Every few years, something shifts in how successful hospitality businesses are run - and the venues that spot it early do not just adapt, they pull so far ahead that the gap becomes almost impossible to close. The ones that miss it spend years catching up, often without fully understanding why the gap opened in the first place. That shift is happening right now, and it is moving faster than most operators realise.
Introduction
There is no shortage of opinion on AI in hospitality. Every supplier, every trade publication, every tech platform and conference panel has a take. Most of it is well-intentioned, occasionally interesting, but is ultimately just noise if you are trying to work out what to actually do on Monday morning. The practical result is that a lot of operators have already tried something, found the output didn't land in any useful way, and quietly stopped using it. The technology often gets the blame. The more honest read is that most of what's being sold today is too generic to do the job, and the distance between what operators are promised and what they actually get is wider than the marketing suggests.
What has been missing is a clear, honest diagnostic built on what operators are genuinely experiencing, rather than what the industry thinks they should be excited about. This is that diagnostic.
It is built on conversations with dozens of operators across the full range of hospitality - what they have tried, what has worked, where they have stalled, and what the venues pulling ahead are doing differently. It will not tell you that AI is the future. It will tell you where your venue stands right now, what is already happening above you, and what it takes to move.
There are three levels to how AI integrates into a hospitality operation. Most venues are at Level 1 or 2. Level 3 is where the operational gap between venues becomes visible and starts to compound. Read on to find out which level sounds like yours - and whether that should concern you.
1Level 1: You know what AI is, and you may or may not be using it
This is where most venues are, and there is no shame in that. Some of you have ChatGPT open in a browser tab right now. You might be using Claude to help draft a difficult email or pull a proposal together faster, perhaps Gemini for the odd image generation to jazz up your pitch deck.
When something comes up that feels like AI could help, you go to it, you ask, and more often than not it does something useful. It has quietly become part of the toolkit without anyone making a formal decision that it should be.
Others are still weighing it up. Not because they are resistant to change, but because the noise around AI has made it genuinely difficult to work out what is worth paying attention to and what is not. If that is where you are, you are in good company - and this report is as much for you as it is for the operator three tabs deep into their second ChatGPT conversation of the morning.
Wherever you are starting from, the picture at Level 1 is broadly the same: AI as a capable assistant for tasks you already knew you needed to do. Drafting communications. Generating ideas. Summarising information. Speeding up work that used to take longer. The output is useful and the time saving is real, but the AI has no meaningful knowledge of your business. Every conversation starts from scratch. It knows as much about your venue as a smart stranger who just walked in off the street.
It's also where a lot of operators have already had their first disappointment with AI. Jill Maddison at Cycle Hub was honest about it: the outputs from the generic tools she'd tried were, in her words, utter nonsense. Her experience is more common than the industry tends to admit. The cause is almost always the same: the tool she was using had no idea about her venue, her trading patterns, or her stock. When AI doesn't know the business, what it gives back is generic, and generic doesn't help operators run venues.
The more significant opportunity at Level 1, and the one that most venues are sitting on without realising it, is data interrogation. If your EPOS is connected, you can already ask it questions that would previously have required a report request, a spreadsheet, and half a morning. This matters more in hospitality than in most other industries. The people running hospitality businesses are doers and problem solvers. They know their venues inside out, they have years of pattern recognition built up, and most of them have never had the time or the formal training to pull that knowledge out of a dashboard.
Questions like: what were my top-selling items at last night's event? Which vendor underperformed this weekend versus last? What are my best and worst margin items over the last 90 days? At what point in the day does my average order value start to fall, and how sharply? No dashboards, no filters, no waiting for a report - just a plain English question and a straight answer. These are not complex analytical questions. They are the questions that good operators are asking anyway, usually slowly, manually, and less often than they should be. The difference now is that you can just ask.
Most venues are sitting on years of transactional data they have never properly interrogated, not because they are not curious, but because no one told them the barrier to entry had dropped this far.
Lauren Richards at Camden Roundhouse put it plainly: rather than having to ask the Dines team how to find something, being able to use an AI assistant to surface that information directly changes how quickly decisions get made.
That shift - from waiting for a report to just asking a question - is what makes Level 1 genuinely valuable when it is used well. Evidence-based decisions made faster, with less reliance on instinct and less dependency on someone else pulling the numbers.
But here is the thing about Level 1: it is reactive by nature. You still have to know what to ask. You still have to go to it. The AI is not thinking about your business when you are not in the room, because it does not know your business. It is a powerful tool that waits to be used, and tools that wait to be used only deliver value when someone picks them up.
If Level 1 is where your venue currently operates, you are in the majority. You are also at the point where a decision about whether to go further is starting to have real consequences either way.
2Level 2: You are feeding it context, and it is starting to have opinions
If Level 1 is asking AI to help you work faster, Level 2 is asking it to help you think better. The shift is subtle but the implications are significant, and it is the point at which AI stops being a productivity tool and starts becoming something closer to a business intelligence layer. Put simply, it's the shift from pulling information to receiving it.
At Level 2, you have moved past generic prompts and open-ended questions. You are bringing external context into the conversation - industry benchmarks, market reports, third-party research - and asking AI to make sense of it alongside your own operational data. You are cross-referencing your transaction data with your invoices, your staff costs, your event or show schedules, and asking what the combined picture tells you that any single data source on its own would not. You are using it to frame performance for stakeholders, to turn a set of numbers into a narrative that means something, to draft the kind of update that used to take hours and now takes 10 minutes.
This is where AI starts to have opinions about your business rather than just answering questions about the world. Feed it enough context and it will start identifying patterns you did not think to look for. A dip in average order value that correlates with a specific event format. A margin problem that only shows up when you look at staff costs and wet sales together. A stock variance that makes no sense until you cross-reference it with the show schedule. These are not insights that require a data analyst or a consultant. They require context, and at Level 2 you are starting to build that.
This is also where AI starts to reason across the whole business at once, holding your stock, your sales, your staff costs and your margins in the same answer and pulling a useful conclusion from across all of them. It is a capability you only get from AI that can see across the whole picture in one place, and a lot of AI being sold into hospitality today simply cannot.
The compounding effect of this is easy to underestimate. The more context your AI has access to, the more useful it becomes - and that relationship accelerates the more deliberately you feed it. Which naturally raises a question worth sitting with: what data does your AI not currently have access to that would make it meaningfully smarter? What is still living in a spreadsheet, a separate system, or someone's head that, if your AI could see it, would change what it is able to tell you?
The manual work is the thing that holds most venues back at this level.
Shauna McCarthy at Delfont Mackintosh Theatres described it well: the problem is not that people do not want the insight, it is that they are doing double the work to get it, maintaining a spreadsheet and then inputting the same information into their system separately.
The same logic applies to the tasks that quietly eat hours without anyone noticing. Updating prices across a large menu. Reconciling revenue across a busy weekend. Stock queries that used to require a request, a wait, and a spreadsheet. At Level 2, these become conversations rather than projects - and the time that comes back is time that goes back into running the venue.
Most venues who consider themselves ahead of the curve are operating at Level 2. And they are right to feel good about it - it represents a genuine step change from Level 1, and the operators who are here are making better decisions more quickly than the ones who are not. But it is worth being honest about what Level 2 still is: a smarter version of you asking better questions. The AI is more informed than it was at Level 1, but it is still waiting for you. You are still the one who has to notice that something is worth investigating, pull the relevant data together, and go to it with a prompt.
Level 2 is a meaningful step. But if Level 1 is working faster and Level 2 is thinking better, then Level 3 is something else entirely - and the venues already there are not waiting for you to catch up.
3Level 3: AI that runs the venue with you
This is where the gap becomes visible. Not in a few years when the technology finally catches up, and not at some vague point when the industry is ready. It is happening now, in venues that look a lot like yours. The operators at Level 3 are not doing more with AI than you are - they are doing something categorically different with it, and the distinction is worth understanding clearly.
At Level 1, AI helps you work faster. At Level 2, it helps you think better. At Level 3, it helps you actually run the venue. With your approval, the system does the work itself, because it carries the context of your business permanently and understands how hospitality workflows actually run.
Three shifts together make it possible.
The first is persistent business context. A Level 3 AI knows your venue and does not start each conversation cold. The things that shape your week, the things you would otherwise have to explain again every time you talked to a generic chatbot, are simply known. The fact that the London Marathon happens every April and your venue sits 200 metres from the route. The comedian's show you booked last August that sold out of pink gin within the hour. The fact that bad weather hurts you because most of your seating is outdoors. That your supplier delivers on Wednesdays. None of that lives in your EPOS, but all of it shapes what is about to happen in your venue this week. Most AI tools throw that context away at the end of every conversation, where a Level 3 AI carries it forward and grows sharper every time you use it.
The second is real understanding of how hospitality work actually runs. A Level 3 AI is not just following generic instructions when you ask it to do something, it understands the workflows themselves. When it reorders stock, it knows your supplier's minimum order quantities, your usual cadence, and what counts as a normal order versus an unusual one. When it re-prices a menu for a busy weekend, it knows how your prices have moved historically and what the margin needs to clear to be worth the change. When it reconciles your end of day, it knows where the discrepancies typically live and what to flag. When it communicates with a customer, a supplier or your team, it speaks in the voice your venue uses. As the understanding is built into the system, the AI carries it through every workflow it touches.
The third is authorised execution, and this is where the real labour saving lives. A Level 3 AI surfaces a proposed change, shows you the reasoning behind it, and waits for your sign-off. When you give it, the system carries the change through: the reorder placed, the prices updated, the reconciliation closed, the message sent. A human stays accountable for every decision, but the work of carrying it out moves off your desk. And if something does get actioned that should not have been, a rollback window gives you a defined period in which any AI-driven change can be reversed.
At Level 2, the AI gives you a better answer and you still have to act on it. At Level 3, the AI acts on its own answer, with your authorisation. That is the difference between being told you should reorder, and the reorder going in.
In practice, that looks like this: Your AI flags that consumption rates on a particular product line are running higher than the sales data would suggest, before you have noticed the variance yourself. It tells you that a regular purchase order has not been placed by its usual day. It connects a sunny weekend forecast to a historically reliable spike in a specific category and prompts you to act on stock before the weekend. Each of those moments has the three shifts behind it: the context already loaded, the workflow understood, and the action ready to go once you approve it.
Nick Crow at Bar Live described the ambition precisely: using your POS to guide behaviour rather than just record it, building a picture of what is about to happen rather than accounting for what already has, and then measuring performance against those expectations in real time.
This is also AI that works the way operators actually work. Not a desktop tool that requires you to sit down and engage with it deliberately, but something that lives on your phone, fits between services, and is there when you are on the floor making decisions rather than when you are finally at a desk with time to think. At Level 3, your AI is not waiting to be consulted. It is part of the operation.
And crucially, it is trusted. The operators at Level 3 have moved past treating AI output as a useful suggestion to be verified, towards treating it as a valid source of truth that informs real decisions on labour, stock and pricing. That trust is built through consistent, accurate, context-rich output over time, which is exactly why the venues that started building it earlier are now so difficult to close the gap on.
What Level 3 looks like for you will vary. For a food hall it means one coherent view across multiple traders, with the patterns that only emerge at the aggregate level surfaced automatically. For a theatre it means an AI that understands the pre-show window, what it historically demands of the bar, and what needs to happen before the doors open. For a festival operator it means site-wide volume management in real time, with the kind of predictive stock and labour intelligence that turns a chaotic peak into a managed one.
The principles are identical. The application is specific to your operation and your role.
Here is the thing that is easy to miss about Level 3: it is not coming, it is here now. The venues using it are not beta testers or early adopters in the traditional sense. They are operational, they are busy, and they are making better decisions faster than the venues that are not there yet. The only meaningful difference between them and you is not budget, not technical resource, and not venue size. It is that they made a decision, and then they started.
What holds venues back
If most of the industry is at Level 1 or 2, and Level 3 is already operational in venues right now, the obvious question is: if the technology exists, and the case for it is clear, and the venues using it are no different to mine, what is the gap actually made of?
It is almost never the technology. In years of conversations with operators across every corner of hospitality, the blockers that come up consistently are not technical ones. They are human ones, and they tend to fall into three categories.
The first is data that lives in disconnected systems. Transactions in one place, staff costs in another, invoices somewhere else entirely, and no single layer that can see across all of it at once. AI is only as intelligent as the information it has access to, and a fragmented data picture produces fragmented insight. The venues stuck here are not failing because their AI is not good enough. They are failing because the foundation it needs to work from has never been properly built.
The second is the pull of gut feel. Hospitality has always run on instinct, and for good reason - experienced operators have pattern recognition built up over years that is genuinely valuable. The problem is that instinct does not scale, does not document itself, and cannot be interrogated at 2am when the numbers from last night's event need to make sense. Teams that have always defaulted to feeling over evidence are not wrong to trust their experience. They are just leaving a significant amount of decision-making quality on the table, and in an industry where margins are wafer thin, that cost compounds quietly. The irony is that the manual work gut feel produces - the spreadsheets, the reconciliation, the end of week reports nobody has time to properly read - is exactly the work that AI at Level 2 and beyond quietly removes.
The third is trust. AI recommendations that arrive without visible logic are hard to act on, especially for operators who have not yet built a track record with the tool. This is the most understandable barrier of the three and also the most self-resolving - trust is built through experience, and experience only comes from starting. As covered in the Level 3 section, the right AI also makes that trust easier to build: it surfaces its reasoning, asks before it acts, and gives you a rollback window if it gets something wrong. The venues that are furthest ahead on trust got there simply because they began earlier than everyone else.
None of these are permanent problems. None of them require a wholesale transformation or a significant capital investment or a team of people who understand machine learning. They require a decision, followed by a series of deliberate and manageable steps in the right direction. The path from Level 1 to Level 3 is not a leap. It is a series of habits built gradually, each one making the next one easier.
But here is what is true about all three barriers: they get harder to close the longer they are left.
What it actually costs
The operators we speak to describe the same problem underneath everything else: there is no time to do the human bit. No time to spend with the customer who's just walked in, no time to perfect a service offering, no time to chase the feedback that builds repeat business. They're firefighting through admin, reconciliation, stock queries and supplier paperwork instead. And here's the part that matters commercially. When an operator is under that level of pressure, it shows on their face.
The customer feels it. It changes whether the customer comes back. So the gap we've been describing throughout this report has a second face. It's also the gap between venues where the operator still has time to be properly present with their customers, and the venues where they don't.
The diagnosis is the starting point. Where you go next is the question.
ConclusionWhat level are you at?
Before we introduce anything, it is worth being honest about why this report exists. We did not write it to add to the noise. We wrote it because we kept having the same conversation with operators across the industry - venues that knew AI mattered, were not sure where they stood, and had no clear way to work out what to do next.
The diagnostic in this report is built on those conversations. It is not a framework we constructed in a meeting room. It is a pattern we kept seeing in the field, with real operators, in real venues, until it became impossible to ignore.
Which brings us to the question the whole report has been building towards.
Cast your eye over the statements below and note how many you can honestly say apply to your operation today:
Level 1
- I use AI tools like ChatGPT, Claude, or Gemini for occasional tasks
- I use AI to draft communications, generate ideas, or speed up administrative work
- I have asked my EPOS or data system a direct question and received a useful answer
- My team is comfortable using AI as part of their working week
Level 2
- I regularly feed external data - benchmarks, reports, industry research - into AI to inform decisions
- I use AI to cross-reference operational data: transactions, staff costs, invoices, event schedules
- I use AI to shape how I communicate performance to stakeholders
- AI is starting to surface patterns in my business that I did not specifically go looking for
Level 3
- My AI carries the context of my venue permanently, without me having to set it up every conversation
- My AI understands the workflows of running a hospitality venue, not just generic instructions
- My AI carries out approved changes (reorders, repricings, communications), not just suggests them
- I trust my AI to act on decisions I approve, knowing changes can be rolled back if needed
- My AI works on a phone, mid-service, on the floor, not just at a desk
The level where you start ticking fewer boxes is your level.
If most of your ticks are in Level 1, you are in the majority. If you are solidly at Level 2, you are ahead of most - but the gap to Level 3 is where the competitive distance is being created right now. If Level 3 sounds like your operation, there is a conversation worth having about what comes next.
ConclusionBeyond Level 3
We have deliberately not published Levels 4 and 5 in this report. Not because we do not know what they look like - we do, and we are already working towards them with a small number of operators. What we can say is that they move into territory that most venues are not yet ready to have a conversation about: AI that does not just surface insight but acts on it autonomously, scheduled and automated reporting that removes the human trigger entirely, and agentic AI that manages workflows end to end without being prompted. The venues in those conversations with us right now are not larger or better resourced than yours. They are just further along the map. If you are at Level 3 and want to know what comes next, this is the moment to say so.
ConclusionTilly
Everything in this report informed how we built Tilly - Dines' integrated AI assistant built for hospitality operators.
Not the other way around. We did not build a product and then write a diagnostic to support it. We spent years in conversation with operators about what they actually needed, what was getting in the way, and what a genuinely useful intelligence layer for a hospitality venue would have to do differently from the generic tools already available. In February 2026, we hosted the Futures Forum: 14 of the most interesting and complex hospitality operators in the UK, brought together across workshops and interviews to shape what we were building. The principles in this report, and the ones that sit underneath Tilly, came directly out of those rooms.
Tilly is the answer to those conversations. It is already live. Operators are already using it. You do not need to know how to build a report or navigate a dashboard - you just need to know what you want to find out. Ask it in plain English and it will tell you. And it is built specifically for the industry this report is written for.
Find out more at dines.co.uk/platform/tilly.
ConclusionStay in the conversation
The operators who will have the advantage are not the ones who found this interesting. They are the ones who read it, knew where they stood, and did something about it.
If you are at Level 1 or 2 and want to understand what moving looks like for a venue like yours, speak to us.
If you are at Level 3 and want to know what comes next, that conversation is already happening with a small number of operators. If you want to be in it, now is the time to say so. Whether you run a theatre, a food hall, a festival or a multi-site bar group, we will tailor the conversation to your operation.