What independent hospitality operators are missing, what it is costing them, and what to do about it.
We have spent our working lives in hospitality. We love this industry. We love the craft of it and, more than anything, we love the people, the teams who make it work and the guests who come back. The Impossible Dream was never only about AI. It is about hospitality as a whole: the standards, the service, the human moments that no tool will ever replace.
That is also why we have spent the last two years living with AI properly. We have tested it, argued about it, and changed our minds more than once. Some days it has felt like it changes everything. Other days it has felt massively overhyped. But one thing has stayed rock solid through all of it. Hospitality should not just adapt to AI. Hospitality should help shape how AI grows up.
Here is the part that bothers us. Almost the whole conversation about AI in hospitality is happening outside the industry. It is happening in tech rooms and on stages, and moving fast. There are brilliant people building this technology, and hospitality needs to be building it alongside them, bringing what only a life on the floor can teach you. The people who understand this industry best are the people who live it every day, and right now too few of them are in the room.
So this paper is not us telling hospitality what to think. It is us asking hospitality what it thinks. It is written with the industry, not about it. It comes from people who have actually run these businesses, with no hype and no false promises. Some of it you will agree with. Some of it you might completely disagree with, and honestly, we hope you do, because the best conversations rarely start with everyone nodding along. We just want to start the conversation, and we want operators front and centre in it, sat alongside the people building this, not left outside while it gets decided.
Co-Founders · The Impossible Dream
Independent operators (restaurants, bars, hotels, cafes, event venues) are being squeezed from every direction. The costs of energy, ingredients, wages and rent have all risen sharply. The cost of getting the marketing wrong, the training wrong, or the reporting wrong has risen with them.
At the same time, a gap is opening up. A small but growing number of operators have started using AI tools to reduce admin, speed up marketing, sharpen their reporting and train their teams more consistently. The gap between them and the ones who haven't started is widening every week.
Every employee who leaves takes their product knowledge, their regulars and a slice of your margin with them. Replacing a single member of staff typically costs between 50 and 75 per cent of their annual salary once you factor in lost productivity, recruitment time and training, somewhere between £12,000 and £18,000 for a worker on the National Living Wage.
The financial pressure is mounting on every line of the P&L. Labour costs are running at 30 to 35 per cent of revenue for most independent restaurants and bars (against an industry target of 25 to 30 per cent) and have been pushed higher still by successive National Living Wage increases and the employer National Insurance changes announced in the October 2024 Budget and introduced in April 2025, when the rate increased from 13.8 to 15 per cent and the secondary threshold was reduced from £9,100 to £5,000. Gross profit margins, which typically run at 65 to 70 per cent on food and drinks combined, are being compressed from below by supplier costs and from above by the labour overhang. The average independent UK restaurant now operates on low single-digit net margins, of the order of 3 to 5 per cent. There is not much room for error.
That pressure has deepened, not eased, since these figures were first compiled. The November 2025 Budget confirmed a further National Living Wage rise to around £12.71 an hour from April 2026, adding an estimated £1.4 billion to the sector's annual cost base. The effect is already visible in employment. UKHospitality estimates that between 89,000 and 100,000 hospitality jobs have been lost since the 2024 Budget, and 3,353 accommodation and food service businesses entered insolvency during 2025, one of the highest counts of any sector in the UK. We are not going to claim that AI is what stands between a business and insolvency; the cost pressures driving those closures are far larger than any admin saving. But for an industry operating on net margins this thin, low single digits and often nearer 3 per cent than 5, avoidable inefficiency is a luxury no one can afford. The capacity you get back is your margin for error: the hours to fix the rota, chase the enquiry, and catch the GP slip before it compounds. It does not remove the pressure. It gives you a little more room to stand it.
Against this backdrop, independent operators have been slow to adopt the one category of tools that could materially reduce the admin burden, improve consistency and give management better visibility of the business. In February 2026, the UK government published its AI Adoption Research: a nationally representative study of 3,500 private-sector businesses commissioned by the Department for Science, Innovation and Technology, based on fieldwork conducted between February and May 2025. Its finding for hotel and catering: 88 per cent of businesses neither use AI nor have any plans to do so. Not "not yet." Not "exploring it." Not using it, and not planning to.
of hotel and catering businesses have not started with AI and have no plans to. Not exploring it. Not planning it.
In February 2026 the UK government published a nationally representative study of 3,500 private-sector businesses; fieldwork February to May 2025, the most recent government data of its kind.
Adoption is accelerating and the gap is beginning to close.
That number is a snapshot, not a fixed state. More recent industry surveys, which apply a looser test of what counts as using AI, report engagement climbing sharply through 2026, with a meaningful share of operators now piloting or exploring the tools. The two readings are not in conflict. The government study measures a strict bar, neither using AI nor planning to, while the newer surveys count anyone experimenting at the edges. Together they tell a clear story: the standstill is ending.
All three worries are fair enough, but in most independent hospitality businesses they are either overblown or built on a misunderstanding of how these tools actually get used. So much of the conversation about AI has been run by tech vendors selling complicated platforms and consultants who have never had to run a service on a Saturday night. We are neither of those, and we are not going to talk to you like they do.
There are no special commands to learn, no coding, no technical qualifications you need first. The one real barrier to getting something useful out of an AI tool is the same one you hit briefing a new member of staff: being clear about what you actually want. If you can write a WhatsApp message, you can use AI.
Think of it as a brilliant apprentice. Incredibly well-read, fast, available at any hour, and never in a bad mood.
It has read more books, reports, menus, marketing campaigns and business plans than any person alive. Give it a clear brief and it will hand you useful first drafts and analysis frighteningly fast, but you still have to use your own judgement, check it, and sign it off before it goes anywhere. Give it a vague brief and you get a vague result. Same as with anyone.
The apprentice does not take the initiative. It will not spot that the rota is wrong or that the new chef has not been shown the allergens. It needs telling. But once you have told it, it moves at a pace no person can match. The jobs that used to eat an hour of a manager's day (writing a staff briefing, replying to a bad review, putting together a week's social posts, summarising last week's sales) take minutes.
Not a strategy. Not a rollout plan. One task that takes your manager an hour today and ought to take ten minutes. A weekly report. A reply to a bad review. A staff briefing for the new menu. You do it once with AI, look at what comes back, and go from there. The 12 areas in this paper give you the map of where the opportunities are; the free Snapshot tells you which ones to start with in your own business. Most operators who try it get their first useful result inside an afternoon.
ChatGPT, Claude, Canva AI, Copilot: these are different tools built for different jobs, the same way a Robot Coupe and a sharp knife both earn their place in a kitchen but you would not use one for the other. You do not need to understand how any of them work under the bonnet. You just need to know which one to reach for. A good part of what the review does is match the right tool to the right problem in your business, so you are not left guessing.
The quality of the output is a direct function of the quality of the brief, which is a skill you can learn quickly.
These 12 areas are the whole of the review. Between them they cover the things that quietly go wrong in almost every independent business we look at.
Be honest about how enquiries get answered in your place. They come in everywhere at once, the phone, email, Instagram, Facebook, the website form, and they get picked up by whoever happens to be free. Some get a great answer. Some get a rushed one. Some never get answered at all. And a customer who waits half a day for a reply on a private hire enquiry has already rung the place down the road. Get every message into one place, answered to the same standard, and none of them slip through.
Here is the part most operators miss. The reply matters, but so does what the enquiries tell you. Once they all run through one tracked process, you can finally see where people drop off: which channels go quiet, how many enquiries never turn into a booking, where the money is walking out before you ever knew it was there. Most of us have never had that view, so we have no real idea how much business we are losing before it reaches the diary.
A review with no reply says nobody is home. Picture someone deciding where to book, and the last thing they see is three one-star reviews sitting there with no response. They will go elsewhere. Not because the reviews were fair, half the time they are not, but because the place clearly is not paying attention.
The problem is never that operators do not care. It is that replying to every one, properly, takes time you do not have on a busy week. AI trained on how you actually sound can draft a response in seconds. You read it, tweak it, post it, thirty seconds of your time. Every review gets a reply while it still matters, and over a few weeks you start to see the patterns in what people keep telling you.
Let us be fair here. Plenty of operators are genuinely good on social now. They know their brand, they take a decent photo, they can write a post that sounds like them. The gap is almost never skill or awareness. It is consistency. The posting holds up in the quiet weeks and falls apart in the busy ones, there is rarely a calendar to work to, and hardly anyone is checking what actually landed.
AI will not replace a good photo or a proper offer, and it will not turn a weak idea into a strong one. Do not let anyone tell you otherwise. What it does take away is the blank page and the guessing about timing. An hour with it gives you a month's content calendar, every social post, email and promo mapped out. Feed it your events, your seasonal menu changes, the offers you have coming, and it hands back a plan with draft copy for each one. You edit it, you approve it, it goes out. Regular, on-brand, and holding up in the busy weeks as well as the quiet ones.
We all know how training really goes. Hospitality churns through people faster than almost any other industry, and the training rarely outlasts the person who did it. A new starter gets shown the ropes by whoever happens to be on that day. The standard moves with the mood. The good stuff lives in a few people's heads until those people leave, and then it walks out the door with them. So the quality a guest gets comes down to who is on shift.
This is one area where AI genuinely changes what is possible. You can now build training around each person, meet them where their knowledge actually is, and take them up from there. Everything a team member needs, the product knowledge, the standards, the way you do things, can sit in their hand on their phone, in short Duolingo-style bites instead of a binder nobody opens. Every new starter gets the same grounding, you can see who is where, and you stop being the only person who knows how it all works. The standard stays in the business even when the people move on.
No-shows cost you real money every single week, and everyone in this industry knows the feeling. A table of four that never turns up on a Saturday night is the drinks, the desserts and the covers you could have resold if you had known in time.
Here is the frustrating bit. Most modern booking platforms already have confirmation and reminder systems built in, and most operators have never switched them on. A guest who gets a confirmation, a reminder three days out, and a quick nudge on the morning is far less likely to simply not show.
Sat on top of your booking system, AI earns its keep in a few ways. It can pick up booking and enquiry messages wherever they actually land, phone, email, social, the website, and reply fast and consistently, so fewer go cold while your team is heads-down on the floor. When a cancellation comes in, an AI-assisted waitlist can offer that table to the next guest straight away, so a late drop-out gets resold instead of lost. And by reading your own booking history it can tell you which reservations look like no-show risks and which nights are about to get hammered, so you plan ahead rather than firefight. A person still owns the guest. The tools just do the chasing and the pattern-spotting around them.
In most independent places the owner finds out how last week went on the following Tuesday, off a spreadsheet somebody had to build by hand. By the time you can see the problem, it has already cost you a week you are never getting back.
Live dashboards that pull your EPOS (the till), your booking system, the rota and the accounts into one view are not some luxury only the groups and chains get to have. The difference they make to a decision is immediate. AI-assisted reporting takes it a step further: it reads the raw numbers, writes you a weekly summary you can actually understand, and flags what has moved before anyone has gone looking. But the reporting that really changes a decision lives below the headline. Not just this week's GP, but GP dish by dish, so you can finally see which plates are carrying the menu and which ones are quietly bleeding you.
The gap between a team that upsells well and one that does not comes out in pounds per cover, plain as that. A well-trained member of staff who naturally offers the matching wine, the sharing board, the room upgrade, never pushy, just part of good service, brings in noticeably more per table than one who lets it slide.
Most operators know all this already. What almost none of them have is a way to make it happen the same way, every shift, across the whole team. That is where AI helps. It builds and keeps the training materials that set the standard, and it reads your EPOS data to show you which items and which people are actually driving the spend per head.
The item worth steering a table towards is the highest-GP one, not simply the most expensive.
The sharper move is to point that effort at margin per dish, not at the headline sale. The matching wine is the obvious one, but steer a whole menu this way and it behaves completely differently, for a change that costs you nothing to make.
On every cover. No supplier haggling recovers it.
Stock is where the margin leaks fastest and shows up latest, and it catches good operators out all the time. Your headline GP will tell you the kitchen is losing money. It will never tell you where. The loss happens dish by dish, in the gap between the recipe as you wrote it and the recipe as it actually lands on the plate. A main you specced at 180g of protein going out at 220g because nobody put it on the scales hands back a chunk of that dish's food cost on every single cover, and no amount of haggling with the supplier gets it back. The control starts one level below that headline number: proper recipe cards with real portion weights, a costed portion, and a plate that looks the same on a quiet Wednesday as it does on a full Saturday.
Three things move underneath all that, and most operators never see them coming. Prices creep, a few pence at a time, and on catchweight goods (meat and fish billed by what they actually weigh, not a fixed pack) the cost of the same dish shifts with every delivery. Yield moves when a supplier or a spec changes, so you get less usable weight off an ingredient and the profit on the dish drops with it, quietly. And wastage, the trim, the over-prep, the stuff that spoils, is real food cost that almost never makes it onto a report. AI does not fix any of this for you. What it does is the maths, the moment a price moves, and it puts the slip in front of you in days instead of at the year-end stocktake. It will draft the allergen matrix too, the one that goes out of date the second a new ingredient comes in unrecorded, but you check that one line by line yourself, because it is a safety document long before it is a cost one.
Count the systems you run. The average independent place has somewhere between three and eight: a till, a booking platform, a rota tool, an accounts package, a stock system, a payroll provider, and usually a couple more nobody quite remembers signing up for. And they barely talk to each other. So someone spends hours every week keying the same numbers from one into another, mistakes creep in, and there is nowhere a manager can stand and see the whole thing at once.
The good news is that joining them up is often less complex and less expensive than it looks. Plenty of the tools you already pay for can talk to each other, the connection has just never been switched on. Where they cannot, a bit of lightweight automation can build the bridge without any custom development. But be straight with yourself about the exceptions. An older system that is past its end of life often has nothing to switch on, and connecting it properly can mean moving to something current. That is a real project with real risk attached, not a flick of a switch. So it is worth knowing which of your systems sit in which camp before you assume the fix is going to be quick.
This one is about what your team is actually typing into tools like ChatGPT, and what happens to it after they hit send. Here is the truth of it: your staff are almost certainly already using AI (ChatGPT, Canva AI, Copilot, whatever they have found) whether you know about it or not. A lot of the time that is completely harmless. But some of the time it is customer details, financial figures or confidential bits of the business going into a third-party system, with nobody stopping to ask where that data ends up or what gets done with it.
The data protection risk here is real, not scaremongering. A one-page policy is a sensible place to start: which tools are allowed, what must never be shared, and who signs off before anyone brings in a new one. A proper review goes further than that. It looks at what customer or business data could end up in these systems, how each vendor's data-retention settings are set, and when the law says you need a data protection impact assessment. The UK data protection regulator publishes detailed guidance on AI, and in a business handling customer data it is worth letting that guidance shape your policy.
The real thing standing in the way of AI in most independent businesses is not budget and it is not technology. It is time. The one or two people who would have to make any of it happen are the same ones already buried in rotas, reports, supplier emails and the weekly figures, week after week. And those are exactly the jobs AI can do in minutes. The same people could be out on the floor, coaching their team, making the calls that actually move the business on.
This is the bit that gets overlooked. AI's best contribution in hospitality is not to the guest experience. It is to yours, as the person running the place. A manager who is not drowning in admin has the time to train people properly, catch problems before they blow up, and build the sort of team that stays and looks after your guests.
There is a second thing here, and it is resilience. At 38.7 per cent turnover a year, good people leave regularly, and in most independent places everything they knew was only ever in their head. Writing down how your business actually works is the dull, unglamorous job you have to do before you can automate any of it. There is no shortcut around it.
This is the foundation everything else sits on: the physical and cyber basics under the rest of the review. And it belongs in an AI readiness review for a simple reason. The same wave of AI that is helping operators is also handing the same tools to the people trying to break in. Phishing is one of the main ways an attacker gets into a hospitality business, and AI has made it far more dangerous, because they now use it to churn out convincing, targeted messages at scale. The old advice, spot the typo, spot the dodgy phrasing, does not work any more. The clumsiness is gone.
Our sector gets its payment card data targeted at a higher rate than almost any other, and ransomware and stolen logins are now two of the biggest threats hospitality faces. The physical side is often no better: CCTV nobody has looked at since the day it went in, access codes that never change when someone leaves, cash procedures written once years ago and never touched since. With people coming and going as often as they do, the door is nearly always ajar for opportunistic theft, and a business sitting on customer emails, booking histories and payroll is a far more tempting target than most operators think.
Every one of the other eleven areas in this review adds to your digital footprint: more connected systems, more cloud services, more data moving about. That is the right direction to be going. But a more connected business is a more exposed one if the basics are not in place first. And for most independent operators the bar here is not high or clever: a written access policy, a password manager, a quick honest chat about phishing with anyone who touches payments, and one check a year on which outside systems can get at your data.
A business running labour at 33 per cent of revenue rather than the 28 per cent industry target is losing around £35,000 a year to excess staffing costs. Not necessarily through overstaffing, but through inefficiency, slow rota management, high turnover, and the absence of the data to identify and fix the pattern.
At 38.7 per cent annual staff turnover, a venue with twelve team members replaces roughly five people a year. At £12,000 to £18,000 per hire, the true cost of replacing those five people is between £60,000 and £90,000, most of which is invisible because it never appears as a single line item on a P&L.
In a survey of over 300 customers, scheduling software provider RotaCloud found that managers spent an average of nearly five hours per week on rota management alone before switching to digital tools, dropping to under two hours after. Combined with review responses, social media posting, weekly reporting and routine email handling, total avoidable admin commonly runs at eight to twelve hours per week. At a manager's salary of £32,000 to £38,000 per year, that represents between £8,000 and £12,000 of management capacity being spent on tasks AI can handle in minutes.
None of this makes you behind. It makes you busy, which is a different thing.
The point of putting numbers to it is not to worry you; it is to show that the capacity to fix it is already sitting in your week, spent on work a tool can now do.
The businesses that act now are not necessarily the most technically capable. They are the ones who recognised the gap early, while moving first is still cheap. With most of the sector not yet adopting, the field is wide open: an early mover pays very little to get ahead and gains a lead that is easy to hold. That will not stay true. As these tools become the norm, catching up stops being a quiet advantage and becomes the price of keeping pace. The cost of starting only goes up from here.
This one is illustrative. It is not a single named client; it is the pattern we see again and again, drawn together into one composite so you can picture it. Picture the owner of a 60-cover restaurant and bar. She wasn't failing. Bookings were solid, GP was at 68%, the kitchen was strong. She was also working six days a week, building her P&L manually, handling private hire enquiries across three channels with no consistent process, and training a team on a menu that changed every season. Two months had gone by without her sitting down properly with her own numbers. Not because she didn't care. Because there was always something more urgent in front of her.
The review identified three areas where the gap between effort and outcome was widest.
Thank you for making it this far. We never set out to write the definitive guide to AI in hospitality. In truth, we don't think one exists. What we've tried to do is ask better questions, challenge a few assumptions, and create space for our industry to think out loud.
One thing feels clear. AI won't shape the future of hospitality on its own. People will. The choices we make, the standards we set, the experiences we refuse to compromise and the values we protect, that's what will define the next decade.
This paper isn't finished on the day it's published. If anything, that's the day it starts. We'll keep listening and testing, and change our minds when the evidence asks us to.
So tell us. What have we got right? What have we got wrong? What did we miss, and what would you never let anyone automate in your business? Whether you agree with every word or disagree with half of it, we genuinely want to hear from you. Because the future of hospitality shouldn't be written for us. It should be written with us.
Thank you for helping shape the conversation.
The Impossible Dream
Get in touch: lee@theimpossibledream.co.uk
10 minutes or less. Your score, instant and free. Answer 36 questions across 12 areas, then enter your email and we send you the full report showing exactly where you stand and where to focus first.
We are hospitality operators who have gone deep on AI.
theimpossibledream.co.uk© 2026 The Impossible Dream. All rights reserved. The content of this paper is intended as general guidance only and does not constitute professional advice. All pricing is indicative and subject to individual assessment. Your information is handled in accordance with our privacy policy.