How to write it
A model with no decision attached is a hobby
The single biggest upgrade available to a data science resume is connecting each piece of modelling work to something that happened as a result. Not the AUC — the decision. Who did something differently because of your model, and what did that produce?
"Built a churn model with 0.86 AUC" and "built the churn model that drove a campaign cutting voluntary churn 18%, worth $6.2M ARR" describe the same project. Only the second one tells a hiring manager that you understand what you're for. Keep the model metric if it's genuinely impressive, but never let it be the only number in the bullet.
Say what reached production
The industry's open secret is how many models never ship. So "deployed", "in production", "retrained weekly", and "still running" are among the highest-value words on a data science resume — they separate you from candidates whose best work lives in a notebook on a laptop.
Name the machinery where you can: Airflow, MLflow, a feature store, a batch job, an endpoint. It signals you can work with engineers rather than handing them a pickle file and hoping.
Experiment design is the most under-sold skill
Plenty of applicants can fit a model. Far fewer can design a trustworthy experiment, spot the sample-ratio mismatch, choose the right unit of randomisation, or explain why an observational result isn't causal. If you can, that belongs high on the page in specific language.
The negative results are the most persuasive of all, and almost nobody includes them. "Led the causal analysis that ended a $400k/year programme shown to have no measurable effect" is a bullet only an honest, rigorous analyst can write, and any good hiring manager knows it.
- A/B testing at scale, and the framework you built or improved
- Causal inference methods you've actually applied — diff-in-diff, IV, matching
- A decision that was reversed or stopped because of your analysis
- Statistical rigour: power analysis, multiple comparisons, variance reduction
Cut the Kaggle rank and the course list
Competition placings and MOOC certificates are how a resume signals that professional experience is thin. Once you have a real job doing this work, they compete with it for space and lose.
The exception is a genuinely elite result — a top-ten finish in a large competition is a credential. Everything else, including the eleven Coursera certificates, comes off in favour of one more line about what shipped.
Communication, shown rather than claimed
Every data science posting asks for stakeholder communication, and every resume claims it in the same dead phrase. Replace the claim with an artefact: the readout you run, the dashboard executives actually use, the recommendation that changed a roadmap.
"Ran the quarterly readout to the exec team; two roadmap changes came directly from the analysis" is evidence. "Excellent communication skills" is filler that a reviewer's eye slides over.