Client name withheld under NDA.
Aircon servicing is a race. Someone's unit stops cooling on a hot Saturday, they message three companies, and the first one to reply with a price and a slot usually gets the job. This company had a strong crew and good reviews, but enquiries came in by WhatsApp, web form and phone to two coordinators who were also scheduling the technicians. Replies often took more than an hour, and anything sent after 6pm waited until the next morning. By then, many customers had already booked someone else.
The coordinators were answering the same questions all day: how much for a general service, what a chemical wash involves, whether they come on Sunday. The owner wanted faster replies without hiring another coordinator. We measured the current workload before building anything, so the results could be compared fairly.
We went through three months of WhatsApp and web form messages and sorted them by type. About two thirds were routine: pricing ranges, what each service includes, areas covered and slot availability. The rest needed a human, such as a leaking unit or a landlord quote for multiple flats.
A chatbot should take the repetitive work and pass the rest on. Knowing the split up front stops it from guessing at questions it should never answer. It also tells you how much time the bot can realistically free up.
We wrote the conversation flows with the coordinators: greet, identify the issue, collect the unit type, number of units, area and preferred time, then offer a slot. Price answers were given as ranges, with a clear line about when a technician must see the unit first.
Singapore customers expect a reply in the language they wrote in. The assistant was set up for English, plus common Mandarin phrases, and hands off to a person when it is unsure. Customers can always type 'speak to someone' to reach a coordinator.
The assistant was connected to the company's WhatsApp Business number and a website chat widget, and linked to the booking calendar so it could offer only real open slots. Every conversation that reaches a booking sends a summary to the coordinators.
A bot that books slots the team cannot serve creates more work. Connecting it to the live calendar removes that risk. Technician travel time between areas was built into the slot rules, so the bot would not offer Tampines at 2pm and Jurong at 2.30pm.
We ran 80 test conversations covering typical, awkward and off-topic questions, reviewed every wrong answer with the coordinators and added human handover triggers for complaints and urgent faults. The bot went live on WhatsApp first, then the website.
A soft launch with the team watching the first week of chats catches problems while the volume is still small. The coordinators kept a note of every answer they would have changed, and those notes became the first week's fixes.
Each week we read a sample of conversations, fixed the answers that confused customers and added new questions as they came up, such as warranty, condo management rules and what to do when a unit is leaking. Answers that customers kept asking a human to repeat were rewritten in simpler words. A monthly report tracked reply time, handled enquiries and bookings.
The first version never covers everything. Reading real conversations is how the assistant becomes more useful over the first few months.
Two months after launch, customers got a reply within about two minutes at any hour, and the coordinators spent far less time retyping the same answers. More enquiries turned into booked jobs, partly because customers were no longer lost overnight. The coordinators also reported fewer missed calls, since routine questions were being settled in chat.
Enquiry audit, conversation flows, integrations, testing and soft launch.
Weekly review of conversations, new answers added, handover rules adjusted.
Reply time, bookings and staff hours compared against the baseline.
Cutting the first reply from 95 minutes to 2 minutes lifted the enquiry-to-booking rate from 24% to 37%. In urgent-repair categories, the first company to answer the question often gets the job.
After-hours enquiries went from 9 to 41 a month. Those customers were always there, but they used to be answered the next morning after another company had replied. A reply at 10pm that offers a Saturday slot is often enough to win the job.
With 62% of enquiries fully handled by the assistant, the coordinators spent 9 hours a week on routine replies instead of 22. That time went to scheduling and calling customers back.
How we help Singapore businesses turn interest into enquiries.
Pages built to turn visits into messages and bookings.
Our full range of services for Singapore businesses.
What a lead generation programme looks like end to end.
Tell us how enquiries reach your team today. We will show you what a chatbot could take off their plate and what it would cost.