Dubai Harbour has nine restaurants open today, and six of them already send their daily sales to eGoal. If their operators agree, they become FoodSight’s first live outlets, with nothing to install. That gives Marriott the live client it asked for, and opens the door to Addmind’s and Sunset’s other venues.
And if you can, one more door: Emaar, Paris Society, ADMO or Kerzner. Groups four and five are the slow part, and an introduction there does the most.
Outlets marked eGoal already send their daily sales to eGoal for Shamal.
Ushuaïa Dubai Harbour Experience is temporarily closed.
SUSHISAMBA Dubai at The Palm Tower is Shamal’s brand but Sunset runs it, so it counts with Sunset. Zero Gravity has closed at its old site and is moving.
A benchmark opens at five groups, with none above 40 percent of it. Dubai Harbour has two groups today, and Addmind runs seven of its nine restaurants. So on its own it is our first live client, not yet a benchmark. For the first, Dubai-wide benchmark, Shamal’s Five Guys and The Espresso Lab make three. Groups four and five come from The Anchor’s new operators and from Emaar, Paris Society, ADMO, Kerzner or Marriott. No one ever sees another operator’s figures.
NDA signed on 22 October; the agreement is in its last round. Connects about 100 outlets a month, and already runs Dubai Harbour for Shamal.
Met the Business Council’s F&B lead on 23 October. Interested. A pilot needs a live client and IT security clearance first.
Yes to its venues, with its Marriott-run hotels to follow Marriott. Next step after 5 November.
Warm after a call on 22 October about its casual dining brands. Next step in the week of 9 November.
Your early idea of restaurant categories and tiers is how we now plan to open the benchmark. We start with location, not with named rivals: a guest picks an area first, then a place in it.
All live outlets together. Opens first.
The coast from Dubai Harbour to Bluewaters, Jumeirah, DIFC. Each opens at five groups.
By average check and format, inside each place.
| Place | Restaurants open | Groups | Biggest group’s share |
|---|---|---|---|
| Dubai Harbour on its own | 9 | 2 | 78% |
| The coast: Dubai Harbour, JBR, Dubai Marina, Bluewaters | 193 | 32 | 29% |
Prepared for Sudhin Siva, Shamal Holding, for Sunday 25 October 2026. Not for circulation.
The model plays the next three years out a thousand times, group by group. The event that matters is the first benchmark opening: five groups live, none above 40 percent of the set. Once it opens, every operator in that lane becomes far more likely to join, and volume follows. The plan starts with the outlets eGoal already connects at Dubai Harbour, and gets to the benchmark without anyone working unpaid.
Each dot is one strategy: how likely the premium benchmark is open by October 2028, against what FoodSight keeps over three years. Up and to the right is better.
Outlets live: median, with the middle 80 percent of runs beneath. Benchmark: the chance the premium comparison set has opened. Unpaid: outlet-months eGoal connects and runs without being paid. Own pocket: the chance FoodSight has to top up eGoal’s fee from its own money.
Pick a basket and the model reruns the market at nine prices, everything else as set in the simulator. The search below tries every pricing rule for every basket and keeps whatever does best on the goal you choose, using the same random draws for each candidate so the comparison is fair.
Run the curve to see the response. The dot marks the price in the simulator now.
Only so many accounts can be worked at once. The order decides who gets that time, and early signatures pull others in. Test the six patterns against each other, or move groups up and down and watch the simulator.
Outlets open in Dubai on the outlet research of 14 October 2026, each counted under the company that runs its till. Change a count, switch a group off, and the simulator updates.
eGoal connects the tills, runs the platform and builds the benchmark engine. Under the terms on the table it is paid a fixed fee on every outlet that has paid that month. These figures follow the strategy loaded in the simulator.
eGoal connects your tills and delivery apps every day. FoodSight turns them into one view: trends, targets and how you compare. Connect it to Microsoft Copilot, Claude, Gemini or Grok, add your own data there, and it looks ahead, suggests what to do and, with your OK, does it.
Your POS feed builds the base in FoodSight. The AI connection takes it into your own Claude or Copilot, where your private data joins it. Each level answers a harder question and rests on the one below.
Your Claude or CopilotWhere the market meets your own data
Done, with your OK.
Acts in your rota, POS and buying systems, inside limits you set.
What should we do?
Rota, prep, buying and offers, each with its reason.
What will happen?
Demand, covers and staffing for the days and weeks ahead.
Financials, payroll, time and attendance, reviews and more.
Stays with you. Never shared or pooled: FoodSight never sees it, and your files never travel to us.
Microsoft Copilot, Claude, Gemini, Grok and others. Structured questions only: outlet, place, period.
FoodSightRules and benchmark by FoodSight, data run by eGoal
Where do we stand?
Trends, targets and comparisons on one screen. Filters for place and period.
How do we compare?
Your outlet against its own lane, by place. Blended, never named, a week behind.
What happened?
Your POS feed, cleaned and tagged with lane and place. eGoal connects every till, every day.
Oracle Micros, Foodics, Talabat, Deliveroo and more, every day. Builds your view and the blended market; never shown by name.
We train the people: hands-on sessions for owners, managers and teams, in their own Copilot or Claude.
Blended and lagged, for landlords, brands and agencies, once sets open.
NoweGoal builds it first, from 26 OctoberNextonce the data has some historyLateragentic priced later; reports once sets open
Illustrative: a fictional quick-service outlet with made-up numbers. Every figure follows the rules further down this page.
AED 128,400
Last week, up 6.2% on the week before
Best week of the eight. Same week last year: up 11%.
79% of AED 560,000
AED 440,200 so far, with 7 days to go
Reached: 79%Month gone: 77%
On pace. Needs AED 17,100 a day; running at AED 18,300.
Sales last week
Checks last week
Average check August, on July
The market is a blend of 7 groups, none above 40%, never named. Sales and checks are a week behind; average check is monthly and 45 days behind.
How did my Jumeirah outlets do last week against Jumeirah quick service?
Sales up 6.2% against 3.9% for Jumeirah quick service, a blend of 7 groups. Checks drove it: up 6.0% against 2.8%. From your payroll file, labour was 23% of sales, down from 25%.
The market figures came through the FoodSight connection, a week behind. The payroll figure came from your own file and never left your Copilot.
Friday covers forecast up 18% on a normal Friday.
Why: a public holiday, and a concert nearby that evening.
From your own history plus public signals: the holiday calendar and event listings. No other operator’s bookings or forecasts.
Add one cashier from 7 to 10 pm, and prep 20% more fries.
Why: the Friday forecast, and how your own busy Fridays ran.
Advice from your own data and public signals only. No competitor’s prices.
Rota change drafted in your rota system. Approve?
Friday: one extra cashier, 7 to 10 pm, from your standby list.
Nothing changes until you approve. Every action is logged with who approved it.
The base of the stack, with the Dubai Harbour feed as the first live data. The aim: real figures in the one view and through the AI connection, ready to show a hotel IT team.
One row per outlet per day, from each outlet’s POS and delivery platforms:
outletlaneplacedaysalescoverschecksitemsPlus an entitlement table that fixes each user’s lane, read by the screen and the AI connection alike. Every load is checked, and a partial or unusual one is held back.
Three panels, as above, with two filters (place and period):
Connect to Microsoft Copilot, Claude, Gemini, Grok and others. The tools already running in the demo:
outlet_pulseday_vs_normaldaily_salesdaily_itemsmenuservicereviewstradingdemand_signalsstaffing_demandStructured questions only (outlet, place, period), checked against the lane lock on the server. Code computes, the model phrases.
Good practice under competition law for any benchmark that pools competitors’ figures. The white paper tab sets out why.
Your own data stays in your own Copilot or Claude: FoodSight never receives it. And no personal or guest data, ever.
The end game. An exit to a buyer like STR: every rule above is what that buyer’s diligence will test. See the five exit scenarios in the White paper tab.
The UK is testing whether STR’s hotel benchmark crossed the line, and the answer will set the rules for every benchmark, ours included. FoodSight can build a benchmark that sits safely inside today’s rules; the risk is in the layers we want to add on top.
What is happening.
What it means for us.
This is a research briefing, not legal advice. UAE-qualified competition counsel should confirm it before we rely on it.
STR is the hotel industry’s scorekeeper. Regulators care because a scorekeeper holds every rival’s numbers in one place.
The business. STR started in 1985 as Smith Travel Research in Hendersonville, Tennessee. CoStar Group bought it for USD 450 million in cash, closing on 24 October 2019, when STR held data from 66,000 hotels with 8.9 million rooms in 180 countries (Lodging).
How it works. Hotels send STR their occupancy, average daily rate (ADR) and revenue. STR sends each hotel a report comparing it with a “competitive set” of rivals the hotel picks, blended so no single rival shows. The set rules (CoStar):
STR already runs the same model for hotel restaurants and function space, F&B STAR, with the shares measured on seats (2018 guidelines).
Why regulators care. Competition law calls this information exchange. When rivals know each other’s recent prices and occupancy, they can predict each other’s moves, so the pressure to undercut fades. The CMA is testing whether sharing through STR let the chains do that on pricing and on hotel development (Watson Farley & Williams).
Why it is not simply banned. Benchmarking helps operators find their own weak spots and decide where to open. The law draws the line on what is shared: how fresh, how blended, how many contributors, and whether it says anything about the future. Those four tests run through every case below.
The CMA opened a Competition Act investigation in February 2026 into whether Hilton, IHG and Marriott shared competitively sensitive information through STR. It has reached no view, and its own timetable runs to January 2027 before it decides whether to object.
The CMA is in its second phase. After January 2027 it closes the case, accepts commitments, or issues a statement of objections.
Who and what. Four parties are under investigation: CoStar (CoStar UK and CoStar Group), Hilton, IHG and Marriott (case page). The suspected breach is of the Chapter I prohibition of the Competition Act 1998, which bans agreements and concerted practices that restrict competition. The CMA announced the case on 2 March 2026 and gave no reason for opening it now (press release, Hospitality Investor).
What data is in question. The CMA has not said. UK lawyers list occupancy, ADR, RevPAR, room supply and future demand data (Knights). Others report a focus on near real-time data and on how STR’s tools turn it into reports that inform pricing (Watson Farley & Williams).
Where it stands. Information gathering ran from February to August 2026. On 5 August 2026 the CMA extended the timetable: further evidence gathering, analysis and review, to an estimated January 2027. CoStar said it was surprised, is cooperating, and expected the case to take about six months; that window has already passed (The Real Deal).
What can happen next.
A breach can bring fines of up to 10% of worldwide turnover, director disqualification for up to 15 years, and follow-on damages claims (Knights).
Why it matters beyond the UK. No other regulator is reported to be examining STR (The Real Deal). The US private suits against STR have failed so far (next section), so the CMA case is now the main live test of the STR model. Whatever STR changes to settle it will likely become the norm for every benchmark, including ours.
In the US, benchmarks with no prices have survived, and live shared data feeding a pricing tool has not. The DOJ’s two 2025 and 2026 settlements turn that line into numbers.
| Date | Case | What it was about | Result |
|---|---|---|---|
| 29 Jul 2026 | Cornish-Adebiyi v. Caesars, Third Circuit | Atlantic City casino hotels fed live, non-public rates and occupancy into Cendyn’s Rainmaker and took its advice about 90% of the time | Dismissal reversed; the case returns to the trial court |
| 7 May 2026 | US v. Agri Stats, DOJ and six states | Benchmark reports for meat processors; the DOJ said recipients could tell who submitted what | Proposed settlement: data at least 45 days old (90 for production plans), at least three contributors, none above 70%, no sales or price reports, no rankings or flags, reports open to any buyer, a monitor |
| 25 Nov 2025 | US v. RealPage, DOJ | Rent software pooled landlords’ non-public data to recommend rents | Proposed settlement: no live use of one user’s data in another’s advice; training only on data at least 12 months old; price and local data largely barred; auto-accept off by default |
| 29 Aug 2025 | Portillo v. CoStar, W.D. Washington | STR benchmarking, including Forward STAR’s 365 days of occupancy on the books (Asian Hospitality) | Dismissed with leave to amend: no actual room prices went in or came out |
| Aug 2025 | Segal (luxury hotels), N.D. Illinois | Luxury hotels shared data through Amadeus Demand360 | Dismissed with prejudice: the hotels received only aggregate occupancy data |
| 15 Aug 2025 | Gibson v. Cendyn, Ninth Circuit | Las Vegas Strip hotels licensed the same revenue software | Dismissal affirmed: no agreement, no confidential data exchange alleged, hotels free to ignore the advice |
What the pattern says. The STR suit failed on a narrow point: no prices were exchanged. Cornish-Adebiyi shows the risk moves to the tool that recommends prices from pooled live data, even when each user keeps the final say.
The old safe harbour is gone. The DOJ and FTC withdrew their information-exchange safety zone in 2023: data at least 90 days old, at least five contributors, none above 25% (Arnold & Porter). Enforcers now judge case by case, so the settlements above are the best guide.
States are legislating. Since 1 January 2026, California bans using or distributing a “common pricing algorithm” (two or more users, fed with competitor data) as part of an agreement, or coercing users to adopt its prices (Cleary Gottlieb).
UK and EU law set no fixed safe harbour for benchmarking; they weigh the same eight tests. The CMA’s own benchmarking example names the fix: a third party that collects, anonymises and aggregates the data (CMA guidance, Example 2 of Part 8).
| Test | Lower risk | Higher risk | CMA guidance, August 2023 |
|---|---|---|---|
| What is shared | Costs, efficiency, results | Current prices, pricing plans, capacity, demand or sales forecasts | Para 8.84 lists these as past “by object” breaches |
| How old | Historic | Current, and above all future | Para 8.56: safely historic if several times older than the pricing cycle; the Commission has treated data over a year old as historic |
| How blended | Hard to trace to any one business | Individual figures, or blends that can be unpicked | Para 8.47 |
| How many, how big | Many contributors in a fragmented market | A few firms with a large share | Para 8.48 and Example 2 |
| How often | As rarely as the purpose allows | Frequent | Para 8.70 |
| Who runs it | An independent third party; each member sees its own data and the final blend | Direct swaps, or a hub that relays individual data | Paras 8.72 and 8.73 |
| Who gets access | Fair, open, non-discriminatory terms | A closed club that leaves rivals at a disadvantage | Paras 8.29 and 8.31 |
| Algorithms | A tool fed with public data and your own data | One pricing tool pooling rivals’ sensitive data | Paras 8.26 and 8.68 |
Current capacity is a red flag. The CMA singles out current or recent data on spare capacity or capacity use as particularly likely to reduce uncertainty (para 8.56). For hotels that is occupancy; for restaurants it is covers and bookings.
The EU says the same. The Commission’s 2023 Horizontal Guidelines call future price or quantity intentions “particularly likely” to be a breach by object, treat historic data as unlikely to raise concerns, and want data pools run by an independent third party (Herbert Smith Freehills Kramer).
Liability reaches the hub. A third party that runs the exchange can itself be liable, not only the businesses that use it (para 8.69). That is why CoStar sits in the CMA case beside the chains.
The UAE has a modern competition law with real fines, but no guidance on information exchange or benchmarking. So FoodSight should build to the UK and EU standard and get a written UAE opinion before any anchor signs, as our own rule already says.
The law. Federal Decree-Law No. 36 of 2023 was issued on 28 September 2023 and took effect on 29 December 2023, replacing the 2012 law (UAE Legislation). It covers every business active in the UAE and conduct abroad that affects competition here. The SME exemption and most sector exemptions are gone; government-owned entities stay exempt (Clyde & Co).
What it bans. Article 5 bans agreements whose purpose or effect is to distort or restrict competition: price fixing, direct or indirect, bid rigging, limiting output, sharing markets and blocking entry. There is no separate information-exchange rule. But “agreements” covers practices, written or verbal, explicit or implicit, so an exchange of sensitive data can fall under Article 5 (Chambers, Cartels 2026).
Algorithms are now named. Ministerial Decision No. 96 of 2026 bars coordination on prices, discounts, costs or margins “including through systems or algorithms”. Lawyers call it one of the first express UAE references to algorithmic coordination (Chambers, trends).
The Ministry is watching F&B. Ministerial Decision No. 32 of 2026 sets the terms for exclusive deals between delivery platforms and restaurants, and other decisions limit emergency price rises, including for food delivery (Chambers, trends).
Enforcement is starting. The Ministry of Economy’s Competition Department enforces. A new Executive Regulation, Cabinet Decision No. 59 of 2026, took effect at the end of July 2026: a two-stage investigation, and complaints from consumers and rivals. On 19 April 2026 the Ministry sent a cartel case on poultry prices to the Federal Public Prosecution; no fines are public yet (Chambers, trends).
Penalties. Fines run from AED 100,000 up to 10% of the business’s UAE sales in the last financial year, or AED 500,000 to AED 5 million where sales cannot be set. A court can also close the business for three to six months. There is no leniency scheme, and injured parties can sue in the civil courts (Chambers, Cartels 2026).
What it means for us.
The benchmark itself is defensible if it stays blended, historic and independent. The risk climbs up the stack, to wherever pooled data meets pricing, so the rules tighten as a layer gets closer to setting a price.
The three lower layers stay safe if the set rules hold. The connector and forecasts need technical guards, and anything that touches price needs the strictest rules.
Proposed defaults, for UAE counsel to confirm. No fixed rule exists in the UAE; each number below borrows the strictest relevant precedent.
| Layer | Rule | Where it comes from |
|---|---|---|
| Data layer (eGoal) | eGoal and FoodSight act as the independent third party. Each operator sees its own data and the final blends, nothing else. Collect only what the product needs. | CMA para 8.73 |
| Benchmark sets | Five groups minimum, none above 40% of the set. No change that isolates one group, checked against past reports. | Tighter than STR (four hotels, 50% and 70% caps) and Agri Stats (three, 70%) |
| Benchmark filters | Place is the only filter. No self-picked rival lists, participant lists, rankings or flags. | Agri Stats settlement |
| Market trends (sales, covers) | Blended indices only, weekly and at least seven days old. Never the current day. | Proposed; current capacity data is the CMA’s red flag (para 8.56) |
| Price and cost blends (average check, discounts, labour) | Monthly, at least 45 days old. | Agri Stats settlement |
| Forward-looking data | Never shared across operators: no bookings on the books, forecasts or planned prices. | CMA para 8.84; Forward STAR was a target of the STR suit |
| AI connector (Copilot, Claude, Gemini, Grok) | Entitlements and set rules enforced on the server for every tool call. The model never receives another operator’s rows. Queries logged and checked for isolation. | CMA para 8.73; STR’s isolation checks |
| Your own data | Financials, payroll, time and attendance and reviews join the blends inside the operator’s own Claude or Copilot. FoodSight never receives them, and no other operator sees them. The connector takes only structured questions: outlet, place, period. | CMA para 8.73: each member sees its own data and the final blends |
| Predictive | Each outlet’s forecast uses its own data plus public signals. Market models train only on blends at least 12 months old. | RealPage settlement |
| Prescriptive | Price and discount advice from own data and public prices only, and it must recommend cuts as readily as rises. | RealPage settlement; CMA para 8.68 |
| Agentic | Acts for one operator, inside ranges the owner sets. No auto-accept by default; every action logged with who approved it. Never deals on prices with another operator’s agent. | RealPage settlement; UAE Ministerial Decision No. 96 of 2026 |
| Market reports | Blended, monthly or slower, sold to anyone on the same terms. | Agri Stats open access; CMA para 8.31 |
| Events and immersions | No pricing talk between rivals; agenda set in advance; counsel briefed. | Cornish-Adebiyi (industry events) |
Lane lock helps commercially, not legally. Showing QSR operators only the QSR blend means each sees its closest rivals. Blends across lanes are safer (CMA para 8.47), so within a lane the set rules and the time lags carry the weight, and a Dubai-wide all-lanes view is the safest first level.
Words count as evidence. Cendyn’s slides about avoiding “price wars” were used against the hotels in Cornish-Adebiyi. Our pitch should talk about efficiency and each operator’s own decisions, never about the market holding prices.
Contracts must match the stack. The pilot NDA bars benchmark use and AI training without a later agreement. Predictive models and market reports therefore need the services agreement terms in place first.
Paper the compliance. A written rulebook, a UAE opinion, logs that prove no isolation and a yearly review are what a regulator, and a buyer, will ask to see.
For a buyer like STR, our competition file is part of what it buys. A clean rulebook adds value; a gap turns into an indemnity, an escrow or a walk-away.
Five ways out. None is realistic before the Dubai benchmark sets open, which the rollout model puts at around 2028.
| Scenario | Who | Why they would buy | What must be true first | Watch-out |
|---|---|---|---|---|
| 1. Benchmark owner | CoStar’s STR, or another hotel-data firm | Restaurants are the gap beside STR’s hotel F&B product, F&B STAR (2018 guidelines). CoStar paid USD 450 million for STR in 2019 (Lodging) | Dubai sets open, recurring revenue, a clean competition file | Under its own CMA case, it will test our rules hardest |
| 2. Restaurant data group | Black Box Intelligence, owned by Diversis Capital (QSR Magazine), or another investor-backed roll-up | A Gulf foothold for its “AI command centre for restaurants”; its April 2025 purchase, Yumpingo, already reaches the Middle East (The Caterer) | The AI connection in daily use in Copilot or Claude, and the benchmark live | It will want proof that the predictive and agent layers earn money |
| 3. Platform or POS company | A delivery app or point-of-sale vendor | The benchmark and the agent, sold across its merchants | As above, plus contributors’ consent to the new owner | Neutrality lost: operators may stop sharing, and regulators may look at foreclosure of rival platforms (CMA para 8.30) |
| 4. Gulf strategic buyer | A regional technology or hospitality group | A ready platform for Saudi Arabia and the rest of the GCC, hosted in the UAE | A second market live, and data rights that travel | Merger control in each country, including the UAE 40% share test |
| 5. Stay independent, then sell | A minority investor now, a buyer later | Funds the agent layer and a second market | Recurring revenue and open sets | Slower, but keeps the benchmark neutral and the price higher |
What sets the price. Recurring revenue, outlets live and share of each market, open benchmark sets, daily use of the AI connection, and a clean competition file. A neutral owner keeps contributors sharing; a conflicted one can empty the pool.
What their diligence will test.
Merger control can catch a small deal. A UAE filing is needed when the parties’ combined share of the relevant UAE market exceeds 40%, and the test has no size floor on its wording (White & Case). If the buyer already sells F&B benchmarks here, closing could wait for clearance: up to 90 days, extendable by 45 (King & Spalding).
Build to the strictest precedent now. Whatever the CMA settles with STR is likely to become the market standard. Meeting it from day one means nothing needs unwinding when we sell.
All pages opened on 24 October 2026.
The UK case
STR
The US cases
The rules on information exchange
UAE
Buyers
It is a simulation of the Dubai market month by month, from November 2026 to October 2029. Each run decides, group by group, who signs, when their outlets go live, who stays when a free or founding period ends and who leaves. Hundreds of runs give a range, not a single guess. Every strategy is run on the same random draws, so the differences between them come from the strategy, not from luck.
Every group in the outlet research of 14 October 2026 that runs outlets in Dubai, counted under the company that runs the till, in three lanes: premium dining (independent premium groups, owner-run hotel groups and hotel-brand restaurants), casual and fast casual, and quick service and coffee.
A lane’s comparison set opens when five different groups have outlets live in it and none holds more than 40 percent of that set’s live outlets. Before that, each group sees only its own numbers. Several outlets of one group count once.
Sources: FoodSight outlet universe research, 14 October 2026 (Dubai, open outlets); the price agreed on 14 October (USD 100 per outlet a month); the term sheet sent to eGoal on 18 October (AED 100 platform, AED 65 engine, paid per outlet that paid); eGoal’s capacity of about 100 outlets a month (19 October); the comparison rule of five groups, none above 40 percent; the Dubai Harbour outlets eGoal already connects (24 October); the Marriott Business Council meeting on live data and IT clearance (23 October). Updated 24 October 2026. Built for Lean On Data FZ-LLC. Not an offer; every figure is a model result.