The State of AI in UK Bars 2026

Foreword, from the founders

Hey, I’m Chanel, a co-founder of Sorino. Over the last few years our team has stood in enough bars on a Friday to know the sound of it. The phone rings behind the bar while the whole team is heads down pouring, plating and running the floor. Nobody can get to it. A booking walks. A message on Instagram sits there until the middle of next week. Nobody did anything wrong. There just were not enough hands at the exact moment the guest reached out.

So we put AI on the channels that were going unanswered, reading from one knowledge base of the venue’s own menu, hours and booking policy. Not to replace anyone, but to cover the hours your people cannot pick up. This report is what we are seeing across UK bars and food-led venues, told plainly, with real numbers where we have them and honest gaps where we do not yet. It sits under a wider Sorino report on all of UK hospitality, and you can read that full hospitality report here.

Section one

The Friday phone never stops

For a booking-heavy bar, a bar-restaurant or a food-led pub, the leak is sharpest at peak. Friday night is exactly when the floor cannot answer the phone, the inbox or the messages, and it is exactly when the bookings are coming in. Four things define this kind of venue, and all four point the same way.

• The phone rings out mid-service. When you are slammed, the calls you miss are the tables you never knew you could have filled. They do not leave a voicemail, they just ring the place down the road.
• The message is now the first step in the booking journey. For this crowd the Instagram or Facebook message is where a booking starts, and it is the one most likely to sit unread for two days. By the time you see it, the party has booked somewhere else.
• The same questions, all day. Are you open, is there a kitchen, do you do gluten-free, can I bring the dog, have you got a table for eight. The same handful of questions, over and over, pulling someone off the floor every time.
• No-shows quietly eat the night. A table held for a party that never arrives and never calls is covers you could have sold twice. Confirmations and reminders that actually go out are what bring that number down.

What operators tell us

• “The phone never stops on a Friday, and we miss calls when we are slammed.”
• “The messages just pile up, we see them two days later.”
• “Same questions all day: are you open, gluten-free, can I bring the dog.”
• “They do not turn up and do not even call.”

The cost of the gap. None of it shows up as a line on the accounts. A missed call is not logged. An unread message is not counted. A no-show is just an empty table nobody sold. Which is exactly why it goes unfixed for years. The language that actually describes it is covers, bookings, no-shows, the floor, the diary and the DMs, not “a chatbot”.

Section two

The squeeze every operator already feels

Two things happened at once. Staffing got harder, and guests stopped waiting. Bars are caught in the middle. On the supply side, the people problem is real and measured.

Read those two together and you get the shape of the problem. There is often no spare person to work the phone, and when the shortage bites, venues start closing the very hours guests want to reach them. On the demand side, patience has collapsed.

An instant reply is no longer a nice touch, it is the baseline, on whatever channel the guest happens to use. The gap between “we are short-staffed” and “answer me now” is exactly where bookings, covers and reputation leak away, and for a Friday-night bar that gap is widest at the worst possible moment.

Section three

What the AI actually does: four channels, one knowledge base

The shape is simple. One shared knowledge base, your menu, hours, booking policy and the answers to those same questions, feeds four channels so every one answers the same way, in your venue’s own voice.

The point that matters to a sceptical operator: it is not four bots. It is one source of truth, four front doors, and a clean hand-off to your team the moment a conversation needs a human.

Section four

What good looks like, and what we will not let AI do

The fastest way to lose an operator’s trust is to over-promise. So the guardrails are part of the product, not an afterthought.

• Allergens and dietary questions answer from the record, or hand off. No guessing on anything safety-critical.
• It sounds like the venue, because it reads from the venue’s own words. Not a generic script.
• A human takes over the second it gets complicated. The agent knows the edge of what it can handle.
• Everything is on the record. Full transcripts, clear permissions, clear escalation rules.
• It works alongside the floor, it does not replace it. Coverage for the hours you cannot pick up, not redundancy.

Section five

Getting started: try it before you trust it

You should never have to take a vendor’s word for any of this. So the route in is simple: you build your venue’s agent live on the site, ask it the awkward questions yourself, and hear how it answers, before anyone talks about a bigger commitment. If it passes, the next step is a fixed-scope trial with clear measures agreed up front:

• How fast guests get a first answer across the channels you choose.
• After-hours and mid-service coverage, the calls and messages that used to go unanswered.
• No-shows, and the bookings and function enquiries that used to leak away.

Low risk is the whole point. You see it working on your own venue, in your own voice, before you commit to anything bigger.

Sources

We do not publish a figure unless it is real and attributable. The numbers in this report are:

• UKHospitality workforce survey, 2023: 61% of businesses reporting staff shortages, 42% cutting opening hours as a result.
• HubSpot, 2025: 82% of consumers say an immediate answer is very important.

The “what operators tell us” lines are drawn from how the buying committee in a bar or food-led venue describes the pain. They are representative phrasing, not verbatim-attributed quotes, and no venue names, deployment counts or results are claimed.