Skip to content
ClickMasters
01 —
Case study

AI Receptionist for Small Business:
Every Call, Text and Email Answered in Seconds

Clickmasters built an AI receptionist and lead response system for trades and service businesses such as plumbers, electricians and HVAC contractors. It answers the phone when the owner cannot, texts back missed calls, replies to texts and emails within seconds, qualifies the job and books it into the calendar. Emergencies go straight to a person, and every lead lands in one record with the recording, the transcript and a summary.

Get the full breakdown

No pitch. Just the playbook we used.

We reply in under 24 hours. No spam, no newsletters.

01
Seconds, not hours
02
AI answers, books, follows up
03
24/7 coverage
04
Tracked to revenue
02 —
At a glance

At a glance
the full brief.

Every fact that makes this project real: what it is, who it's for, and what's inside.

Project typeAI receptionist and lead response system by Clickmasters
ProblemSmall businesses lose leads because nobody answers the call, text or email in time
SolutionAI phone agent, missed-call text-back, AI replies to texts, emails and web forms, booking, hand-off and follow-up
Core technologyGemini 3.8 Live, Gemini 3.8 Flash, Pipecat, Twilio, Next.js, Supabase
Build[15] working days, 2 developers
Headline resultFirst text reply in [ ] seconds; [ %] of test leads booked or handed to a person
03 —
The Challenge

The challenge: the
job goes to whoever answers first

A plumber on a roof cannot answer the phone. An electrician under a house does not see the text. By the time they call back, the customer has rung the next name on the list.A Harvard Business Review audit of 2,241 US companies found that 23% never responded to a web lead at all, and the rest took 42 hours on average. A companion study of 1.25 million leads found that firms that tried to make contact within an hour were nearly seven times as likely to qualify the lead as firms that waited one hour longer.

Why speed to lead is hard for a small business

The owner is the receptionist.

Whoever does the work also answers the phone, so jobs and enquiries compete for the same pair of hands.

Three channels, no system.

Calls, texts and emails land in three places, and nobody sees the whole picture.

A bad answer is worse than no answer.

An AI answering service that invents a price or books a job outside the service area costs more than a missed call.

AI receptionist platform answering calls, texts and emails for a trades business
04 —
Gaps We Closed

What the brief left out:
nine gaps we closed

The brief described the goal: answer leads before they go elsewhere. A product that small businesses can trust with their phone line needs nine more things, and we added each one to the scope.

No definition of a "captured" leadNobody can say whether the system worksNine measures with pass thresholds, fixed before the build started
No limits on what the AI may sayA model will happily invent a price or a promiseA business profile as the only source of answers, with rules enforced in code
No hand-off or failure planEmergencies and upset callers need a person, and carriers and AI services failLive transfer and alerts for emergencies; voicemail and text-back if anything breaks
No bookingA lead that is answered but not booked is lost a second timeReal calendar openings, a confirmation message, no double bookings
No follow-upMost leads do not decide in the first replyThree timed nudges that stop on a reply, a booking or an opt-out
No single recordThe same person calls, then texts, then emailsOne timeline per customer, matched by phone number and email address
No carrier or consent rulesTexting rules differ by country and can block a launch for weeksAI and recording disclosure on calls, opt-out on texts, sender registration planned per country
AI receptionist platform answering calls, texts and emails for a trades business
05 —
The Solution

The solution: an
AI receptionist that answers, qualifies and books on every channel

Seven steps, one system. Every channel lands in the same place.

Solution illustration
01
Answer the call.The business keeps its number. The owner's phone rings first for 15 seconds, and if nobody picks up, the AI phone agent answers. It says it is an AI assistant, says the call is recorded and asks how it can help.
02
Text back missed calls.If a caller hangs up early or a call fails, a text goes out within seconds asking what they need. The conversation carries on by text.
03
Reply to texts and emails.An inbound text gets an AI reply in seconds, and an email gets one within two minutes, in the same thread. The AI asks one question at a time.
04
Qualify the job.On every channel the AI collects the same six details: name, contact number, the job, how urgent it is, the suburb and a preferred time. It answers only from the business profile: services, service area, hours and listed prices.
05
Book it.The AI offers two real openings from the calendar, books the one the customer picks and sends a confirmation.
06
Hand off what it should not handle.A burst pipe or a flooded kitchen triggers an immediate transfer to the on-call number, with a text alert that summarises the call. Anything outside the profile gets a promised callback and an alert to the owner.
07
Follow up and record.A lead who goes quiet gets three polite nudges over three days, which stop the moment they reply or opt out. Every call, text and email lands on one timeline per customer, with the recording, the transcript and a summary.
06 —
Our Rules

Rules the AI cannot break

For texts and emails, code checks each reply against these rules before it is sent. On calls, bookings and transfers pass through the same checks, and every transcript is audited afterwards.

01It never states a price that is not in the business profile.
02It never books outside the service area or the opening hours.
03It says it is an AI at the start of every call and whenever it is asked.
04It stops messaging the moment someone replies STOP.
06 —
Test Lab

The Test Lab: proof before a customer calls

Forty test leads, 15 calls, 12 texts, 8 emails and 5 web forms, cover routine jobs, emergencies, price shoppers, out-of-area requests, wrong numbers and spam. Each script has an expected outcome. A test run plays them all and reports the nine measures in the results table below.

40
Test Scenarios
9
Pass Thresholds
AI receptionist platform answering calls, texts and emails for a trades business

What a handled lead looks like

Sample for layout only. Replace these rows with a real timeline from a live session before publishing.

  • 1
    10:02:11Call

    Owner's phone rings for 15 seconds, no answer

  • 2
    10:02:27Call

    AI agent answers, says it is an AI and that the call is recorded; caller reports a leaking tap

  • 3
    10:04:05Call

    Visit booked for Thursday at 9:00; call ends

  • 4
    10:04:09Text

    Confirmation sent to the caller; summary sent to the owner

07 —
Tech Stack

Tech stack
and open-source components

The system combines carrier and AI services with open-source parts, so the call flows, the data and the test method stay under the client's control.

Number, calls and texts
Twilio Voice and Messaging
Rings the owner first, streams call audio to the AI, sends and receives texts
AI phone agent
Gemini 3.8 Live
Speech-to-speech model with interruptions, 70 languages and tool calls for booking
Voice pipeline
Pipecat
Open-source framework that connects phone audio to the voice model
Text and email replies
Gemini 3.8 Flash
Every reply arrives as JSON: the message, the lead details and the next action
Email in and out
Postmark
Delivers inbound email as a webhook and keeps replies in the same thread
Booking
Cal.com
Open-source scheduling with real availability and no double bookings
Web app
Next.js, shadcn/ui
Lead inbox, timelines, business profiles and dashboard
Data, files, login
Supabase
Postgres, recordings storage and authentication in one open-source stack
Timers and follow-ups
pg-boss
Job queue inside Postgres for text-backs and nudges
Prompt testing
promptfoo
Catches regressions when a prompt or model changes
Tracing and cost
Langfuse
Tokens, cost and latency recorded for every reply
AI receptionist platform answering calls, texts and emails for a trades business
08 —
Results

Results: how fast
and how accurately the AI handled test leads

We fixed the pass thresholds before writing any code, tuned the system on 20 practice leads, then measured it on [40] test leads it had never seen.

Every value below is a threshold, not a measured number yet. Replace the bracketed values with the real numbers from the Test Lab before publishing.

01First reply to a textReply sent time minus inbound received time, median10 seconds or less
02Text-back after a missed callText sent time minus call end time, median15 seconds or less
03First reply to an emailMedian2 minutes or less
04Voice response delayEnd of caller speech to start of agent audio, median per call1.5 seconds or less
05Lead details capturedMatched fields + expected fields90% or more
06Bookings madeBooked + scenarios that should book; double bookings must be zero85% or more
07Emergencies handed offEmergency scenarios transferred or alerted within 30 seconds100%
08Invented prices or promisesGuard breaches found in transcripts and messages0
09Cost per leadAI plus carrier cost per scenario, median$0.30 or less
AI receptionist platform answering calls, texts and emails for a trades business
10 —
Backing Document

Read the
full working.

Twelve parts behind the case study, from the opportunity to the sources. Pick a tab.

PART 01

Case study

A Clickmasters AI Automation project for service and trades businesses: Every call, text, email and web form answered in seconds, qualified, booked, followed up and tracked to revenue. This document covers the solution design and the build. Every number in it is a target, a modelled scenario or a test result, and is labelled as such.

Document map

Twelve parts, ordered from what a buyer reads first to what an engineer builds from.

PartWhat it holdsStatus
01
1. Case study
Executive summary, business challenge, architecture, lead lifecycle, before and after, roadmap, deliverables
Complete
02
2. Upwork brief
The short version that accompanies a proposal
Complete
03
3. Solution design
Capture, voice agent, missed-call recovery, SMS, email and web agents, qualification, scoring, routing, booking, CRM, follow-up, human hand-off, knowledge base
Complete
04
4. Architecture
Stack, build vs buy, data model, failure handling, monitoring, analytics, attribution, compliance and security, multi-tenant design
Complete
05
5. Workflow library
32 workflows
Complete
06
6. Feature matrix
64 features and the automation inventory
Complete
07
7. Industry playbooks
Qualification flows for 14 trades
Complete
08
8. Scenarios, ROI and QA
Three scenarios, the ROI model, 48 test cases, final quality audit
Complete
09
9. Build spec
The build: scope, stack, data model, workflows, prompts, tests, plan, costs
Complete
10
10. Website page
Public, shorter version of the case study for the Clickmasters site
Draft, needs the test numbers
11
11. SEO pack
Metadata, keywords, links and schema for the website page
Complete
12
12. Sources
Every page opened for this document
Complete
12 rows3 columns

Executive summary

Service businesses lose jobs they have already paid to attract, because nobody answers the call, the text or the form in time. The AI Lead Response & Revenue Recovery Engine closes that gap. It answers every enquiry on every channel within seconds, qualifies it, books it or hands it to the right person, follows up until there is an outcome, and ties the result back to the marketing source that produced it.

The design rests on five decisions:

  • One intake layer for every channel. Calls, missed calls, voicemail, SMS, email, web forms, web chat, WhatsApp and ad lead forms enter through the same gateway and land on the same record.
  • Two AI agents, one brain. A real-time voice agent and a text agent share the same business knowledge, qualification rules and guardrails.
  • Rules in code where it matters. Service area, opening hours, prices, consent, opt-outs and emergency routing never depend on a prompt.
  • A system of record. Every lead becomes a contact, an opportunity and a timeline, with transcript, summary, score, owner and source.
  • Production discipline. Queues, retries, fallbacks, monitoring, a regression suite of test leads, and a human takeover path for every conversation.

The case for speed is well documented. A Harvard Business Review audit of 2,241 US companies found that 23% never responded to a web lead and the rest averaged 42 hours. Its companion study of 1.25 million leads found that firms that tried to make contact within an hour were nearly seven times as likely to qualify the lead as firms that waited one hour longer.

What a technical buyer needs to seeWhere the working is
01
That we understand how leads are lost
The business challenge, below
02
That the architecture is credible
Part 4: stack, build vs buy, data model
03
That every workflow has a failure path
Part 5: 32 workflows; part 4: failure handling
04
That the AI is controlled
Part 3: guardrails and human-in-the-loop
05
That it can actually be built
Part 9: build spec
06
That value can be measured
Part 4: analytics and attribution; part 8: ROI model and QA
6 rows2 columns

Status. The solution design is complete and the build is fully specified.

The business challenge: enquiries arrive everywhere and are answered late

A service business rarely has a lead problem. It has a response problem. On an ordinary day the same company receives:

  • A call while every technician is on a job, and another at 9:30 pm
  • A Facebook Lead Ad submission and a website form
  • A text to the office mobile, an email to info@ and a website chat message

A WhatsApp message asking for an urgent quote

Without a central system, each of those enquiries can fail in nine different ways.

How the lead is lostWhat it looks likeWhat it costs
01
Missed
The call rings out during a job or after hours
The caller rings the next business on the list
02
Answered slowly
A form or email is read hours later
The prospect has already booked elsewhere
03
Forgotten
A text is read on site and never answered
A lead already paid for in ad spend is wasted
04
Duplicated
One person calls, texts and fills in a form: three records
Two people call back, or nobody does
05
Poorly qualified
No address, urgency or job detail is captured
Wasted call-backs and site visits
06
Routed wrongly
An emergency sits in a shared inbox
The job that needed speed gets none
07
Never followed up
One reply, then silence
Undecided prospects drift to competitors
08
Never in the CRM
The lead lives in a phone's call log
No pipeline, no forecast, no history
09
Never attributed
Nobody knows which campaign produced the job
Marketing budget is steered blind
9 rows3 columns

Why this is revenue leakage

The business has already paid for these enquiries through advertising, SEO, reputation and referrals. Every one that is missed or answered late is revenue that was earned and then dropped.

The size of the leak is simple arithmetic: enquiries × share not answered in time × booking rate × close rate × average job value. Part 8 turns that into a transparent model with conservative, moderate and aggressive scenarios, all clearly labelled as illustrations.

The HBR figures above measured web leads at larger companies in 2011. The mechanism has not changed and applies even more sharply to a trade: the first business to hold a real conversation is in the strongest position to win the job.

Solution overview: one engine behind every channel

The engine is a lead core with channels above it and the business's own systems below it. AI agents do the talking, but they can only act through coded rules and the business's own knowledge.

Architecture diagram showing communication channels feeding a lead core, with AI agents, business rules and a knowledge base connected to integrations and analytics.

Read it top to bottom. A channel event is verified and queued, matched to one person in the lead core, handled by an AI agent, and written back to the calendar, the CRM and the analytics store. Part 3 details each agent and part 4 the stack behind each layer.

Lead lifecycle: from first contact to the next job

A lead passes through 17 processing steps, grouped here into five stages. The loop at the bottom is what turns a response tool into a revenue system.

Architecture diagram showing communication channels feeding a lead core, with AI agents, business rules and a knowledge base connected to integrations and analytics.

The full sequence is: lead source, omnichannel intake, identity resolution, CRM contact, intent detection, qualification, urgency detection, spam detection, lead scoring, AI conversation, the booking, quote or dispatch decision, calendar and CRM integration, automated follow-up, human escalation, conversion, attribution and analytics, then reactivation, review and referral.

Before and after

BeforeAfter
01
Calls missed during jobs and after hours
Every call answered, by the owner or by the AI agent, around the clock
02
Replies that take hours
First response in seconds on every channel, measured per lead
03
Details retyped by hand, or never recorded
Contact, opportunity and timeline created automatically
04
Calls, texts, emails and forms in separate places
One intake layer and one record per person
05
No consistent qualification
The same questions asked every time, tailored to the trade
06
No idea which marketing produced the job
Source, campaign and first and last touch stored on every lead
07
Follow-up when someone remembers
Timed sequences that stop on reply, booking, opt-out or human takeover
08
Leads forgotten
SLA timers and a callback queue, with alerts when a lead is waiting
09
No numbers
A dashboard from lead to revenue, with cost per conversation
10
AI seen as a risk
Guardrails in code, a human takeover path and a regression suite
10 rows2 columns

Implementation roadmap

Implementation roadmap showing six phases with exit gates

Each phase ships something a business can use, so value does not wait for phase 6. Indicative durations are in part 2, and the phase 1 discovery checklist covers ten items: lead sources, current CRM, phone system, business hours, service areas, qualification logic, calendars, staff, escalation rules and compliance requirements.

What Clickmasters delivers

The engagement covers 22 deliverables in six groups.

GroupDeliverables
01
Strategy and design
Automation discovery, solution architecture, documentation
02
AI agents
AI voice agent, AI SMS agent, AI email processing, knowledge base
03
Lead handling
Web lead automation, missed-call recovery, lead qualification, lead scoring, lead routing, follow-up workflows
04
Integrations
CRM integration, calendar integration
05
Measurement
Analytics dashboard, attribution, monitoring
06
Delivery
Testing, staff training, deployment, ongoing optimisation and support
6 rows2 columns

Why Clickmasters

We would rather show the working than make the claim, so this section is short.

  • One team for the whole funnel. Clickmasters is a 13-year-old digital agency and software house. The people who build the automation also run the SEO, paid media and CRM work that produces the leads.
  • Engineering before prompts. Decisions live in code, every step has a failure path, and test leads are replayed before any change goes live.
  • Honest numbers. Targets, modelled scenarios and test results are labelled as such.
  • A build path, not a slide. Part 9 is a specification our developers can start on tomorrow.

Conclusion

A missed call is not a telephony problem. It is a revenue-operations problem that happens to start on a phone.

This engine treats lead response as infrastructure: capture every enquiry, decide in code and tested prompts, act in the systems the business already uses, record every step and prove the result. The next step is the build in part 9, a number that anyone can call.

09 —
FAQ

Frequently
asked questions

Straight answers, no marketing fluff.

What is an AI receptionist?

An AI receptionist answers a business's calls, texts and emails automatically, in natural language. It answers questions from the business's own information, takes the caller's details, books appointments and passes urgent matters to a person.

How fast should a small business respond to a new lead?

As fast as it can. Harvard Business Review research found that firms that tried to contact a web lead within an hour were nearly seven times as likely to qualify it as firms that waited one hour longer. This system replies to texts within seconds.

What is missed-call text-back?

When a call goes unanswered or the caller hangs up, the system sends a text within seconds asking what they need. The conversation continues by text, so the lead is not lost to the next business on the list.

Will callers know they are talking to an AI?

Yes. The agent says it is an AI assistant at the start of every call and whenever it is asked, and it says the call is recorded. A caller can ask for a person at any time.

What happens when a caller has an emergency?

Each business sets its own emergency rules. When a caller describes one, the agent stops qualifying, transfers the call to the on-call number and sends an alert with a summary.

How much does an AI phone agent cost per call?

By our estimate, a three-minute call costs roughly 15 to 20 US cents in AI and carrier fees on a US number at October 2026 rates. Texts and emails cost less. The measured figure from our test run is in the results table.

Does it work with my existing phone number and calendar?

Yes. The business keeps its number and forwards unanswered calls to the AI, or moves the number across. The system books into Cal.com, and Google Calendar or a job-management tool can be connected instead.

Does an AI receptionist work outside the United States?

Calls work on local numbers in many countries. Texting rules differ: the US requires carrier registration, and New Zealand requires a dedicated short code that takes five to six weeks to set up. We plan the sender setup for each country before launch.

Want every lead

answered in seconds?

Book a free automation audit. We will map how calls, texts and emails reach your business today, show you where leads leak and estimate what an AI receptionist would cost per lead.

AI voice agents · AI lead generation · AI chatbots · AI voice agents for construction and trades

CLICKMASTERS · CLICKMASTERS · CLICKMASTERS · CLICKMASTERS · CLICKMASTERS · CLICKMASTERS · CLICKMASTERS · CLICKMASTERS ·