Interviewing an AI-Native FP&A Analyst: 15 Questions That Actually Predict Fit

TL;DR: Most FP&A interview questions still test for pedigree and Excel trivia. Neither predicts fit in 2026. Below are 15 questions I use to screen AI-native FP&A analysts, each with the reason it works, a strong-answer signal, and a weak-answer signal. Plus a live exercise, questions to skip, and the FAQs I get from other CFOs.

You are hiring someone whose day is going to include prompting, source-checking, writing, and pushing back. The interview should look like the job. Fifteen questions, one live exercise, done in a 90-minute loop. That is enough.

The 15 questions

1. Walk me through the last LLM workflow you built into a monthly task. What was the before and after?
Why it works: Filters for shipped work, not exposure. Strong: Specific task, specific tool, honest time savings (2 hours to 20 minutes), still uses it. Weak: Generic “I use it for research” or “I tried Claude once.”

2. Where does your prompt library live and who else uses it?
Why it works: Distinguishes a workflow from a novelty. Strong: Names a shared repo, Notion page, or Google Doc; can list two colleagues who use it. Weak: “In my head” or “I retype the same prompts.”

3. Give me a finance task you refuse to automate. Why?
Why it works: Filters for judgment. Strong: Specific example (comp review for a specific person, audit-committee narrative on a bad quarter). Weak: “I automate everything I can.”

4. How do you verify LLM output before it leaves your inbox?
Why it works: Filters for source-of-truth discipline. Strong: Names the specific check (tie to GL, cross-check against source doc, run the sum manually). Weak: “I read it over” or “the model is usually right.”

5. Difference between a paid enterprise LLM tier and a free consumer tier for finance work?
Why it works: Filters for data-privacy awareness. Strong: Names data-retention policy, training-on-input trade-offs, admin controls. Weak: “The paid one is smarter.”

6. Walk me through your last monthly variance analysis. Start with the first email you got and end with the meeting.
Why it works: Tests process, not slides. Strong: Sequenced, names the data pulls, describes the reforecast conversation. Weak: Jumps straight to “I sent a deck.”

7. A department head disagrees with your forecast and CCs the CEO. What do you do first?
Why it works: Tests political sense. Strong: Reply-all with the source data, offer a 15-minute call, keep the CEO out of the details. Weak: “I escalate” or “I defend the number in the thread.”

8. Model this on the fly: 100 customers, $1K ARR each, 3% monthly logo churn, 5% net revenue retention on the survivors. Year-one revenue?
Why it works: Tests mental math and modeling framing under pressure. Strong: Sketches the formula, gets to a defensible range, calls out the assumption tension. Weak: Freezes or reaches for Excel immediately.

9. What is your close-day target and what would you do to compress it by 2 days?
Why it works: Even FP&A candidates should have a view on this. Strong: Names one accrual policy change, one automation, one prep item. Weak: “That is accounting’s job.”

10. Show me the last board slide you built and walk me through the choice of what to leave out.
Why it works: Tests editing judgment. Strong: Names one number they cut and why. Weak: “I put in everything the CFO asked for.”

11. When would you build in Excel vs a planning tool?
Why it works: Filters for tool fluency without dogma. Strong: Names the trigger (multi-user editing, driver-based rebuild, versioning). Weak: “Always Excel” or “Always Pigment.”

12. Tell me about a time you were wrong on a forecast. What did you change?
Why it works: Filters for calibrated confidence. Strong: Specific miss, root cause, what changed in the next cycle. Weak: “I have not really been wrong.”

13. How would you use an LLM to prep for this interview?
Why it works: Meta-check on tool fluency. Strong: Honest description of prompt (company research, likely questions, practice answers). Weak: “I did not use one” (respectable but suggests low daily use).

14. What is a finance opinion you hold that most of your peers do not?
Why it works: Filters for point of view. Strong: Anything specific and defended (“close should be 5 days, not 10, even at the cost of a small accrual miss”). Weak: Generic platitudes.

15. What do you want to be doing in 3 years?
Why it works: Alignment check. Strong: Specific (“run FP&A at a $100M business” or “be a director”). Weak: “I want to grow” with no detail.

The live exercise

Send this 30 minutes before the on-site or on-camera round:

Attached is a one-page P&L for a $12M SaaS business, actuals for June with a $180K miss vs plan. You have Claude or ChatGPT open. Walk me through your first 5 prompts, in order. Then tell me what you would refuse to trust the output on, and what your first email to the CEO says.

What you are testing: prompt sequencing (they should not start with “explain this P&L”), source-checking discipline (they should call out where the miss traces back to), edit judgment (the email should be 4 sentences, not 40).

Questions to skip

Brainteasers. “How many golf balls fit in a 747.” Filters for people who like brainteasers. Not the job.
“What is your greatest weakness.” Everyone has rehearsed a fake answer. Ask about a specific miss instead (question 12).
Puzzle questions. Same as brainteasers.
Resume walkthrough as the main filter. Use it as a 5-minute warm-up, not the whole first round.
“Why do you want to work here.” They want a job. Ask what they would change in the first 90 days instead.

Sample scorecard

Dimension	Signals I look for	Rating (1-5)
AI fluency	Shipped workflow, prompt library, verification discipline
Modeling	Question 8, ability to build without a template
Judgment	Questions 3, 7, 10, refusal to automate the right things
Communication	Live exercise email, response to question 7
Ownership	Question 12, question 15, tone during pushback

Anyone below a 3 on AI fluency or Judgment is a pass in 2026, no matter how strong the modeling is.

How to build the interview kit itself

The interview kit is worth 3 to 4 hours of your time once, and it saves you 20 hours over the next four candidates. Start with these 15 questions, but rewrite two or three in your own voice each cycle. Candidates now share interview questions with each other on Blind, Reddit, and Slack communities within a week of the loop starting. If your kit is copied word-for-word from a template, the answers will be too.

The other thing to build once and reuse: a written rubric for what a 5 looks like on each dimension in the scorecard above. Without a written rubric, your team’s 4s and 5s drift toward whoever the last interviewer liked. With a written rubric, calibration takes 10 minutes at the debrief, not 45.

One last piece of the kit: a written 90-second intro you deliver at the start of every loop. What the business does, who the person will report to, what the first 90 days look like, and the two hardest problems in the seat right now. Candidates who ask sharp follow-up questions to that intro are the ones who will thrive. Candidates who just wait for you to ask them questions are the ones you regret in month four.

Push back on this.

Every operator’s situation is a little different. If you run this differently, disagree with the methodology, or think we got something wrong, tell us. We publish the best counter-approaches on our Reader Contributions page, credited or anonymous, your call. Email hello@thepragmaticcfo.com.

Frequently Asked Questions

How long should the interview loop be?

90 minutes total for FP&A analyst and manager roles. One 45-minute screen with the hiring manager (questions 1 through 8), one 45-minute live exercise plus questions 9 through 15. Anything longer than 3 hours across the whole loop is a signal you cannot make up your mind.

Should I pay for a take-home?

If the take-home is over an hour, yes. $150 to $300 is fair. Paid take-homes get better candidates because the ones with three offers stop dropping out.

Can I use AI to help write the questions?

Yes, and you should. Feed it your JD, your last three hires’ interview notes, and the profile of the person you wish you had. Use its output as a first draft, then rewrite in your voice. If your interview kit sounds generic, your candidates will too.

What if they refuse the live exercise?

Pass. In 2026, refusing to open an LLM in front of a hiring manager for a finance analyst role is a self-select out. If it is a senior candidate with a strong network reference, ask them to walk you through a recent workflow instead. Refusal to do either is a pass.

How do I calibrate “strong” if I have never hired AI-native before?

Do the exercise yourself first. Sit with Claude or ChatGPT and prompt the P&L miss for 30 minutes. Now you know what a strong 5-prompt sequence looks like from your own hand. Your calibration is set. If you feel unqualified to do this, that is the tell that you need to spend a weekend on it before you interview anyone.

Sources

Robert Half, “2026 Salary Guide for Finance and Accounting Professionals.”
Brilyant, “State of AI in Finance 2026” survey.
Michael Page, “Finance and Accounting Talent Trends 2026.”
CFO.com, ongoing coverage of AI-native FP&A hiring, 2025-2026.
Association for Financial Professionals, “AFP FP&A Benchmarking Survey” 2025.

Written by The Pragmatic CFO. 15+ years running P&Ls and building finance teams across portfolio companies.