You are the first data platform engineer at a 200-person company. The CFO's board pack is assembled in a spreadsheet from three CSV exports, the product team queries a read replica directly, and two teams have started separate warehouse trials. Your manager asks what you will do in the first quarter. Walk me through how you would answer.

What the interviewer is testing

Whether you start from technology or from the decisions the business is already making badly. The trap is a tool answer: warehouse, ingestion product, transformation framework, catalogue. It is a fine stack and it does not tell anyone what breaks first.

The clarifying questions that change the answer

• What decision is currently made on a number nobody trusts? The board pack is the obvious candidate, and whether it is wrong or merely manual leads to different work.
• Is the read replica the product's replica? If a product query can slow the customer path, that is an availability risk sitting in your scope on day one.
• What is the regulatory position on customer data — any obligation to delete, any restriction on where data lands? This decides the storage design before anything else does.
• Who owns the two warehouse trials, and is either already paid for? Consolidating a procurement decision is a political task with a technical component, not the reverse.

A strong answer's arc

1. Stop the bleeding. Move the product's ad-hoc queries off the replica the customer path depends on. It is a week of work and it removes the failure that would otherwise define your first quarter.
2. Pick one number and make it reproducible end to end. The board pack. Ingest its three sources, model it, publish it, and let the CFO's team reconcile against the spreadsheet for a month. One trusted number in production teaches the organisation more than a platform diagram.
3. Choose the stack by the second use case, not the first. After the board pack you know the real update patterns, volumes and access requirements. Then consolidate the two trials into one, publicly, with the reasoning written down.
4. Set the contract early: who owns a dataset, what an SLA on it means, how a schema change is communicated. It is cheap now and unaffordable after twelve teams are on it.

Common weak answers

• "Build the lakehouse." No consumer, no constraint, no evidence. Six months later there is a platform and the board pack is still a spreadsheet.
• "Interview every stakeholder and produce a roadmap." A quarter of listening produces a document. One shipped, reconciled number produces credibility, and credibility is what funds the roadmap.
• "Standardise on the tools I know." Sometimes right, but it has to be argued from the workload, and here nobody has measured the workload yet.

What a strong answer adds

The organisational layer: a platform's constraint at this size is attention, not compute. One engineer can support roughly one well-run ingestion path and one modelling layer. Say what you will not do — no self-service catalogue, no streaming, no reverse ETL — and say what evidence would make you change that. Naming the second-order cost of each tool you decline is the part that reads as senior.