Starbucks says its service times are on target and its stores feel warmer. Now it wants AI to personalise the app and the drive-thru. CEO Brian Niccol said this week that after two years of its Back to Starbucks plan the company is “back on its front foot”. It has finished more than 1,700 store uplifts aimed at warmth and comfort, average service times are on target, and it has 36 million 90-day active Rewards members. For 2027 it plans a menu that recommends based on what you usually order, AI search and discovery in the app, delivery inside the app from the second quarter, and a test of AI at the drive-thru to help baristas take orders more accurately. It is also exploring digital menus that recognise a customer on arrival.
The claim: personalisation and AI are the next lever
The headline reads like one more big brand reaching for AI. What I noticed is the order of events.
Two years ago Starbucks had slow service and stores people did not want to sit in. It did not reach for AI then. It fixed the queue, the furniture and the barista’s working day first. Only once service times were on target and the stores felt like somewhere to stay did it move AI onto the app and the drive-thru.
Ask the question I always ask: is there a need? At the drive-thru, yes. Taking an order accurately while a queue builds is repeatable, time-bound work. A system that hears the order and gets it right every time solves a real problem for the barista and for the car behind. In the app, a menu that remembers your order removes taps. Neither of these needs a person. Both were already working well enough that AI can improve them instead of covering for something broken.
What did not change: the person at the counter, the coffee, the store you sit in. Starbucks spent two years and 1,700 refits on that layer, and it is keeping people there.
That is the order I would use anywhere
My view is that customer service teams in their current form will stop existing. A competent AI customer service automation can already handle most enquiries, I would say 80 to 90%. Order status, booking changes, refunds within a limit, delivery questions: this is transactional work and it no longer needs people, even when the workflow has several steps.
That is the easy half of the argument. The other half is that people should go where a human experience makes the difference. Starbucks is a clean example of both halves in the right order. The drive-thru and the app are transactional surfaces. The seat by the window and the barista who knows your name are the human layer.
Take a bakery that sells online and in a shop. Most of the messages it gets are the same twenty questions: is this available, can I collect on Saturday, where is my order, can I change the delivery address. For one bakery merchant, AeroChat handles 87 to 94% of conversations with no person, through festive peaks. The owner’s time goes where the machine cannot go: the wedding cake consultation, the corporate order that needs a tasting, the regular who comes in on Saturday mornings.
A clinic has the same shape. Booking, rescheduling, reminders, deposits and pre-visit questions can run on AI inside WhatsApp. The consultation itself, and the front-desk person who remembers a nervous patient, are the parts a competitor cannot copy with software.
Fix the transactional layer before you personalise it
This is where many operators get the order wrong. They buy an AI tool to personalise or to sound friendlier while the basic promise is still broken: replies take a day, the order status is wrong, the booking system double-books. AI on top of a broken process makes the broken process faster. Starbucks did not personalise a slow drive-thru. It made the drive-thru fast, then asked how AI could make it accurate and personal.
The pattern I would copy:
- List every customer interaction and sort it into two piles: repeatable and transactional, or human and differentiating. Be honest. Pile one is bigger than most owners expect.
- Fix pile one until it works without AI: accurate order data, clear limits on refunds and promises, one place the answers live. Then automate it fully. The rules about what the automation can commit to belong in the system, not in the prompt.
- Decide, on purpose, where a person shows up, and make that visibly better than before. That is where your service becomes something a customer talks about.
- Read the data from the automated conversations. It tells you which human moments customers actually want, instead of guessing.
“We only use humans” is a dumb position. Human-only service does not scale, and a competitor running the transactional layer on AI will have more people free for the moments that matter. The question left for an operator is which surfaces run on AI and where the people go.
Starbucks put AI on the app and the drive-thru and kept people in the store. Work out the equivalent split for your business, fix the transactional side first, and only then spend on personalisation.
Source: CX Dive: Starbucks puts greater personalization on the menu for 2027