The thesis
The instinct with interview prep is to collect questions and write an answer for each one. It fails in the room for two reasons. Memorised answers sound memorised — the delivery flattens, the eye contact goes, and a good interviewer can hear it. And the moment a question arrives that isn't on the list, you have nothing, because you built a lookup table instead of a capability.
The alternative is to treat the interview as a recombination problem. Almost every question is asking for one of four things: how you find out what's needed, how you get something built, how you handle a person, or what you'd do in a situation. Each of those has a shape. Each pulls from a small library of your own experience. Prepare the library and the shape, and you can answer a question you've never seen — which is the actual goal, because the questions you've never seen are the ones that decide the outcome.
What "speak to anything" really means
Not that you have an answer for every topic. It means that when a topic arrives cold, you have a reliable way to produce a good answer in the four seconds before you start talking. That's the live protocol below, and it's the single most important thing on this site.
The three assets
Everything you study reduces to building these. Nothing else is worth your time.
1. A story bank — eight to ten, no more
Real situations from your own history, each written down with a number in it. The leverage is that one story serves four or five different questions: a story about a failed rollout answers "tell me about a failure," "how do you drive adoption," "tell me about a time you were wrong," "how do you gather requirements," and "what would you do differently." Ten well-chosen stories cover the entire behavioural surface.
This is the heaviest lift and the highest return. The story bank builder gives you ten slots, each tagged with the questions it covers, and saves as you type.
2. Six frameworks
Repeatable ways of thinking you can apply live to a problem you've never seen. When someone asks "how would you approach the accruals redesign," you don't need to have thought about accruals — you need a discovery ladder you can walk down out loud. Frameworks are what make an improvised answer sound structured rather than rambling. All six are here, each with the sentence that introduces it in the room.
3. Domain fluency in three areas
The team named three projects — the accruals redesign, the headcount dashboard, and consolidated spend for forecasting. Those are almost certainly going to appear as case questions, because interviewers reach for the work in front of them. You need enough fluency to talk about each for two minutes without bluffing, plus the finance vocabulary to not sound like a tourist. That's the domain page.
| Asset | Time to build | Covers |
|---|---|---|
| Story bank | 3–4 hours | Every behavioural question, most judgment questions |
| Frameworks | 2 hours | Every "how would you approach X" question |
| Domain fluency | 2–3 hours | Case questions, credibility, your own questions back |
How to actually study
Five rules. They are boring and they are the difference between feeling prepared and being prepared.
Recall, never re-read
Reading an answer you wrote feels like studying and mostly isn't. The retrieval — pulling it out of your head with nothing in front of you — is the part that builds the memory. Every drill card on this site is collapsed by default for exactly this reason. Read the question, answer it out loud, then reveal.
Out loud, standing up, timed
An answer that reads well silently often collapses when spoken, because writing tolerates subordinate clauses and speech doesn't. Stand up, say it to the wall, and use the timer in the corner of the drill page. Target ninety seconds. Most people's problem is not being unable to answer — it's not being able to stop.
Traffic-light triage
After each drill, rate yourself green, amber or red. Then filter to amber and red and only work those. Your second pass should be a third the size of your first. If it isn't, you're rating yourself too kindly, which is a much more common failure than the reverse.
Spread it out
Four one-hour sessions on four days beats one four-hour session, by a wide margin, for how much survives to the interview. If you only have one evening, use the compressed plan below and accept that you're buying fluency rather than depth.
One unscripted mock
At some point before the day, have someone ask you fifteen questions in a row, in a random order, with follow-ups, and don't stop to correct yourself. It is uncomfortable and it surfaces things no amount of solo prep will — the answer that only works if you get to set it up, the story you can't actually remember the numbers for. If there's no one to do it with, record yourself and play it back once. Once is enough.
The live protocol
This is the part to over-learn. When a question lands that you haven't prepared, run this. It takes about four seconds of thinking and then you talk.
Step 1 — Classify it
Every question is one of four kinds. Naming which one, silently, tells you which framework and which stories to reach for.
| Kind | Sounds like | Reach for |
|---|---|---|
| Discovery | "How do you find out…", "A stakeholder says…" | The discovery ladder; a requirements story |
| Delivery | "How would you build…", "What if it's infeasible…" | Scope & trade; a shipped-project story |
| People | "How do you handle someone who…", "Tell me about conflict…" | Adoption triangle; a conflict story |
| Judgment | "What would you do if…", "How do you decide between…" | Automate/redesign/eliminate; a decision story |
Step 2 — Four beats, then stop
- Name the failure mode. "The way this usually goes wrong is…" It buys you three seconds of thinking time, and it signals experience in the first sentence, because only someone who has done it several times knows how it fails.
- State your move. One or two sentences. What you actually do.
- Ground it once. One example with one number. Not a full story unless they ask for one — a story is the answer to "tell me about a time," not to "how do you."
- Land it. What it cost, what changed, or what you'd do differently. Then stop talking.
Nearly every model answer in the drill bank is built on those four beats. Once you notice the pattern you can generate them yourself, which is the point.
Step 3 — Escape hatches
You genuinely don't know. "I haven't hit that exact situation. The closest I've had is [X] — let me tell you how I'd reason about yours." Confident and honest beats a bluff, and interviewers at this level are specifically watching for whether you'll bluff.
It's past your technical depth. One clause to mark the boundary, then move to what you'd do about it: "That's past where I'd trust my own judgment — I'd want your engineers in that conversation before I promised anything. What I'd bring to it is…"
The question is vague. One clarifier, not three. "Do you mean X, or Y?" Asking twice reads as stalling; asking once reads as precise.
You've started badly. Don't apologise and restart. Land the plane: "…the short version is [one sentence]." Recovering cleanly is itself a senior signal; visible flustering is the only real damage a bad answer does.
The one trap to watch
Trying to sound overqualified is the fastest way to sound junior. Claiming scope, name-dropping frameworks, and answering a small question with a big answer all read as insecurity. What actually reads as fifteen years: short answers, one unflattering detail, naming the failure before they do, and being visibly calm about things that would panic a less experienced person.
Your three things
Decide, before you walk in, the three things you want them saying about you afterwards. Then make sure each one lands at least twice, in different answers. People remember three things from an hour; if you don't choose them, they choose themselves badly.
A defensible set for this role:
Turns fog into something buildable
The person who takes an undefined problem and produces a spec an engineer can estimate against.
Ships things people actually use
Adoption is treated as a design constraint, not a phase at the end. Has the usage numbers to prove it.
Easy to work with across a seam
Won't relitigate other people's decisions, won't over-promise to stakeholders, won't surprise the build team.
Swap any of these for something truer to your history — but pick three, write them on the card you take in, and check them off mentally as they land.
Study plans
The seven-day plan
Roughly an hour a day. This is the version that works.
Domain immersion
Read the domain page end to end. Don't memorise — you're building enough scaffolding that finance vocabulary stops being noise. Pay most attention to the three named projects and to why each is harder than it looks.
Done when: you can explain what an accrual is, and why the accruals process is painful, in two minutes without notes.
Story bank — the heavy day
Fill in all ten slots. Situation, what you did, the number, what it cost or changed. Go and find the real numbers — open old decks, old dashboards, old emails. A story without a number is half a story.
Done when: eight of ten slots are filled and at least six contain a specific figure.
Frameworks
Read the six frameworks. Then the exercise that makes them stick: pick five random questions from the drill bank, and for each one say out loud which framework applies and how, without giving the full answer. Ten minutes, high return.
Done when: you can name all six from memory and say what each is for.
Drill pass one
All of the drill bank, out loud, timed. Answer before revealing. Rate every card honestly. Expect a lot of amber — that's what the pass is for.
Done when: every card has a rating.
Mock, unscripted
Fifteen questions, random order, someone else asking, follow-ups allowed, no stopping. Then re-drill only what went red. This is the day that most changes your performance, and the day most people skip.
Done when: you've been through it once and written down the three answers that fell apart.
Positioning and your questions
The panel page: the boundary answer, the overqualified handling, and the questions you ask them. Get the boundary answer to the point where it comes out the same way twice — if two interviewers compare notes and hear two versions, that's the fail.
Done when: you can deliver the boundary answer in under forty-five seconds, twice, consistently.
Taper
Thirty minutes on amber and red only. Then stop. Don't learn anything new the day before — new material the night before displaces what's already settled and adds nothing you can retrieve under pressure.
Done when: you've stopped. Sleep is a study technique.
The 48-hour compression
If the interview is the day after tomorrow:
- Evening one (2 hrs). Story bank — six slots minimum, with numbers. Then read the frameworks once.
- Morning (45 min). Domain page, skim-then-focus: the three named projects and the vocabulary table.
- Evening two (90 min). Full drill pass, out loud, rate honestly. Then the boundary answer and your questions, twice each.
- Morning of (15 min). The day-of checklist below. Nothing new.
The two-hour emergency version
If that's all you have, in this order, and stop when the time runs out:
- 40 min — six stories with numbers. Non-negotiable; everything else is worth less.
- 25 min — the live protocol above, until you can run it from memory.
- 25 min — the three named projects on the domain page.
- 20 min — the boundary answer and the overqualified answers.
- 10 min — the five questions you'll ask them.
The first ten minutes
Disproportionately important, and the easiest thing to prepare, because you know roughly what's coming.
"Tell me about yourself." Ninety seconds, three beats, and it is not a career history. Where you've been, in one sentence. What you've become good at, in two. Why you're in this room, in one. Then hand it back with a question — "…which is why the FIT remit interested me. Where would you want someone to start?" The handback is what separates a rehearsed intro from a conversation.
Set your own frame early. If you have a view of the role — the seam between design and build, solution definition, whatever you've settled on — say it in the first ten minutes rather than the last. Everything afterwards gets interpreted through it. Say it late and it sounds like a correction.
Ask one question early. Not at the end. Something like "before I answer that — how much of the work today is greenfield versus fixing things that already exist?" It changes the register from interrogation to conversation, and every subsequent answer of yours can be aimed better.
Day-of checklist
Fifteen minutes before, and nothing more
- Read your three things. Just the three.
- Read the story bank index — titles only, not the content. You want the pointers hot, not the text.
- Say the boundary answer out loud once.
- Read the five questions you're asking them.
- Check the one number you're most likely to be asked to defend.
Then close the laptop. Cramming in the last hour reliably makes delivery worse — it loads working memory with material you'll reach for instead of listening. The single highest-value behaviour in the room is listening to the actual question rather than the one you prepared for.