Home › Evidence
Show your working
Every number we put in front of a client carries its source, its sample and its date. This is the whole evidence base for what we sell, graded — including the parts that do not help us.
Show your working
Most people selling this will show you a statistic. Almost none of them have read the study behind it. Here is the whole evidence base for what we sell, graded — including the parts that don't help us.
Researchers at UC San Diego, George Mason, Vanderbilt and Cornell analysed 11.6 million real enquiries on a services marketplace and backed it with randomised experiments. A one-hour delay in replying cut the chance of being hired by 46%. A day's delay cut it by roughly 90%. The authors report that delays of just 5–10 minutes already reduce the chance of being hired. Buyers read a fast reply as a signal you'll be reliable later, and there was no evidence that appearing in-demand helps.
One finding matters more than the headline: the speed effect held independently of what the message actually said, and independently of the provider's quality rating. Speed is doing the work on its own.
What it doesn't show — and this one is against us: it's a freelance digital marketplace, not UK trades or professional services, so treat the mechanism as transferable and the exact percentage as not. More importantly, the study never tested an automated reply. Every fast response in it came from a human. Whether an instant answer from software carries the same signal is an open question that nobody has yet measured — including us. We'll know more once our first installs have run.
VanEpps, Hart, Sezer & Amir, "Speed Is a Signal: When Faster Replies Increase Hiring Likelihood", Management Science, published online 2026. Peer-reviewed. doi:10.1287/mnsc.2024.06185 · Read itThis has been measured repeatedly by mystery shop rather than by vendor platform data, which is why we trust it. 55% of 433 B2B firms hadn't responded within five working days (Drift, 2017). Only 365 of 1,000 firms replied at all (RevenueHero, 2024). 40% of 466 home services firms never responded (Valve+Meter, 2018).
What it doesn't show: all US. The nearest UK equivalents are older and smaller — 213 letting agents in 2014, 85 law firms in 2016 — though both found roughly a third to a half of enquiries unanswered for days.
Multiple independent mystery-shop studies, 2014–202428% of UK businesses with fewer than 10 employees use any AI technology; construction is the lowest sector at 13%. From the ONS Business Insights survey, 38,637 responding businesses, published July 2026.
What it doesn't show: adoption, not effectiveness. The ONS does not measure customer service, chat or sales use cases at all, so this says nothing about whether any of it works.
ONS, Artificial intelligence in UK businesses, 20 July 2026You will see this everywhere. It comes from a 2007 analysis of six companies, using call data supplied by a firm selling lead-response software, co-authored by that firm's chief executive. It has never been peer-reviewed or replicated. It measures the odds of starting a conversation at 5 minutes versus 30 — not conversion, and the authors say explicitly that they did not examine close rates. It is routinely mislabelled as an MIT study; MIT neither funded nor published it.
The detail that settles it: the same company's later dataset — 5.7 million leads instead of 15,000 — produced 8×, not 21×. Their own numbers, at 380 times the scale, more than halved their own headline.
Elkington & Oldroyd, Lead Response Management Study, October 2007There is no independent measurement of this category. None. Every booking rate, containment rate and after-hours figure in circulation traces back to a single vendor's own client book. Claimed after-hours enquiry shares range from 26% to 91% across vendors, which tells you nobody is measuring the same thing.
So we won't quote you one. What we'll do instead is measure yours, before and after, and show you the method.
Assessed across the published vendor material in this category, August 2026We looked properly, across construction, trades and professional services, in academic literature, official statistics and trade bodies. There is no independent study anywhere measuring quote turnaround against win rate. The one figure in circulation comes from a proposal-software vendor with no sample size, no method and no dates attached.
It's a reasonable inference from the reply-speed research above. It is not a fact, and we won't dress it up as one. The only number that settles it is your own: win rate on quotes sent within 24 hours against quotes sent in week two. Most firms have never pulled it. We do that in the fit assessment.
Searched August 2026. If you know of a study we've missed, tell us and we'll publish it here.Two projects, described
We don't publish case studies until an install has run long enough to compare against its baseline. Until then, here are two projects described honestly. Both are composites: each is drawn from several real owners we have sat with this year, with the details blended so nobody is identifiable. The baseline numbers are real measurements. There are no outcome numbers, because it is too early to claim any.
The business. Around 40 staff, about £4m turnover, mostly reactive and planned maintenance for commercial landlords and facilities managers. The owner and one estimator hold all the pricing. Every quote waits for one of them.
What we measured first. Five test enquiries over a fortnight, sent the way a customer would send them: two by web form, two by phone out of hours, one by email with a photo attached. First reply averaged 19 hours. A price took six working days. Two of the five were never answered at all. The owner's estimate before we started was "same day, usually."
What was built. The rate card and the estimator's rules of thumb were written down in half a day, then chat on the existing website and a voice agent on a tracked number, both asking the questions the estimator asks. Enquiries are answered within the minute, qualified, and booked for a survey where one is needed; the draft quote is produced the same day from the rate card and goes to the owner or estimator to approve before it is sent. Callers are told at the start they are speaking to an automated assistant. Anything unusual or urgent goes straight to a person with the context attached.
What we will report. The same five-enquiry test repeated monthly, alongside enquiries captured, quotes issued within 24 hours, and win rate on those quotes against the ones that took longer. Those numbers go on the case studies page once there are enough of them to mean something.
Composite of several owners, 2026. Baseline figures are measured. No outcome claimed.The business. Fourteen staff, about £1.8m turnover, built over twenty years by the owner. Every quote that leaves the business is priced by him, because the pricing is twenty years of instinct and it lives in one place. When he is on site, quotes wait. When he is on holiday, quotes wait.
What we measured first. Quotes took four days on average and up to eight. The owner spent roughly twelve hours a week pricing. The lowest score in the diagnostic was owner dependency, and quoting was the single biggest driver of it. The 11pm thought, in his words: "I built this thing, and now it can't function for a day without me."
What was built. Half a day getting the pricing instinct out of his head and onto paper: about sixty rules of thumb, the kind of "stainless, add forty per cent" and "that customer, always allow for two revisions" that had never been written down. Then a quoting system around them. An enquiry arrives by email with drawings; the system reads it, pulls the materials take-off, applies the sixty rules, checks current steel prices and drafts the quote. It goes to the owner to approve. He is still the brain. He is no longer the typist, the calculator and the filing system as well.
What we will report. Quote turnaround and the owner's weekly pricing hours against the baseline above, and win rate on quotes sent within 24 hours against the rest. Separately, whether the general manager can approve routine quotes without him, because that is the point at which the pricing rules stop being one man's instinct and become an asset of the business.
Composite of several owners, 2026. Baseline figures are measured. No outcome claimed.Straight answers
Grade A is peer-reviewed or official statistics with a stated sample and method. Grade B is repeated independent measurement, such as mystery shops, without peer review. Rejected means we traced the claim and it did not hold up.
Because a claim you cannot check is worth nothing, and because the omissions are usually what an experienced buyer wants to know. The reply-speed research, for instance, has never tested an automated reply.
Thin, which is the honest answer. Most of the mystery-shop work is American. That gap is why we are building a UK response benchmark from measured audits rather than quoting someone else's figures.
They are composites: each is several real owners we have worked with this year, blended so nobody can be identified. The baselines are real measurements. We describe what was found and what was built, and stop there, because publishing an outcome before it has been measured against the baseline is exactly what we tell clients not to trust.
Yes. Each entry names the study, the sample and the date, and links to it where a public link exists.
Twenty minutes, no pitch. Bring your enquiry numbers and we will work through the arithmetic together.
Book a call