone up.
- Not reading the summaries. The morning team must act on them.
09 questions
Common questions
- Can it make outbound calls too?
- It can, but in the US AI-voice calls need the person's prior express consent. Inbound answering does not have that problem.
- What about medical practices?
- It works the same way, but clinical questions are always taken as a message for a clinician.