McDonald's and AI: lay the foundation first
Originally written in Chinese. Translated with AI, reviewed by Gary. Read the Chinese original →
Today I came across one company's foresight and strategy, and honestly, it stunned me a little.
The story starts five years ago. Before the pandemic, McDonald's replaced the ordering boards in its restaurants with LED screens. Lots of people in the industry talked about it at the time, and the mainstream reading was the same everywhere: cut costs, digital upgrade, faster ordering. Back then I also thought it was a standard big-company process-optimisation case.
Only when I looked back recently did I realise I had completely underestimated that step.
It was never just "changing screens". It was the starting point of a long-term strategy to bring AI into 40,000 restaurants.
Lay the foundation first
McDonald's is the world's largest restaurant chain: 40,000 stores serving 70 million customers a day. The problem is that it isn't a company with "one unified system". Every country, every region, even every store has a different data structure. Run AI across an organisation like that? Almost impossible.
So they didn't rush to "use AI". They first did the hardest and least sexy thing: laying the foundation.
In 2019, McDonald's invested in AI company Dynamic Yield.
In 2020, it formally launched its Accelerating the Arches strategy.
That wasn't a product plan. It was an organisation-level decision: make digital the growth engine of the future.
For the next three years, McDonald's built almost no "cool AI applications".
They were doing something the outside world could barely see: unifying systems, integrating data, rebuilding the underlying architecture.
Only in 2023 did they fully move their global data to Google Cloud.
And only then could AI truly "run".
AI actually enters the store
AI really arrived in the stores after 2023.
The scenes you see now are the result of those three years of hard, unglamorous work.
In the kitchen, AI monitors orders and food preparation in real time. Cameras automatically check whether items are missing, wrong or short. When staff hit an equipment problem, they don't look for a supervisor; they ask the AI, which talks them through it step by step.
Based on past orders, time of day, weather and local habits, the system plans the day's rhythm of best-sellers for the kitchen in advance. What to prepare when, and how much, is all worked out.
The result?
More consistent service, less waste, and actually less stress for staff. It isn't about pushing people to run faster, but letting the system think for them.
It's already running in 400 restaurants, with the goal of rolling it out to stores worldwide by 2027.
More importantly, the back end is connected to the supply chain at the same time.
What's selling, how much has sold, how much to restock: the feedback is almost real time.
The changes at the front end are even more worth studying for founders.
Many people think an LED menu is just an "electronic menu board". But at McDonald's, the screen isn't static.
When you order at the drive-thru, the menu you see is generated in real time.
On a rainy day, the system leans towards recommending hot drinks. When there's inventory pressure, it gently puts certain items in front of you. When there's a sales target, it nudges you naturally through combos.
Not a hard sell, not an ad bombardment, but "guidance that follows your behaviour".
The app is the same.
It remembers what you eat, when you eat, and whether you respond to deals. You've just claimed a breakfast voucher, and it's already warming you up for lunch.
This isn't "we have an AI feature". It's as if every store has an intelligent agent that understands its customers.
McDonald's AI-native thinking
McDonald's didn't put AI into one department, or build an AI app to show the boss.
They wove AI into every action in the organisation: ordering, food prep, inventory, supply chain, loyalty. All the same logic, the same data chain.
AI isn't a tool. It's part of the process.
We love to say big companies are clumsy, slow and can't turn.
But McDonald's started on this back in 2019.
Google is the same. Only with the recent Gemini release did many people realise: never underestimate a big company that's truly patient.
They're slow not because they don't get it, but because once they move, they're paving the road for the next ten years.
For founders, this is a bit uncomfortable to look at.
We're used to chasing the next wave, chasing features, chasing "can we ship faster". But what really decides long-term competitiveness is often the choices that look most boring, most tiring and least applauded at the time.
Laying foundations is never sexy, but without them, any AI is just a performance.