Reply first, but reply right: speed wins the lead, judgment keeps it
A famous Harvard Business Review study showed that enquiries go cold within hours. The usual fix is to reply faster to everyone. The twist is that a fast reply only helps if something first decides which enquiries deserve one, and which need a person.
Picture this. It's 9pm, and a message lands on your business WhatsApp: "hi my aircon not cold since yesterday, 3 units at tampines, how much for chemical wash? can come this sat morning?"
You see it at 8am the next day. By then the customer has messaged three other companies, and one of them replied at 9.05pm with a price.
Nobody was lazy. The enquiry arrived when nobody was watching, and it had a short life.
The short life of a lead
In 2011, Harvard Business Review published The Short Life of Online Sales Leads by James Oldroyd, Kristina McElheran and David Elkington. They looked at how 2,241 US companies responded to web enquiries.
Two caveats: the study is from 2011, and it covers US firms, before WhatsApp became a common way customers message small businesses in Singapore and across Asia. Don't treat the exact multiples as true for your shop today.
The shape of the finding is the part worth keeping: interest fades in hours, not days, and a surprising number of businesses never reply at all.
Where the speed advice goes wrong
The obvious lesson is "reply faster", so owners buy a chatbot that answers every message within seconds. That fixes the clock and breaks something else. Speed alone just sends fast replies to the wrong people:
- A spam message offering "first page Google rankings" gets a warm, friendly reply.
- An unhappy customer whose aircon is leaking worse after your visit gets a price list.
- Someone reporting a burning smell and smoke gets asked to "share your preferred timing".
What looks like a speed problem is often a sorting problem. The real question isn't "how fast can we reply?" It's "how fast can we decide what this message is?"
So my version of the rule has three parts:
Decide quickly: is it real, urgent and a fit?
Real means a customer, not spam. Urgent means they want it this week, or something is unsafe. A fit means it's a service you actually offer, in an area you cover.
Answer the good ones fast
If the enquiry has enough detail, reply with something useful: an indicative price, a site visit, or the one or two details you still need. Not "thanks, we'll get back to you".
Hand the tricky ones to a person
Complaints, billing disputes, safety risks and anything the system isn't sure about go straight to a human, with the message summarised so they can act in minutes.
What I learned selling a WhatsApp sales agent
Before my current work I ran Elisha, a WhatsApp sales and lead-generation agent. It served insurance and banking sales teams in Singapore.
The part users valued most wasn't the AI writing their messages. What they valued was finding and qualifying the lead: knowing which conversations were worth their time and which weren't. A salesperson can write a decent message. What they can't easily do at 9pm, across every open chat, is work out which few matter.
The second lesson was about friction. When setup took long, people drifted away before they ever saw the benefit. The qualification had to be short: a few questions that matter, not a long form.
Both lessons point the same way: the value is in the judgment, not the writing.
What this looks like in practice
I built an interactive demo to show the idea: Can this enquiry get a price in 30 seconds? You pick a sample aircon, pest-control or moving business in Singapore and paste in a messy enquiry, Singlish included.
The system asks a few narrow questions about the message. What does the sender want: a new job, a reschedule, a complaint, a billing question, or is it spam? Which service, how big is the job, where is it, did they say when, and is there a safety risk? A fast decision model answers each one with a probability. When it's unsure, a general AI model is asked the same question.
Then plain code, not AI, makes the decision:
- Enough detail: an indicative price range from the business's own price table.
- Missing detail: a reply asking only for what's missing, such as the number of units.
- Can't be priced by message (termites, new installations): offer a site visit.
- Wrong service or outside Singapore: a polite decline. Spam: archived, no reply.
- Complaints, billing and follow-ups on past jobs: always a person, never an auto-reply.
- Safety risk, like a burning smell: a person calls back now.
- The two models disagree on what the message is about: a person decides. It never guesses.
It's a demo with illustrative prices, not a client deployment, so I'm not claiming results for it. It shows where the line sits: fast where the answer is clear, deliberately slow where a mistake would cost you a customer. (For checking every answer before it's sent, see chatbots that check their answers.)
The handover is part of the speed
Handing a message to a person isn't the system failing. Done well, it makes the person faster too.
A large study of 5,172 customer support agents by Erik Brynjolfsson, Danielle Li and Lindsey Raymond, published in the Quarterly Journal of Economics in 2025, found that AI assistance increased issues resolved per hour by 15% on average, and by 30% for less-skilled and less-experienced staff.
That's the point of a good handover: your newest staff member shouldn't open a complaint cold, but see what the customer wants, how urgent it is, and what's already known.
In the MummyReady funnel I built for an insurance team, every booking went to the right advisor instantly, with the context attached. Routing is speed.
None of this is about an AI salesperson pushing out more messages (a different problem). It's the simpler, earlier job: not losing the customers who already asked.