Call · 15 min
AI systemsMattia Esposito2 September 20267 min read

Inbox AI. The 23:10 enquiry doesn't have to wait until morning.

Hardly any business loses enquiries because it replies badly. It loses them because enquiries arrive through four different channels, and there's nowhere to see them all together.

In brief

The problem is the list. One enquiry comes in on WhatsApp, one by email, one through the website form, one by phone. None of those four places knows what's in the other three, so nobody can tell which enquiry still hasn't had a reply.

Outside hours it replies on its own; during hours, you decide. Recurring questions get an instant reply built from information you've already approved, and the reply says it comes from a system. Quotes, prices and confirmations always go to a person, with the context already gathered.

There's one metric, agreed up front: the average time to first reply to an enquiry. We measure where it stands today, set the threshold the work has to beat, and compare after thirty days.

This page covers a single piece of the system. The other pieces, and how we choose which one to start with, are on the services page.

Four channels, no list

The number of places an enquiry can come from has grown, and each one is another inbox to check. In 2025, 63.57% of European businesses used at least one social network, and in Italy the share was 58.97%: that's the figure from Eurostat's survey of ICT use in enterprises, which states “In 2025, 63.57% of EU enterprises used any type of social media”.

In accommodation and food service, the same survey puts it at 78.25%, the second-highest of any sector. People in those trades get enquiries through one more channel than everyone else, and that channel doesn't talk to the inbox or the business software.

On the other side is the phone in the customer's hand. Istat, in its 2025 release on citizens and ICT, reports that 79.1% of people aged 6 and over go online using a smartphone. The enquiry goes out when the customer has their phone in hand, not when the office is open.

The most expensive enquiry is the one nobody ever saw. Before you buy anything, it's worth running the numbers: what replying by hand really costs, in six lines and three figures of your own. It arrives in the evening, on a channel nobody is watching at that moment, and by the next morning it's buried under ten other notifications.

The practical problem is that the business has no way of noticing. To know which enquiries went unanswered you'd need a single list, and that list doesn't exist in any of the four channels, because each one only knows about itself.

An unanswered enquiry shows up nowhere. It's the only loss that leaves no trace to read.

The window is narrower than it looks

Contacting a potential customer within 5 minutes of their enquiry makes you up to 100 times more likely to reach them than waiting 30 minutes. The figure comes from the InsideSales.com/MIT study by James Oldroyd, presented in 2007.

That number measures phone call-backs on enquiries from the web, not an email from a food buyer or a message from a couple looking for a venue, and the difference matters. We've written a whole page on what it really says and what it doesn't: the five minutes that decide an enquiry.

What holds up even outside that sample is the direction: the chance of making contact drops fast, and it drops in minutes, not days. For where automated enquiry handling is heading, with the necessary caveats, see Gartner's forecast on customer service.

What the system does, step by step

Every incoming enquiry, from whatever channel, lands in the same place and gets an ID. From there the route is always the same, and every step leaves a log entry recording what happened and when.

StepWhat happensWhat you get
Captureevery channel

WhatsApp, email, website forms and the phone all feed into one list, with the date, the channel and the content.

One place to look, and an answer to the question nobody can answer today: which enquiries haven't been dealt with yet.

Sortingas they arrive

Each enquiry is sorted by urgency and type, and given a priority.

The enquiries that matter are seen first, and people's time goes there instead of on whichever message happened to be opened first.

First replyautomatic outside hours

Recurring questions get an instant reply built from information the business has already approved, and the reply says it comes from a system.

Someone writing in the evening gets a real reply straight away, and the next morning you find a conversation already under way instead of a message sitting there waiting.

Fallback routewhen the model gets it wrong

If the classification doesn't come back, or comes back garbled, fixed rules still sort the enquiry and keep it on the list.

No enquiry disappears when something breaks, and that's the worst damage there could be.

Handover to a personwhen it commits the business

Quotes, prices, discounts and confirmations are prepared with the context gathered, and wait for approval.

You make the call on prices and confirmations, with the context already in front of you and no need to piece it together.

The fallback route, and why it matters more than the model

A system built around a language model has a typical way of failing: the model doesn't respond, or responds in a format the rest of the system can't read, and the enquiry slips through the cracks. The failure makes no noise, just like the problem the system was supposed to solve.

Here there's a second route that doesn't use the model at all. It sorts the enquiry with fixed rules on urgency and content, and keeps it on the list. Every row records whether it was handled by the model or by the fallback, so errors get counted instead of vanishing.

The engine is built and has been tested on its test branches, with no real customers involved. That's a functional check, and we describe it as exactly that.

What goes out on its own, and what waits for a person

Outside working hours the reply goes out on its own, and that's the reason this piece exists. The 23:10 enquiry gets availability, opening hours and the basics straight away: content the owner approved once, which the system simply repeats.

During working hours the owner decides, channel by channel. Some businesses prefer to answer in person while someone's in the office; others let the system run during the day too and step in where needed. It's a switch you can turn on and off whenever you like, without anyone touching the system.

What never goes out on its own is anything that commits the business: a quote, a price, a discount, a confirmation. Those replies are prepared with the context already gathered, and they wait, because they're decisions.

Opening hours can be repeated. A price has to be decided.

The time you get back is the time spent waiting and retyping, never the time spent thinking. If an enquiry needs a person, that person sees it sooner and with more context, not later.

The automatic reply says it comes from a system

From 2 August 2026, Article 50 of Regulation (EU) 2024/1689 on artificial intelligence applies, and for anyone using a system like this it means a duty to inform: people have to know they're dealing with a system.

When it has to be said is set out in paragraph 5 of the same article: “The information referred to in paragraphs 1 to 4 shall be provided to the natural persons concerned in a clear and distinguishable manner at the latest at the time of the first interaction or exposure”. So it goes in the first reply, not partway through the conversation.

So the automatic reply says so, in one line, within the message itself. It costs nothing and removes the only serious risk the feature carries. What this really means for a small business is covered in the AI Act and Italian SMEs, and the systems we use are listed on the AI transparency page.

The same disclosure keeps the tone honest. Nobody finds out halfway through a conversation that they've been talking to a machine when they thought they were writing to the owner, and the reasoning behind this choice is on our page about the principles we build by.

Same mechanism, different name in every trade

The mechanism is identical; the enquiry that gets lost isn't. It's worth looking at your own case, because that's where you see what it's costing you today.

SectorThe enquiry that gets lostWhere you see it
Food and agricultureand export

The email from a foreign buyer that came in at 23:10, from a different time zone, asking about pack sizes and the price list.

A typical day for a small food producer who exports

Hospitalityaccommodation and events

The three WhatsApp enquiries that came in between 10pm and midnight, when people actually plan their holidays.

The run-up to the season for a guest accommodation business

Restaurantsand bars

The call nobody answered during service, which went to the restaurant down the road without leaving a trace.

A dinner service with the phone ringing unanswered

What this piece doesn't do

It doesn't replace your business software, CRM, channel manager or phone system. It works in the layer that doesn't exist yet, the one that gathers enquiries scattered outside those systems and brings them together with the rest.

It doesn't bring in new enquiries. It works on the ones already coming in, which is why it's almost always the first piece to build: the return is measured on your current volume, without depending on a campaign that has yet to work.

It doesn't decide for anyone. It repeats information approved in advance, sorts and flags, and stops where a decision that commits the business begins.

Questions and answers

What's the difference between Inbox AI and a chatbot on the website?

A chatbot lives on a single channel, the website, and answers people who have already landed there. Inbox AI gathers enquiries from WhatsApp, email, web forms and the phone into one list, and its main job is making sure none of them goes unanswered.

The conversation is one part. The other part, the one that really counts, is the list of people nobody has got back to yet.

Do replies go out to customers on their own?

Yes, and you choose when. Outside working hours, replies based on information the owner has already approved, such as opening hours, availability and basic details, go out on their own, and that's where this piece earns its keep. During working hours the owner decides, channel by channel, with a switch that turns the feature on and off without anyone touching the system.

What never goes out on its own is anything that commits the business: quotes, prices, discounts, confirmations. And the automatic reply states that it comes from a system, as required by Article 50 of the EU AI Act, which applies from 2 August 2026.

What happens if the model misclassifies an enquiry?

There's a fallback route that doesn't use the model at all. If the classification doesn't come back, or comes back garbled, the enquiry is still sorted using fixed rules on urgency and content, and it stays on the list.

Every row records whether it was sorted by the model or by the fallback, so errors show up instead of vanishing. A lost enquiry is exactly the damage the system exists to prevent.

Do we have to change our business software, CRM or phone system?

No. Your business software, CRM, channel manager and phone system stay where they are: your team knows them and they work. The work goes into the layer that doesn't exist yet, the one that gathers enquiries scattered outside those systems and brings them into one place.

We only suggest replacing a tool if the analysis shows it's the bottleneck, never out of preference.

How do you measure whether it's working?

By the average time to first reply to an enquiry, agreed as the metric before we start. We measure the current figure over the previous thirty days, set the threshold the work has to beat not to count as a failure, and compare afterwards.

The second number, where a reliable list exists, is how many enquiries in the month never got a reply.

Notes on sources

  1. The five-minute figure comes from the InsideSales.com/MIT study by James Oldroyd, presented in 2007, and measures phone call-backs on web enquiries, not enquiries from food buyers or guests of a hospitality business. We say so because the sample isn't the reader's.
  2. The figures on business use of social media come from Eurostat, Social media: statistics on the use by enterprises, data extracted in June 2026 from the 2025 ICT survey (code isoc_cismt), covering about 157,000 of 1.53 million enterprises. The survey only covers enterprises with 10 or more employees, so micro-businesses aren't included, and it's an EU-wide figure with Italy's value given separately.
  3. The share of people going online by smartphone comes from the Istat release Cittadini e ICT, anno 2025. It measures people, not a business's customers: it tells you which device people use to go online, not which channel a sales enquiry comes through.
  4. The text of Article 50(5) is quoted from the official English version of Regulation (EU) 2024/1689; the link leads to the Italian version in the Official Journal. It's a legal provision, so there's no sample, and the transparency rules apply from 2 August 2026.
  5. This page doesn't report results achieved for a client, because this piece hasn't yet been delivered to a client. The tests mentioned are functional checks run in a test environment, which is why you won't find a recovered revenue figure here.
  6. We don't publish a percentage of lost enquiries that can be recovered: it depends on volume, channel and sector, and we haven't found a reliable sector average for this particular figure. The number is worked out from your previous months, before any quote.

The other pieces in this group

Hours spent doing repetitive work by hand

All the pieces, in the six groups

·The next step

Fifteen minutes, with your case in front of us.

How many enquiries last month never got a reply? If the answer is that nobody knows, that's already the most useful thing to find out, and it can be counted within thirty days. In fifteen minutes on the phone we'll look at it together and tell you where it makes sense to start, even if we never end up working together.

You'll speak to Mattia Esposito, who then builds the system: there's no salesperson in between. If you'd rather measure things yourself before talking, the Diagnostico (in Italian) is twenty questions and five minutes.