Graduate Marketing Scientist in London

Fospha is dedicated to building the world's most powerful measurement solution for online retail. For over a decade, we've helped teams make smarter decisions with full-funnel marketing insights, forecasting, and optimisation. With Fospha, every team moves faster and grows smarter.

We're looking for a Graduate Marketing Scientist to join Fospha's Marketing Science team in London. Fospha builds marketing measurement products for ecommerce brands - attribution, marketing mix modelling, incrementality testing, and brand impact measurement. Marketing Science owns the applied end of that: designing and delivering incrementality tests and MMM engagements for clients, and standing behind the numbers when a client challenges them.

Team: Marketing Science

Level: Graduate - Entry (Data Science Career Development Framework)

Location: London

Salary: max £35,000

What you'll do:

• Assemble and validate test data - geo-level spend and conversion series, checking pre-period parity between treatment and control, spotting the coverage gaps that invalidate a design before it launches.
• Support test design under review - market matching and control selection, power and minimum detectable effect sanity checks, and identifying contamination risks such as geo-targeting settings that don't behave the way the platform's documentation claims.
• Run analysis and read the results honestly - pre-treatment fit diagnostics, lift estimates with their intervals, and what a null result does and doesn't tell you.
• Qualify client data for MMM - spend coverage across channels, whether there's enough variation in spend to identify an effect at all, series length and granularity, collinearity between channels, and gaps that will bias the result.
• Assemble and validate model input datasets, and investigate the discrepancies that surface when you do.
• Support model runs and read the diagnostics - fit, residuals, convergence, and whether a channel's estimated contribution is plausible.
• Contribute to output-extension work under review - building on an existing MMM result, for example forecasting or budget scenario work derived from it.
• Compare results across methods - where MMM, incrementality, and platform-reported figures disagree, understanding why is the interesting part of the job.

Model trust and diagnostics:

• First and second line on client trust queries - investigating why a number changed, working in SQL against client data to isolate the cause.
• Distinguish a bug from a methodology change - attribution window changes, model recalibration, data feed gaps, and platform reporting shifts all look similar from the outside and have very different signatures underneath.
• Triage PSPs on model trust, resolve what you can, and escalate.
• Reconcile platform-reported figures against our measurement - why walled-garden ROAS disagrees with ours is the hardest recurring question in the business, and you'll be learning it from the inside.
• Log and tag incidents consistently, so recurring failure patterns become visible and can be automated away rather than repeatedly handled.

Client communication and enablement:

• Run templated explainer sessions under review, walking clients through how our measurement works.
• Draft documentation and presentations above the core explainer content, and feed recurring query themes back into the source material.
• Fact-check methodology claims in product marketing collateral before it goes out.

What we're looking for:

• Someone with a strong foundation in maths and stats with clear communication who is looking to grow their skillset.

Technical:

• Working proficiency in SQL - you can investigate a discrepancy yourself rather than asking someone else to pull the data.
• Python, or a demonstrated ability to pick it up quickly. Most of our analysis tooling sits there.
• Grounding in inferential statistics - hypothesis testing, uncertainty, statistical power, and what a null result means.
• Some exposure to experimental design - randomisation, control groups, confounding, and why a badly designed test is worse than no test.
• Strong AI fluency - you use AI tools to get moving on unfamiliar problems and plug gaps in your own knowledge, and you QA the output before you rely on it.

Communication:

• Clear, concise written communication - a large share of this job is explaining something technical to someone who isn't.
• Composure in client-facing conversation, including when the client is unhappy with a number.