20% of customers are lost after they have already written to you.

That many people were given a price and not a single invitation to come in. Formally the question is answered and the conversation looks successful — while the person walks off to compare prices elsewhere. This is the cheapest loss of them all: they had already written, they had already been answered, all that was missing was a time.

• 2–19%were told to call instead of being booked right there in the chat
• 6–16%show a booking in the chat — and what happened next is visible nowhere
• $230–350is what the person who was never invited had been ready to spend

Take two ads with the same impressions, the same click-through rate and the same cost per conversation. In the ad manager they look equally successful, and both keep running. But connect every message to a business outcome: did the person book, how often did they come back, and how much did they spend? It turns out that the cost per booking can differ by five to thirty-one times between them, while some ads never produce a booking at all. The ad manager cannot show this because the outcome of a conversation sits outside of it. Delivery, meanwhile, watches whether the conversation goes anywhere after the click, so a slow inbox quietly raises what you pay for impressions.

• 8–22%of the budget goes to ads that produce no bookings at all — $80–220 of every $1,000
• 61%of ad spend lands in conversations that were answered late, or never
• 60%of paid clicks never name a service — a click was bought, not a customer
• $1.25–2.08the real cost of a single paid conversation

That share of messages comes in at night and on weekends — at every business in the sample. This is not junk traffic: where these enquiries started getting answered straight away, the night began producing up to half of the hottest leads.

• up to 35%of all bookings started at night once night messages were answered
• halfof the hottest enquiries arrive outside working hours

Counting every enquiry as it is, nights and weekends included, that is the average time to a first real answer. The typical case looks perfectly fine — the losses live in the tail, and the tail is exactly what an owner never sees. Where these messages started getting answered immediately, those same hours turned into seconds.

• 3–29%of enquiries get no answer at all — roughly one in ten at a typical business
• 9–41 his what the slowest tenth of answers waits — for most, more than a day
• 11 min – 5.4 his what the typical case shows — which is exactly why the problem stays invisible
• 5–8%asked for a price and never received a single number

That many "first answers" in the conversations we went through were a template. Your own statistics look fast afterwards, the conversation is marked as answered and slides down the list — and your manager stops seeing it. Meta, incidentally, does not count an auto-reply as an answer at all: until somebody replies properly, as far as it is concerned you are silent.

• 100%of the businesses in the sample were running one of these templates
• Not countedMeta does not treat an auto-reply as an answer — it will not make your impressions any cheaper

That many conversations at a single business had genuine interest, then the person went quiet and nobody ever came back to them. Follow-up is either not done at all or it is dry and templated, and nobody replies to that. We tried writing individually, reading each conversation properly: it works markedly better than a template, but it is a full-time job that a manager cannot carry mid-flow, reminders or not.