You are asked to 'use ML to improve the user experience on our platform.' How do you approach this completely open-ended problem?
Open-ended ML problems require scoping before modelling: translate the vague ask into a measurable business objective, identify which user interaction has the highest improvement potential, formulate it as a concrete ML task with a defined label and evaluation metric, then propose the simplest viable model first. Jumping to model architecture before this scoping is the most common interview failure mode.
How to think about it
The five-step scoping framework
Step 1 — Define the business objective
Restate the vague ask as a measurable outcome: “improve user experience” could mean reduce churn, increase session depth, increase revenue per user, or reduce support tickets. Ask clarifying questions: “What does a 10 % improvement look like for this team? What metric are they held to?”
Step 2 — Map user interactions to impact
Draw out the main user journeys and identify the highest-friction or highest-dropout steps. Use data: which funnel step has the biggest drop-off? Which user segment churns fastest? This localises where ML can have the most impact.
Step 3 — Formulate as a concrete ML task
Pick the single user interaction identified in step 2 and ask:
- Prediction task: what are we predicting? (e.g., will this user churn in the next 30 days?)
- Label: how is the target defined from logs? (churn = no login in 30 days)
- Features: what signals are available at prediction time?
- Evaluation metric: AUC-ROC, precision@k, NDCG? Chosen based on how the model output will be used.
Step 4 — Propose the simplest viable model first
A logistic regression or gradient-boosted tree is the starting point. State why: interpretable, fast to iterate, strong baseline. Neural networks come later only if there is a clear advantage (sequence modelling, image/text features) and sufficient data.
Step 5 — Address operationalisation
How does the model output translate into a product action? A churn-risk score feeds a re-engagement campaign — the threshold for triggering the campaign is a business decision (step back to the cost-matrix framework). Define how success is measured post-deployment.
Worked example — e-commerce platform. Vague ask: “improve UX.” After scoping: the checkout funnel has a 68 % cart-abandonment rate, highest among mobile users. ML task: predict whether a user will abandon within the current session (binary classification, label derived from session logs). Features: cart value, session duration, device type, prior purchase history. Model: XGBoost, evaluated on precision@top-10 %. Product action: trigger a discount pop-up for high-risk users. Guardrail: margin impact of discounts must not exceed revenue recovered.