Your questions, and what the answers reveal

Why this matters more in this field than most

An “AI role” at a company with no data warehouse, no pipelines and nothing in production is a year of plumbing sold as machine learning. That is not always a bad job (early plumbing is how a lot of good careers start) but you should know which one you are accepting.

Your questions are the only instrument you have.

The seven questions

”Where does the data live today, and who owns it?” The strong answer names systems and people. A vague answer usually means it lives in application databases and spreadsheets, and that extracting it is your first year.

”What is in production that uses a model, and when did it last change?” The most diagnostic question in the list. “We are exploring several initiatives” means nothing is in production. A specific answer (“the fraud scoring model, retrained monthly, last deployed in July”) tells you there is a working system to learn from.

”How do you know when a change made things better?” If nobody can answer, there is no measurement culture, and your work will be judged on how it looks in a meeting.

”What did this team ship last quarter, and how long did it take from idea to live?” Cuts through everything. Six weeks means a functioning team. “We are still waiting on approvals from a project started last year” means your work will queue.

”Who would I ask when something breaks at nine in the evening?” Tests whether there is anyone senior, whether there is an on-call arrangement, and whether it is written down.

”What does the first ninety days look like for this role?” A manager who has thought about you arriving has an answer. One who has not is hiring a pair of hands for an unspecified backlog.

”Why is this role open?” Growth, backfill, or a reorganisation. All three are fine and they are different jobs. If it is a backfill, ask, pleasantly, what the previous person went on to do. A predecessor who lasted seven months is not disqualifying, but it is a thread worth pulling.

Reading the answers

Hesitation matters more than content. Everything you are told is the best available version of the truth, so the reliable signal is behaviour: how quickly they replied to your emails, whether the process matched what they described, whether the interviewers had read your CV, whether anyone was late without saying so.

A company that runs a shambolic interview process runs shambolic projects. The interview is the sample.

What not to ask

Anything on the website. Holiday policy in round one: ask the recruiter later, when it costs nothing. “What does the company do.” And avoid the performed question that is a statement about how diligent you are; interviewers can tell, and it wastes the slot.

Asking about money

Ask the recruiter, early, plainly: “What is the band for this role?” It is a normal question, it is legally required to be answered in a growing number of places, and it prevents four weeks of interviews for a job that cannot pay you. Nobody senior thinks less of a candidate for asking. The awkwardness is entirely in your head, and it costs real weeks.

Checking the answers independently

Free sources for pay and culture signals: Levels.fyi, Glassdoor, AmbitionBox in India, national salary surveys published by statistics offices and industry bodies, and public job adverts from the same company that do list a band.

Better than any of them: ask a current or former employee directly, using the outreach method from the previous module. A ten-minute conversation with someone who left eight months ago is worth more than every review site combined, and people are usually willing.

The question to ask yourself afterwards

Not “did I do well” but “would I want to spend two years learning from these people”. A job where the work is dull but two colleagues are excellent teaches more than a job with a fashionable stack and nobody to learn from. At the start of a career, who you sit next to is the largest single factor in how fast you improve, and it is the one candidates weigh least.