Your Phone Knows Your Daily Routine — AI Now Guesses What You're Doing
Location data alone can now sort people into six daily routine groups.
Phone location data is everywhere. Every time you open a map, order a coffee, or scroll social media, your phone quietly logs where you stopped and for how long. What that data has never told anyone is *why* you were there. A new paper from researchers in London introduces a model called ACE (Activity Chain Encoder) that fills in that gap. It combines where you stopped, what time it was, and how long you stayed, then uses a Transformer — the same type of AI engine behind chatbots — to spot the shape of your day. Crucially, it's "self-supervised," meaning no one had to hand-label thousands of trips as "commute" or "grocery run." The AI teaches itself from patterns.
Applied to real mobile app location data from London, the model identified six distinct weekday activity groups. Each group has its own rhythm — think early risers with long commutes versus late starters who stay close to home — and each lines up with a different kind of neighbourhood and different census demographics like age, income, and household type. The researchers call the result "activity-pattern personas." In plain terms: they can describe how a slice of a city lives its day, without ever asking anyone a single question.
So what? This kind of tool is genuinely useful. City planners could see which areas need better bus routes at 7am. Retailers could spot where evening foot traffic is quietly growing. Public health teams could target campaigns at the right neighbourhoods. Because the code is publicly available, other cities and companies can run the same pipeline on their own data tomorrow.
The catch is privacy, and it's a big one. Location data is among the most sensitive information a phone produces. Even when researchers strip out names, movement patterns are famously easy to re-identify — most people have one distinctive place they visit. The paper works with aggregated groups, not individuals, but the underlying raw data exists, and the tool for interpreting it is now free for anyone to use. Whether that's used to plan better buses or to profile you is a policy choice, not a technical one.
- A new AI model called ACE reads phone location data and infers what you were doing, not just where you went — no human labelling required.
- Using real London data, it grouped people into six weekday routine types, each tied to specific neighbourhoods, ages, and income levels.
- The full code is public, so any city or company can run the same analysis on its own location data — useful for planning, risky for privacy.
Why It Matters
Cities and companies can now profile how whole neighbourhoods live their days from location data alone.