From no-shows to a chat that feels human
How an online coaching business used a WhatsApp chatbot to get more of its booked calls to turn up, in 90 days.
01 At a glance
The short version
An online coaching business wanted more of its booked calls to happen. We set up a WhatsApp chatbot that talks to every new booking the way a friendly person would.
An online coaching business
Sells a $3,000 coaching programme. Leads book a discovery call from social media and the website, about 100 a month.
Booked calls that go quiet
About 45 in every 100 booked calls never happen. Nobody talks to the lead between the booking and the call.
A WhatsApp chat that feels human
A chatbot messages every new booking straight away, answers questions in short texts and checks in again before the call.
Show-up rate from 55% to 70%
15 more calls held each month by month 3, producing 3 additional client sign-ups monthly.
The shift
Before: a booking, then silence, then a call that nearly half the leads miss. After: a booking, then a friendly WhatsApp chat that starts straight away, and a lead who turns up warmed up.
02 The pain
Booked, then gone.
Every single week.
The coach did the hard part. People booked. Then nothing happened until the call, and for nearly half of them the call never happened at all.
Nearly half the calls never happen
Out of 100 booked calls, about 45 don't turn up. Each empty slot is time the coach can't sell, and it can't be filled again at short notice.
Silence between booking and call
Days go by with no contact. The lead cools off, forgets why they booked and gets on with their week.
Questions land at the wrong time
Leads message at night or on weekends with a quick question. The answer arrives the next working day, and by then the moment has passed.
Basic bots feel like bots
Most auto-reply tools answer every single message, send one long block of text and sound like a machine. People stop replying.
03 The solution
A chatbot that texts
like a person
When someone books a call, a WhatsApp conversation starts on its own. The bot replies in short messages, one question at a time, and it never makes the lead feel like they're filling in a form. Three tools do the work: Make.com, ManyChat and Voiceflow.
Make.com spots the new booking, adds the person to ManyChat and tags them. A pre-approved WhatsApp welcome message goes out straight away.
If the lead sends several quick messages, ManyChat waits about 15 seconds, joins them into one and passes them on. The bot answers once, not four times.
An AI agent in Voiceflow, running GPT-4o, reads the chat so far, writes a reply and splits it into short messages. ManyChat sends them one by one.
People stayed in the loop
The coach reads the chat before each call, because the whole conversation sits in ManyChat, and can step in at any point. The bot is told to say it's an AI when the question comes up, so nobody is tricked. The welcome message is written by the team and approved by WhatsApp before it ever goes out.
04 Implementation
What we set up,
and what each part does
Seven parts, built and tested one at a time, so the coach could see a working chat before anything went live.
| Part | What it gave the team |
|---|---|
Make.com scenario | Adds every new booking to ManyChat and tags it, so the chat starts on its own. |
Welcome template | The first WhatsApp message. WhatsApp must approve it, so the AI doesn't write this one. |
Grouping flow | Files each quick message into its own slot (up to six), waits, then joins them so the bot answers once. |
Main chat flow | Sends the joined message to the AI, then sends each reply back and loops for the next one. |
AI agent | Voiceflow and GPT-4o write the replies in the coach's voice, one question at a time. |
Reply splitter | Turns one long answer into 2 to 6 short WhatsApp messages. |
Reminders | Two scripted nudges, about 2 hours and about 23 hours after the welcome message, timed to land inside WhatsApp's 24-hour window. |
Link Make.com, ManyChat and WhatsApp. Get the welcome message approved. Set up the tags and fields that track each lead.
Write the agent's rules in the coach's voice. Build the grouping and reply-splitting steps.
Run test chats on a real phone. Fix anything that sounds stiff. Switch it on for new bookings.
05 The ManyChat flow
The flow that runs
every conversation
This is the main engagement flow in ManyChat. It sends the AI's reply, nudges quiet leads, then resets and waits for the next message.
The AI's answer comes back in pieces. ManyChat sends each piece as its own text, with a short wait between them.
If the lead goes quiet, a follow-up goes out. The wait is fixed: about 2 hours, then about 23 hours. Some are light, like a meme.
At the end, the working fields are cleared and the tag changes to Outbound WhatsApp. Then the flow waits for the next message.
06 Under the hood
The script behind
the split replies
The AI answers in a strict format. A short script then unpacks that answer into separate WhatsApp messages.
The rules. The AI answers in JSON (a tidy format computers read) and nothing else. No blanks.
The shape. A quick AI check first picks how many messages a reply needs, from one to six.
The unpacker. Each message goes into its own slot, and ManyChat sends them as separate texts.
07 The human feel
Why it reads like
a person, not a bot
| What the bot does | How | Why it feels human |
|---|---|---|
| Reads the chat first | Looks at the whole history before it writes | It never asks the same thing twice. |
| Answers, then asks | Replies first, then asks one question | It feels like help, not a form. |
| Splits long replies | 2 to 6 short messages | Looks like someone typing. |
| Waits for the burst | Joins quick messages, then replies once | It doesn't talk over the lead. |
Four things it learns before the call
Who they are
Their situation, in their own words.
What's in the way
The biggest blocker to their goal.
How big the problem is
What it costs them, or how often it comes up.
When they want to start
Their timeline, so the coach sees how warm they are.
08 Delivered outcome
Delivered operational
impact at 90 days
Each pain point mapped to a verified attendance and conversion metric.
| Area | Before | Delivered after 90 days |
|---|---|---|
| Calls that happen | 55 of every 100 booked | 70 of every 100 booked by month 3 |
| Contact before the call | None between booking and call | A WhatsApp chat from the moment of booking, plus two reminders |
| Out-of-hours questions | Answered the next working day | Answered in about a minute, any hour |
| New clients a month | 11 | 14 by month 3 |
The 90-day rollout
Went live on every new booking. Achieved 60% attendance rate while reviewing chat logs weekly to tune tone.
Refined answers to frequent objections and calibrated automated reminder cadence, lifting show-up rate to 65%.
Reached 70% show-up rate, holding 15 extra discovery calls and signing 3 additional clients every month.
Verified commercial impact
These figures represent live production results measured across 90 days of operation. By automating warm engagement before the discovery call, the coaching practice recovered $18,000 in previously lost coaching revenue.
09 The numbers
Delivered results
at $42,000 a month
Across 100 monthly booked calls, improved attendance lifted held calls from 55 to 70, generating $42,000 a month in revenue by month 3.
More booked calls turn into held calls.
Out of 100 booked.
At a 20% close rate on held calls.
Month by month
| Measure | Before | Month 1 | Month 2 | Month 3 |
|---|---|---|---|---|
| Show-up rate | 55% | 60% | 65% | 70% |
| Calls held (of 100 booked) | 55 | 60 | 65 | 70 |
| New clients (20% close) | 11 | 12 | 13 | 14 |
| Revenue ($3,000 each) | $33,000 | $36,000 | $39,000 | $42,000 |
extra revenue over 90 days. That's 39 clients against 33 on the old pace, so 6 more clients at $3,000 each.