The Sovereign Restaurant — Blog Series - Post 6: The Sovereign Choice

Six weeks ago I posted two lines on LinkedIn and left them there without explanation.

A few hundred people engaged with it. A few dozen sent messages asking what I meant. The thesis that followed — The Sovereign Restaurant — has been the most substantive thing I have published in this industry, and the response has confirmed something I suspected when I started writing it: the operators and investors who are thinking clearly about the 2030 horizon already know something is wrong. They just haven't had a framework for naming it yet.

So let me end this series with the argument I consider most important. Not the most technically interesting, not the most immediately actionable, but the one that I think will look most consequential in retrospect.

The sovereign choice is not about technology. It is about which side of a structural divide you end up on.

On one side are the operators who will, over the next 24 months, gradually migrate their operational intelligence into platforms that are offering AI analytics as a free or near-free inclusion. They will do this because it is the path of least resistance. The integrations will be clean. The dashboards will be good. The IT overhead will be lower than building an alternative. Each individual decision will look rational.

At the aggregate level, those operators will have permanently transferred their most durable competitive asset to vendors whose commercial interests are structurally misaligned with theirs. They will have analytics. They will not have sovereignty. And when the platforms move — as they will — to monetize the corpus they have built from operator data, those businesses will discover they funded someone else's intelligence layer with their own operational history.

On the other side are the operators who make a different decision. Who treat their data as an asset rather than an overhead. Who build — or procure, or contractually protect — the inference layer that sits on their own data and stays theirs. Who arrive at 2030 with a proprietary corpus that cannot be replicated retroactively, no matter how sophisticated the platform's offering becomes.

I am not pretending this is a costless choice. Building proprietary data infrastructure is harder than signing a SaaS contract. It requires investment, internal capability, and the organisational patience to treat a long-term asset as a genuine priority before its value is obvious.

What I am saying is that the window to make this choice is open right now, and it will not stay open indefinitely. The platform announcements are coming. Once the default becomes seamless enough, the friction of the alternative will increase and the commercial case will become harder to make to a board that wants simplicity.

The delivery aggregator decision looked obvious in 2005. One contract at a time, the industry built a dependency it is still paying for twenty years later.

The data sovereignty decision is being made now. In every contract renewal. Every platform onboarding. Every time someone clicks agree without reading the terms.

None of what I've written in this series is certain. The agentic discovery timeline could slip. The data monetization thesis could prove harder to execute than the logic suggests. I have tried to be honest about where the argument is speculative and where it is grounded.

What I am not uncertain about is this: the operators who treat their data as a liability will arrive at 2030 having subsidised someone else's intelligence layer. The ones who treat it as an asset will have something worth owning.

That is the sovereign choice. It is available right now.

This is the final post in the Sovereign Restaurant series. The full white paper — covering all four pillars in detail — is available [here]. If this thesis is relevant to your business and you want to discuss it further, my details are below.

— MB

The Sovereign Restaurant — Blog Series - Post 5: The Macro Lab

There is a number sitting inside your operation right now that your purchasing team does not have.

It is not in your ERP. It is not in your supplier portal. It is not in any market report that lands in your inbox on a Friday morning. It exists only in the aggregated pattern of your own operational history — in the delivery variances, the waste events, the procurement shifts, and the menu mix changes that your sites have been recording, mostly without reading, for years.

That number is a forward indicator for your own cost base.

Here is the logic. A 50-site operator running at scale is conducting thousands of small economic transactions every week across the full food supply chain. Each of those transactions is a data point. A delivery that came in 3% above the invoice estimate. A waste event on a protein line that has happened three Thursdays running. A purchasing decision that was made at a price point that, in retrospect, preceded a supplier price increase by six weeks.

Individually these are operational footnotes. Aggregated across sites, reconciled against your own historical patterns, and read against the broader procurement cycle, they are something more useful: they are your operation telling you, in advance, where your input costs are heading.

Most operators are not reading this signal. They are receiving their cost base as a monthly surprise — discovering in the P&L review that beef moved, or that a key supplier has repriced, or that a category they thought was stable has started behaving differently. The response is reactive. The margin impact has already happened before the conversation starts.

The practical application of what I'm calling the Macro Lab is not complicated. It does not require a data science team or a proprietary algorithm. It requires a decision to structure and read the operational data you are already generating — to build the internal intelligence layer that converts your procurement history into a forward-looking view of your own cost exposure.

Done well, this lets you time your purchasing decisions rather than react to them. It lets you adjust your menu pricing ahead of the squeeze rather than in response to it. It lets you move from a posture of reactive cost management to something that resembles a considered position on your own supply chain — knowing when to lock in and when to let contracts run.

I want to be clear about the limits of this argument, because I think clarity matters more than a clean narrative.

The version of this thesis that involves selling your operational data corpus to institutional investors or FMCG partners as a commercial signal product is a genuine long-term possibility. It is also a near-term distraction that requires legal architecture, GDPR compliance, and a sales function most operators do not have and should not build right now. That is a 2029 conversation.

The 2026 conversation is simpler: use your own data to run a smarter operation. Anticipate your cost base. Make better purchasing decisions. Stop being surprised by your own P&L.

If it becomes a revenue stream later, that is the upside you never had to promise anyone.

The bin data is not waste. It is a dataset. The question is whether you are treating it like one.

Part five of the Sovereign Restaurant series. Full white paper available [here].

The Sovereign Restaurant — Blog Series - Post 4: The 70/30 Rule

I want you to imagine the Tuesday morning of a good Area Manager. Not a bad one — a genuinely capable, experienced operator who knows their sites, knows their people, and has good instincts about what's working and what isn't.

They are at their desk by 7:30. By 9am they have pulled last week's labour report for six sites, identified three variances they need to investigate, sent a message to the two GMs who haven't submitted their weekly summaries, started reconciling this Tuesday's opening numbers against last Tuesday's, and begun building the slide they need for Thursday's ops review.

They have not yet spoken to a single human being about the actual state of their business.

By 11am they are in a call reconciling a food cost discrepancy that turns out to be a data entry error. By 2pm they are in a car driving to the site with the worst variance, which they could have identified at 8am if the data had surfaced it automatically rather than requiring two hours of manual retrieval to locate.

The coaching conversation they need to have with the GM of that site — the one about the kitchen culture that is technically hitting its numbers but quietly burning through people — happens at 4:30pm, forty-five minutes before they need to leave, when both parties are tired and the conversation gets half the attention it deserves.

This is not a failure of the Area Manager. It is the structural reality of managing multi-site operations in an industry that has never seriously invested in its information architecture. The data exists. It is in the POS, the scheduling tool, the supplier invoices, the waste logs. But it exists in formats that require a human to retrieve, reconcile, and repackage before it can be used.

That reconciliation work — the 70% — is consuming the majority of a highly experienced, highly paid person's working week. And it is the least valuable thing they do.

The 30% is the work that actually compounds. The coaching conversation. The culture read. The judgment call about the junior manager who is ready for more responsibility before anyone else has noticed. The honest conversation with a site that is performing adequately on paper and deteriorating in reality.

That work requires presence, relational authority, and the kind of contextual intelligence that no inference layer will replicate. It is also the work that retains teams, builds culture, and creates the operational resilience that shows up in performance over years rather than weeks.

The argument is simple: automate the retrieval and you get the 30% back.

But I want to be direct about what that actually means organisationally, because the clean version of this argument leaves something important out.

When you remove the 70%, you find out quickly who has the commercial intuition and who was using the retrieval work as a shield. The spreadsheets are not just inefficiency. For a meaningful number of Area Managers, they are a way of being visibly busy that defers the harder, more exposed work of actual performance leadership.

This is not a technology project with a change management workstream attached. It is, in part, an organisational reckoning. Some people in the Area Manager role will make the transition from report-runner to commercial leader. Some will not. That has HR implications that need to be planned for honestly before the system goes in, not managed reactively after it does.

The 30% is recoverable. Getting there is not painless.

Part four of the Sovereign Restaurant series. Full white paper available [here].

The Sovereign Restaurant — Blog Series - Post 3: Auditioning for the Agent

There is a couple somewhere in London right now trying to decide where to go for dinner on Friday night.

One of them opens an AI assistant — it might be Gemini, it might be a voice interface on their phone, it might be something that didn't exist six months ago — and says: "Find somewhere good for Friday night. Italian or Mediterranean, somewhere quiet enough to have a proper conversation, under £80 a head, doesn't need to be booked weeks in advance."

The assistant does not open a browser. It does not scroll Instagram. It does not read reviews the way a human would, weighing the witty one-star against the suspiciously perfect five. It queries structured data. Menu information. Real-time availability. Pricing signals. Consistency across platforms. It evaluates the candidates against the brief and returns a recommendation.

The couple largely accepts it.

Your restaurant is either in that recommendation or it isn't. And whether it's in there has almost nothing to do with your photography, your Google Business Profile copy, or the campaign you ran in February.

It depends on whether your data is machine-readable.

This is the shift that most operators aren't structurally prepared for. The entire marketing function of the hospitality industry has been built around optimising for human attention — the moment when a person scans search results, looks at images, reads a headline, and makes a choice. That moment is still the majority of discovery in 2026. But its share is declining, and the trajectory is unambiguous.

I want to be honest about the timing, because intellectual honesty matters more than a sharp narrative. AI-mediated discovery is currently strong at transactional decisions — the parameters above, where the brief is clear and the answer is structural. It is genuinely poor at contextual judgment. The restaurant that is technically correct but emotionally wrong still gets filtered out by a human, not an algorithm. That gap will close, but it will take longer than the optimists suggest.

What this means practically is that the near-term case for machine-readability is not "prepare for the AI agent." It is simpler and more immediate: clean, structured, consistent data improves your performance in human-mediated discovery right now.

Your Google snippet quality improves when your menu data is properly structured. Your booking conversion improves when your availability is accurate in real time. Your cross-platform trust improves when your pricing is consistent. The operators doing this work for future AI readiness will get the human-experience dividend immediately.

Machine-readability is the new curb appeal. The question is not whether to invest in it. The question is whether you are doing it before it becomes the standard — or after, when the early movers have a twelve-month corpus advantage and you are catching up.

Most operators, if they audited their data footprint today, would find noise. Inconsistent menu information across platforms. Availability that updates slowly or not at all. Pricing that contradicts between the website and the aggregator listing.

That is fixable. It is also not being fixed, because the industry's marketing function is still optimising for a discovery paradigm that is declining in relevance.

The audition is already happening. The question is whether you are in the room.

Part three of the Sovereign Restaurant series. Full white paper available [here].

The Sovereign Restaurant — Blog Series - Post 2: The Cloud Trap

Sometime in the next twelve months, your inbox will receive an announcement.

A platform you already pay for — your POS provider, your workforce management tool, your ERP — will launch an AI analytics product. The email will be well-designed. The language will be confident. The headline will say something about unlocking the power of your data.

The price will be included in your existing contract, or close enough to feel like a no-brainer.

And if you sign up without reading the data terms carefully, you will have made the same mistake the industry made twenty years ago with delivery.

Cast your mind back to 2004. The choice operators faced then was whether to build delivery infrastructure — own the fleet, manage the logistics, control the channel — or outsource it to an aggregator that would handle the complexity in exchange for a percentage of every order.

Almost nobody built the fleet. The aggregator fee seemed manageable. The operational simplicity was compelling. And one contract at a time, the industry permanently transferred a revenue layer to platforms that now have structural pricing power over the operators who created their customer base.

The data sovereignty question is that decision, one layer up — and the downside is materially worse.

Here is what happens when you hand your operational data to a platform to train its models. The insight that data produces belongs to the platform, not to you. They will benchmark your performance against the anonymised aggregate of every other operator on their system. They will productize those benchmarks. They will sell the resulting intelligence to your competitors, your suppliers, and eventually to the investors evaluating your business.

You will receive a dashboard. It will be good. You will not own anything.

But here is the version of this argument that tends to land hardest in a boardroom: the platform that owns your best-selling item data, your peak trading windows, your highest-margin menu combinations, and your customer frequency patterns has a complete operational brief. It has everything it needs to design a ghost kitchen that competes directly with you — using your own historical data as the product development document.

That is not a paranoid scenario. It is the logical commercial endpoint of a platform whose relationship with operators is already deteriorating over margin, and whose cost base is structurally lower than yours.

The sovereign position — owning the inference layer, keeping the thinking inside your own architecture — is not a technology prescription. It is a contractual and commercial decision. It means ensuring, in writing, that the data your operation produces is yours: that it cannot be used to train models you don't control, benchmark you against competitors, or productize insights you generated.

Whether that means a private cloud with explicit data sovereignty clauses, an edge computing architecture, or something else will depend on your scale and your IT capability.

The principle does not depend on either.

Own the model. Or be owned by the platform.

Part two of the Sovereign Restaurant series. Full white paper available [here].

The Sovereign Restaurant — Blog Series - Post 1: The Death of the Price Taker

Picture a multi-site operator in a board meeting sometime in late 2024. Costs are up. Menu prices are at the ceiling of what the market will bear. The aggregator is taking 30% off the top of every delivered order. The CFO is asking why EBITDA keeps compressing despite record covers.

The answer nobody gives — because nobody has the language for it yet — is this: the business model itself is the problem.

The operator in that room is a price taker. Always has been. The industry has been built around the assumption that the job is to execute brilliantly within market conditions you cannot influence. Source well, schedule tight, reduce waste, sweat the asset. Survive.

That model worked when the primary competitive variable was product and location. It is failing now because the cost structure has been permanently reset in ways that operational discipline alone cannot absorb.

The April 2026 National Insurance changes, the £12.71 National Living Wage floor, and the structural 30% aggregator tax on delivered revenue are not a bad cycle. They are the new floor. Operators waiting for conditions to normalise are waiting for something that is not coming back.

But the more interesting problem — the one that will define which businesses are still standing in 2030 — is not the cost reset. It is the data problem.

Every multi-site restaurant operation running at scale is generating thousands of data points every week. Transaction data. Procurement data. Waste data. Labour data. Delivery variance data. Behavioural data from every customer interaction the business has ever logged.

Most of that data is being treated as a byproduct. An administrative overhead. Something that gets reported on weekly and acted on monthly, if at all.

Here is the reframe: that data is not a byproduct. It is the most valuable asset on the balance sheet. More valuable than the lease. More durable than the brand. And almost no operator in the industry is treating it that way.

The shift from Price Taker to Signal Provider is not a technology decision. It is a strategic one. It requires a deliberate choice to treat the operational data layer as an asset class — to invest in its collection, its structure, and its interpretation with the same seriousness applied to property, equipment, or people.

The operators who make that choice now will have two to three years of structured, inference-ready data before the platforms catch up. The ones who wait will have a vendor's dashboard and someone else's margin.

The price taker era is not ending because the market got kinder. It is ending because the information layer finally got cheap enough to exploit.

The only question is who does the exploiting.

This is the first in a six-part series expanding on The Sovereign Restaurant thesis. The full white paper is available [here].

Five Things I'm Watching After Easter (And One I'm Trying Not To Think About)

There's a particular kind of quiet that settles over the hospitality industry in the days before a bank holiday weekend. Operators are heads down, covers are up, and nobody has time to read anything longer than a rota. So I'll save this for Tuesday, when the dust has settled, the egg hunt is a distant memory, and everyone is staring at their Q2 targets wondering what just happened.

Here are five things I'm watching closely going into the second quarter of 2026 — and one I'll admit I'd rather not think about at all.

Five Things I'm Watching After Easter (And One I'm Trying Not To Think About) - Michael Baglieri

1. Labour costs hitting harder than operators expected

Let's be honest about the numbers, because they deserve to be said clearly.

In April 2025, employer National Insurance Contributions rose to 15% and the threshold at which employers start paying was lowered — a double hit that landed hard across multi-site operations. What most operators modelled as a one-off adjustment has now become the new baseline. And as of this week — 1st April 2026 — the National Living Wage has increased again, to £12.71 per hour.

That is not a rounding error. For a hospitality business running 300 or even 600 hours of labour per week, per site, this is a material shift in the cost structure. Annualised across a portfolio, it is the kind of number that changes conversations with investors, banks, and boards.

What I'm watching isn't the cost itself — it's the response. There are broadly two types of operators right now: those who are making structural decisions about how they deploy labour, and those who are quietly trimming hours and hoping nobody notices until the next period closes. The first group will be in a stronger position by Q3. The second group will be having a more difficult conversation before the year is out.

The operators who treat this as a design problem — how do we build a labour model that genuinely works at these rates — will come out ahead. The ones who treat it as a line item to manage will keep having the same conversation every quarter.

2. The delivery margin squeeze reaching a tipping point

I've been saying for a while that the economics of third-party delivery are broken for a significant number of operators. Not all of them — there are businesses for whom the channel genuinely works. But for too many, it's a revenue line that flatters the top of the P&L and quietly destroys the bottom.

Something has to give. Either the commission model changes — and there are early signs that some platforms are starting to feel competitive pressure — or operators start making harder decisions about which channels they actually want to be in. The brands that have built genuine direct relationships with their customers are going to have options. The ones that outsourced that relationship entirely are going to find themselves with very little leverage.

This isn't an argument against delivery. It's an argument for knowing your numbers and making a deliberate choice — rather than being on every platform because it feels like you should be.

3. AI moving from pilot to P&L

The operators who ran AI trials in 2025 are now being asked to show the return. Some of them can. Most of them can't — at least not in a way that satisfies a CFO or a board.

The ones who are struggling tend to share a common characteristic: they deployed AI against the wrong problems. Customer-facing tools, marketing copy, chatbots. Visible, easy to demo, difficult to quantify. The ones who are showing genuine returns built around operational decision-making — labour, waste, throughput, yield. Boring problems with measurable outcomes.

I've written about this before and I'll keep writing about it, because I think it's the most important distinction in the space right now. The question isn't whether AI works in restaurants. It does. The question is whether you've pointed it at a problem that matters to your business model.

4. US brands arriving with high expectations and low local knowledge

This is the one I find genuinely exciting — and the one that makes me a little nervous at the same time.

The scale of American QSR investment coming into the UK in 2026 is significant. To name just a few of the brands either arriving or accelerating:

• Jersey Mike's Subs — planning 400 locations across the UK and Ireland. That is an extraordinarily ambitious target.

• Raising Cane's — making its long-anticipated UK debut this year, bringing its cult following and laser-focused menu with it.

• Chick-fil-A — accelerating its UK expansion following its Leeds opening, with more sites in the pipeline.

• Chuck E. Cheese — also confirmed for a 2026 UK entry, targeting a family dining market that has been underserved for years.

These are serious brands with serious capital behind them. The best American QSR operators know things about throughput, consistency, and unit economics that the UK market is still learning. I have genuine respect for what several of these businesses have built.

But — and this is a significant but — the UK is not America with smaller portions and worse weather. The planning system is different. The labour market is different. Consumer expectations around value, service, and brand authenticity are different. The cultural assumptions that work in Texas or Tennessee don't always translate cleanly to Manchester or Edinburgh.

The brands that invest in genuinely understanding that — before they start signing leases — will do very well here. The ones that treat the UK as a copy-paste of their domestic model will relearn some expensive lessons. I say this with warmth, not scepticism. I've watched it happen enough times on both sides to know it's not a criticism. It's just a pattern.

5. Data becoming the asset nobody has valued yet

This is the one I find most interesting, and the one I'm spending the most time thinking about professionally.

Restaurant groups are sitting on behavioural data — purchasing patterns, visit frequency, menu preferences, daypart distribution, channel behaviour — that has genuine commercial value beyond the four walls of their own operation. Most of them don't know it. Some of them are starting to suspect it. Very few have any kind of framework for how to think about it, let alone monetise it responsibly.

That's changing. Not overnight, and not without some interesting questions around data governance and consumer trust. But the operators who get ahead of this — who start treating their data as an asset on the balance sheet rather than a byproduct of their POS — are going to find themselves in a very different conversation with investors and partners in three to five years.

I'll be writing more about this specifically in the coming weeks. It's where most of my advisory work is heading, and frankly it's the most intellectually interesting problem in the sector right now.

The one I'm trying not to think about this Easter?

Egg pricing. Which? found a Galaxy Extra Large Easter egg went from £4.98 for 252g last year to £5.97 for 210g this year — a 44% increase per gram, achieved by making the product smaller and the packaging identical. In restaurants we call that portion distortion. In confectionery they call it innovation. Whoever handles Cadbury's PR deserves a bonus.

And possibly a Wispa.

Have a good break. Back properly next week.

Michael Baglieri is a restaurant consultant and Managing Director (EMEA) with 20 years of experience across the USA, Europe, GCC, and Asia-Pacific. He writes about restaurant strategy, data, and the future of the operator at michaelbaglieri.co.uk

The Shift I Nearly Missed.

Twenty years ago, I was running a busy lunch service in New York when my general manager handed me a printout. Labour was 4% over. I told him I already knew — I'd felt it on the floor.

That instinct, built over thousands of shifts, is what most operators rely on. And for a long time, it was enough.

It isn't anymore.

Not because operators have lost their edge — but because the business has grown faster than any individual's ability to track it. More sites, more dayparts, more data points than any human brain can hold simultaneously. The instinct is still there. The information it needs to work with has outgrown the format.

I spent years thinking technology was the enemy of that instinct. That dashboards and data tools were for people who couldn't read a room. I was wrong.

The operators I've worked with who are pulling ahead right now aren't the ones who've abandoned their gut feel. They're the ones who've given it better information to work with.

That's what good technology does in a restaurant. It doesn't replace the operator. It brief the operator — faster, more accurately, and across more variables than any morning meeting ever could.

The shift I nearly missed wasn't on the floor that day in New York. It was in how I was thinking about the relationship between experience and data.

Twenty years later, I'm still learning to read the room. I've just got better tools to help me do it.

What restaurant operators actually need from AI — and why most tools miss the point entirely

The AI tools being built for restaurants are mostly built by people who've never run a shift. They're designed for tech companies, repackaged for hospitality, and sold to operators who are too busy to notice the difference until they've wasted six months and a significant budget.

Here's what I've learned after 20 years operating and consulting in restaurants across four continents: operators don't need AI that's clever. They need AI that's useful. There's a meaningful difference.

What operators actually need:

Labour percentage is the number that tells you everything. If your labour is wrong, nothing else matters. What operators need is a tool that looks at their rota, their covers, their revenue — and tells them in plain English where the problem is and what to do about it. Not a dashboard. Not a report. An answer.

Weekly reporting consumes hours that Area Managers and GMs don't have. The average multi-site operator spends 3-4 hours per week producing reports that nobody reads in full. A well-designed AI prompt can do that in 60 seconds — and produce something sharper than the manual version.

Site performance comparison across a portfolio is where the real insight lives. Not averages — outliers. Which site is quietly underperforming? Which one has a labour problem hiding behind strong top-line revenue? AI is genuinely good at finding patterns humans miss in data they're too close to.

What most tools get wrong:

They're built for scale before they're built for usefulness. A tool that works for a 500-unit chain doesn't work for a 12-site operator — and the 12-site operator is where most of the real need is.

They require clean data. Restaurant data is never clean. Any tool that can't handle messy, inconsistent, multi-format data isn't a restaurant tool. It's a proof of concept.

They solve the wrong problem. Operators aren't asking for AI that predicts consumer trends or optimises their marketing funnel. They're asking for something that gives them their time back. Start there.

What I'm building:

I'm developing a set of practical AI tools specifically for restaurant operators — starting with a weekly ops report generator and a labour diagnostic tool. Both are designed to work with the data operators actually have, not the data they wish they had. Join the waitlist if you want early access.