There is a statistic that gets quoted in every sales deck ever written: respond within five minutes and you are dramatically more likely to convert. It is repeated so often that it has stopped meaning anything, which is a shame, because the underlying observation is true and almost nobody acts on it.
The reason is not laziness. It is that "respond within five minutes" is a target, and targets do not do work. A target tells you where you would like to be. It does not tell you what happens at 7:15pm when the enquiry arrives, the office is empty, and the only person who checks that inbox is at their daughter's swimming lesson.
What closes the gap is not trying harder. It is deciding, in advance and in writing, what the first five minutes contains) and then building it so it happens whether or not anyone is available.
Why the tail is the problem, not the average
Pull three months of enquiries out of any small business and plot the time between arrival and first outbound reply. You will not get a nice curve. You will get two humps: a large one under an hour, and a smaller, stubborn one measured in days.
The average of those two humps is a number nobody experiences. Half your enquiries get a genuinely good service. The other half get something that, from the customer's side, is indistinguishable from being ignored.
And the tail is not random. It correlates with exactly the conditions under which enquiries are most valuable: evenings, weekends, holiday periods, and the weeks when the team is busiest because work is going well.
So the design question is not "how do we get faster?" It is "what happens when the usual person is not there?"
What the first five minutes should actually contain
A reply is not one thing. It is three, and they have different requirements.
Acknowledgement. The enquirer needs to know a human organisation received their message and it is not sitting in a spam folder. This has to be immediate and it does not need to be clever. It needs to be accurate: reference what they actually asked about, not "your enquiry".
Qualification. Two or three questions that decide whether this is a job you want, and what kind. Postcode. Rough job type. Timescale. Whether they are the decision-maker. These questions are boring, they are always the same, and they are almost never asked consistently by humans under time pressure.
A next step with a time on it. Not "we will be in touch". Either a booked slot, or a specific commitment ("one of our engineers will call you before 10am tomorrow". The commitment is the thing that stops them working down the list of competitors.
Notice that only the third of these requires judgement, and even that is often rule-based. The first two are pure process. That is why they can be designed and why, once designed, they can run at 7:15pm.
Channel matters more than people expect
The other thing an audit turns up is that businesses respond in the channel that suits them rather than the one the enquirer used.
Someone who sends a WhatsApp message expects a WhatsApp message back. Sending them a formal email with a PDF attached reads as a different company entirely. Someone who filled in a web form at 11pm is not necessarily expecting a phone call at 8am the next morning; a text confirming receipt, followed by a call at a stated time, converts better and annoys fewer people.
Getting this right costs nothing and is mostly a matter of writing the rule down. Getting it wrong quietly halves the effect of everything else you do.
Where AI genuinely helps, and where it does not
Being specific about this matters, because "AI-powered lead response" is the sort of phrase that means nothing.
It genuinely helps with:
- Answering the call you would otherwise miss. An AI voice agent covering evenings and overflow turns a voicemail into a captured enquiry with a job type and a callback slot. This is coverage, not replacement) the hard conversations still go to people.
- Writing the acknowledgement in the enquirer's own terms. Pulling the specifics out of a free-text form and reflecting them back is a job models are good at and templates are bad at.
- Structuring messy input. Turning "hiya do you do bathrooms im in llanelli" into a contact record with a postcode, a job category and a source is unglamorous and enormously useful.
- Keeping the sequence alive. Reminders that stop when someone replies, escalation when they do not, and a clean hand-off to a person at the point where the rules run out.
It does not help with:
- Deciding whether you want the job. That rule has to come from you. A model will produce a confident answer to a question you have never actually decided, which is worse than no answer.
- Pricing. Anything with real commercial consequence stays with a person unless the pricing rules are genuinely mechanical.
- Anything you have not written down. This is the recurring theme of every automation project. Undefined process automates into undefined behaviour.
The failure mode: a fast reply into a void
The most common way this goes wrong is that the front end gets built and the back end does not.
An enquiry is answered in ninety seconds by an agent that qualifies beautifully (and writes the outcome into a transcript that lands in an inbox nobody owns. The customer has been handled. The business has learned nothing, the CRM has no record, and the follow-up sequence never starts because nothing triggered it.
This is why we do the CRM and integration work before or alongside the answering layer rather than after it. The rule is simple: nothing gets a mouth until it has somewhere to write.
What good looks like, concretely
For a typical home-services firm, a designed first five minutes ends up looking like this:
- Enquiry arrives on any channel and is normalised into one record with a source, a postcode and a job type.
- Within sixty seconds the enquirer gets an acknowledgement in their own channel, referencing what they asked about, with a stated next step and a time.
- If it arrived in hours and passes the qualification rules, it routes to whoever is on duty with the context already attached.
- If it arrived out of hours, the voice or messaging agent captures the missing details and offers real slots from the live diary.
- If nothing happens by the stated time, the sequence escalates rather than politely waiting.
- Every step writes back, so next month's report can show the distribution rather than the average.
None of that is exotic. Most of it is configuration and a written rule set. The AI-shaped part is a fairly small slice in the middle) which is usually the right proportion.
Start by measuring, not buying
If you take one thing from this: before you evaluate a single tool, pull the numbers. Missed calls by hour. Time-to-first-response as a distribution. Enquiries by channel. Three months is enough.
Most businesses find the answer is not the tool they were about to buy. That is a good outcome, and it costs nothing but an afternoon.
If you would rather someone else pulled them, that is what the free AI Opportunity Assessment is for, and it feeds straight into the 12-month roadmap if you decide to go further.