The stages, and what each one is really testing

The standard sequence

Most data and AI hiring processes are some subset of seven stages. Knowing what each is for stops you giving a round-four answer in round one.

1. Recruiter screen: 20 to 30 minutes. Tests three administrative facts: can you legally work there, are you in the salary band, and can you describe your experience in plain language. It is not a technical conversation and trying to make it one is a common error. Give a two-minute version of your background, ask for the band, ask about the process.

2. Hiring manager screen: 30 to 45 minutes. Tests one thing: would this person be useful on my team within a few months? This is where relevance and communication decide everything, and where depth matters least. Talk about what you built and who used it.

3. Technical exercise: a take-home, or 45 to 60 minutes live. Tests whether you can actually do the work. Its own lessons follow.

4. Deep dive on your own work: 45 to 60 minutes. Tests whether the things on your CV are yours. Expect “why did you do it that way” three levels deep on a project you named. Prepare the two projects you will be asked about; do not let them pick a third you barely remember.

5. Design or case conversation: 45 to 60 minutes. Tests whether you think in constraints and trade-offs rather than tools.

6. Behavioural: 45 minutes. Tests whether you can work with people and whether you tell the truth about failures.

7. Offer.

Timelines, and what they tell you

Two to eight weeks from first contact to offer is normal. Under two weeks usually means a small company or an urgent gap. Beyond about ten weeks, with long silences between rounds, you are learning something real about how that organisation makes decisions, and you will experience the same thing as an employee waiting for a decision about your work.

Ask at the recruiter screen: how many stages, and what is the expected timeline. A recruiter who cannot answer is a signal in itself.

The thing nobody tells candidates about feedback

Panels write down their assessment after each round, and those notes travel forward. The interviewer in round four has usually read what round two thought before they meet you.

Two practical consequences. First, a weak first round is not erased by a strong third; prepare the early rounds harder than instinct suggests, because they are cheap for the company and decisive for you. Second, if you gave a poor answer, you may raise it later (”I have been thinking about the join question from Tuesday and I got it wrong; here is what I would do”) and it is read as strength rather than weakness, because the notes are already there and you are showing you can revisit your own work.

Between the rounds

Send three lines afterwards, naming one concrete thing from the conversation. Not for politeness. It puts a written trace of you into the loop, in your own words, at the moment the panel is deciding.

Thanks for the time today. The point about the nightly load
overlapping with the batch export was the part I keep thinking
about — I had not considered that the retry window and the
export window could collide. Enjoyed the conversation.

Four sentences, specific, no request.

Rejection, and what you can learn from it

You will usually be given no reason. This is legal caution rather than rudeness; a specific reason is a document that can be argued with, and most companies have decided not to produce them.

What you can still learn is the stage. Rejected before any human conversation means the route or the paper is wrong: go back to the employer categories and the referral routes. Rejected after the technical means a skill gap you can name. Rejected after the final round repeatedly means you are close, and it is usually a small number of specific things; that is the point at which asking a friendly interviewer for one piece of feedback is most likely to succeed.

Practising without paying

Interview coaching costs from about a hundred to several hundred per hour, and much of what it sells is rehearsal with an audience. A peer who does the job (swapping mock interviews in a community, an hour each way) is free and often better, because they know the questions your market actually asks. Free mock-interview swaps exist in most large technical communities. Record yourself on a phone and watch it back once; it is unpleasant and it works.