About the role
The platform rests on a claim: what store teams know shows up in what stores sell. The intelligence layer exists to test that claim, house by house, and to make it useful. It joins content and quiz data from the platform to sales data from the maisons, at POS, market and country level. This role owns the models inside it.
The work runs from correlation to prescription. First come honest answers about what moves with what, at which lag, in which markets, with the confidence stated plainly. Then come the models that suggest which content to push to which stores, and the evidence that the suggestions were worth taking. You build them with the engineers in Shenzhen, who ship them into the product.
What you will do
- Join data of very different shapes and tempers: platform events, quiz results, POS feeds, market and country data.
- Build the correlation work that gives each maison an honest read on the knowledge-performance link.
- Design the prescription models that suggest what content to push where, and measure whether they earn their keep.
- Ship with the Shenzhen engineers; your models run in the product, not in a notebook.
- Stand behind the numbers with the account managers when the quarterly reviews come round.
What we look for
- A strong statistical foundation. You know what a correlation can and cannot say, and you say it.
- Years of applied work on commercial data, ideally retail or POS data with all its gaps and quirks.
- Fluent Python and solid SQL, and the habit of shipping models with engineers instead of handing them over a wall.
- The judgement to tell a client the data does not show what they hoped, and the tact to make that useful.
- Clear writing. Your findings travel between London, Shenzhen and client HQs, and they have to survive the trip.
Practicalities
Full-time, based in our London office, with three days a week there as the default. The working language is English. Start as soon as you can.
How to apply
Write to careers@tolduntold.com with 'Data Scientist' in the subject line. Send whatever shows your work best: an analysis you stand by, a repository, a short note on a model that survived contact with production. We read every message and reply to all.