GTM engineer technical interview questions

• 01 You could use a low-code automation platform to build a workflow that chains four automations, or you could write fifty lines of code calling API endpoints to do the same thing. Which would you choose? Whether they ask who maintains it before choosing a shape

  Follow up with

  • In what context would you choose the other one?
  • Which of the two would you hand to someone else on the team?
  • What breaks first when the vendor changes a field?

  Strong answer

  A strong candidate will not answer this yet. They will ask who else has to maintain the thing once it exists. Tell them they are the only one and they will usually take the code, because it is more expressive and quicker to change. Tell them a team shares it and they will ask two more questions: does the low-code platform genuinely make collaboration easier, and is there a development environment with review and permissions to write code in at all?

  Weak answer

  A weak candidate picks one straight away, usually whichever they are personally better at, and argues it is the better choice in general. They treat the question as one that has an answer.

• 02 You own the lead router. About 4,000 inbound leads a month come in, and roughly 60 a week fail classification because the company name does not resolve to anything you hold. Nobody works an exceptions queue today. You could drop those records and raise an error, or route them to a default owner and keep going. Which would you build? Whether the volume in the question changes their answer

  Follow up with

  • When would you switch to the other one?
  • Who finds out, and how long does that take?
  • What volume of unclassified records would change your answer?

  Strong answer

  A strong candidate uses the numbers. Sixty a week is small enough to work through by hand, so they will build the queue and name who owns it, because a default owner silently absorbs leads nobody looks at again. Expect them to say what happens if that number grows: at six hundred a week the queue stops being workable and the answer flips to a default owner plus a weekly report on what landed there.

  Weak answer

  A weak candidate picks whichever option keeps the pipeline moving and never comes back to the sixty records. They talk about the router rather than about who eventually deals with what it could not classify.

• 03 You could move the same data with a real-time sync or with a scheduled batch. How do you decide which one to build? Whether they ask who consumes the data before choosing a shape

  Follow up with

  • What does real-time cost you that a batch does not?
  • What does a retry look like in each?
  • Who is waiting on this data, and how fast do they actually need it?

  Strong answer

  A strong candidate turns the question back to whoever reads the data. They will ask what decision it drives and how stale it can get before that decision changes. Tell them a rep acts on it within the hour and they will argue for the sync, then name what it costs you: retries, ordering, and a failure you cannot cleanly re-run. Tell them it feeds a weekly report and they will say the batch gets the same answer with far less to operate.

  Weak answer

  A weak candidate argues that real time is simply better technology, or that batch is cheaper, without asking what the freshness is for.

• 04 Two systems hold different values for the same field on the same record. How do you decide which one wins? Whether they ask what the field is used for before picking a winner

  Follow up with

  • What happens to the value that lost?
  • Where do you write that rule down so the next person applies it the same way?
  • What would you do if the losing value turned out to be right?

  Strong answer

  A strong candidate will ask what the field is used for and who edits it in each system before picking a winner. Tell them a person maintains it in the CRM and a vendor guesses at it, and they will keep the human value. Tell them both are vendor guesses and they will want a sample checked against something they can verify. Either way they end up with a rule per field rather than per system, and they keep the losing value somewhere they can get it back.

  Weak answer

  A weak candidate picks whichever system is newer, or whichever they trust more in general, overwrites the other, and leaves no rule behind for the next disagreement.

• 05 You have a nightly job that enriches about 200,000 records against a vendor API. Tonight it stopped two-thirds of the way through because you hit the rate limit, and the job cannot resume: a rerun starts from the first record. What do you change? Whether they fix tonight and the design separately

  Follow up with

  • What happens to the records the job already processed?
  • How would you make the second run cheaper than the first?

  Strong answer

  A strong candidate separates tonight from tomorrow. Tonight they want the run finished without paying again for the two-thirds already done, which means making it resumable before they rerun it. Then they go after the volume itself, by requesting only records that changed, batching where the vendor allows it, and caching what does not move month to month. Expect them to notice that 200,000 records every night is itself the problem, since most of them cannot have changed since yesterday.

  Weak answer

  A weak candidate adds a delay and reruns from the top, paying twice for two-thirds of the records. They treat the rate limit as the vendor being stingy rather than as a signal about how much they are asking for.

• 06 You send about 30,000 cold emails a month across eight domains. Reply rate has held steady just under 2% for the last quarter, but seed testing shows inbox placement has slipped from the high 80s to the low 60s over six weeks. Nothing about the copy has changed. Would you cut volume, add domains, or rewrite the content? Whether they read reply rate and placement against each other

  Follow up with

  • How would you tell which one worked?
  • What are you measuring placement with?
  • How long do you wait before you decide a change helped?

  Strong answer

  A strong candidate notices that a steady reply rate alongside falling placement means the mail is reaching a shrinking group of people who still like it, which makes the content the least likely culprit. They will ask what else moved in those six weeks, the list source first, then cut volume because it is the one move you can undo. Expect them to name adding domains for what it is, which is buying room rather than fixing a cause.

  Weak answer

  A weak candidate adds domains because it is the fastest lever, or rewrites the copy because that is the visible thing, and changes several of them at once so the next seed test cannot be attributed to anything.

• 07 Your outbound numbers have been sliding for six weeks. Reply rate is down about a third and bounce rate has crept from 2% to 5%. You buy contact data from a single vendor, and you have not changed your targeting or your copy. How do you work out whether the source has gone bad or the market is just slow? Whether they use the bounce rate rather than the market

  Follow up with

  • How would you know inside a month rather than a quarter?
  • What did you have in place beforehand that lets you tell?
  • Would you check the source or the segment first?

  Strong answer

  A strong candidate goes to the bounce rate first, because that number does not care what the market is doing: 2% to 5% means the data got worse, not that buyers got busy. From there they want a controlled comparison, running the same segment and the same messaging against a sample from a different source at low volume. Expect them to ask what changed on your side that you have not mentioned.

  Weak answer

  A weak candidate reasons about the market, waits another month, or switches vendor on a hunch with nothing to compare against. The bounce rate is sitting in the question and they never pick it up.

• 08 You run outbound across four domains with five inboxes each, sending about 20,000 emails a month at good placement. Sales wants to triple that. You could add inboxes to the domains you have, stand up more domains, or move to a different provider. What do you do? Whether they do the arithmetic before spending

  Follow up with

  • What is the limit you are actually hitting?
  • What does tripling the send do to your reply handling?

  Strong answer

  A strong candidate works out what is actually capping you before spending anything. Twenty inboxes carrying 20,000 a month is roughly 45 a day each, comfortably inside what an inbox can carry, so per-inbox throughput is not the limit and adding inboxes is the cheap first move. Expect them to name domain reputation rather than provider capacity as the thing that breaks as you climb, and to raise what tripling the send does to whoever answers the replies.

  Weak answer

  A weak candidate adds domains because that is what scaling outbound looks like, without doing the arithmetic the question hands them or noticing that the existing inboxes are running at a fraction of what they could carry.

• 09 You enrich accounts from three providers. They disagree on employee count for about a third of your accounts, sometimes by a factor of three. That field decides which of two segments an account lands in, and the boundary sits at 200 employees. What lands in the CRM? Whether they shrink the problem to the boundary that matters

  Follow up with

  • How would you check which source is right, rather than which is most confident?
  • What do you do with the two that lost?

  Strong answer

  A strong candidate narrows the problem to the boundary. Providers disagreeing between 40 and 120 employees changes nothing here, because both land in the same segment; only accounts near 200 need resolving, and that is a far smaller set than a third. They will check that set against something verifiable rather than voting on it, and keep the disagreement recorded so the next person can see the field was contested.

  Weak answer

  A weak candidate resolves all of it, by majority vote or by trusting the priciest vendor, and writes a single number in. The 200-employee boundary is in the question and never gets used to shrink the work.

• 10 You want to use an LLM to infer whether an account runs a product-led motion, and route leads into different sequences based on the answer. A wrong route means the prospect gets a less relevant first email. Nothing else downstream depends on the field. When is it good enough to route on? Whether the threshold scales to what a mistake costs

  Follow up with

  • What would you need to see before you trusted it?
  • How would you build that check without hand-labelling a thousand rows?
  • What happens when it drifts?

  Strong answer

  A strong candidate scales the bar to the consequence, and here the consequence is small and reversible, so the threshold is far lower than instinct suggests. Expect something like labelling a couple of hundred accounts by hand, checking whether the model beats whatever you do today, and shipping it if it does. They will assume the field drifts and say how they would notice, usually by re-checking a sample on a schedule.

  Weak answer

  A weak candidate demands production-grade accuracy for a low-stakes routing decision, or reads the model's own confidence score as though it measured correctness. Either way the cost of being wrong, which the question states plainly, never enters the reasoning.

• 11 You are enriching 50,000 cold accounts a month for work email addresses. Three providers are on contract at 8, 5 and 3 cents a match. The expensive one matches about 70% on its own; the cheap one about 40%. How do you order them? Whether they reason about overlap rather than match rate

  Follow up with

  • What tells you to stop paying for another hop?
  • What does the third provider have to add to be worth it?
  • How do you know a provider is only finding what the last one already found?

  Strong answer

  A strong candidate runs the cheap provider first, because every match it finds is one you did not pay 8 cents for and the expensive one still catches what it misses. Then they will say the number that actually decides the order is not either match rate but the overlap between them: if the cheap provider's 40% sits inside the expensive one's 70%, the order is free money, and if the expensive one is the only route to a distinct slice, the order matters less than it looks.

  Weak answer

  A weak candidate orders by match rate and puts the strongest provider first, treating the three rates as though they add up. They quote the 70% without asking what population it was measured on.

• 12 The same company sits in your CRM three times under three different domains, because it acquired two smaller businesses and kept their websites live. Each record has its own open opportunities and its own owner. How do you handle it? Whether they treat it as a hierarchy question or a matching one

  Follow up with

  • Where does that strategy break?
  • What do you do about subsidiaries that should stay separate?
  • Would you merge automatically, or queue it for a person?

  Strong answer

  A strong candidate works out what should be true commercially before touching the data, because this is not really a matching problem. Three records with three owners and three open opportunities may be exactly right if the acquired businesses still buy separately. If it should be one account, they will reach for a hierarchy rather than a merge, so the opportunities and the history survive. Expect them to be wary of anything automatic here.

  Weak answer

  A weak candidate proposes fuzzy matching on company name and a merge, without asking whether the three should be one account, and without saying what happens to two owners' open pipeline.

• 13 Someone wants a new required field on the opportunity object. What do you check before saying yes? Whether they ask what writes to the object before answering

  Follow up with

  • What happens to the records that already exist?
  • Who is creating opportunities in ways you are not thinking of?

  Strong answer

  A strong candidate asks what creates opportunities today before they ask who wants the field. They will list the ways in, from integrations and imports to automations and whatever the reps use on a phone, because a required field breaks every one of them that does not supply it. Expect them to raise the backfill for records that already exist without being prompted.

  Weak answer

  A weak candidate evaluates whether the field is a good idea, adds it, and discovers the broken integration when a rep cannot save a record.

• 14 You inherited a lead score. It is a gradient-boosted model over about forty features, it was trained eighteen months ago and has not been evaluated since, and reps have stopped looking at it. Would you retrain it, simplify it, or make it explainable? Whether they measure the score before choosing a fix

  Follow up with

  • How would you find out whether the score is actually wrong?
  • What would make a rep act on it tomorrow?

  Strong answer

  A strong candidate will not choose until they know whether the score is wrong, and the question gives them grounds to suspect it: eighteen months without evaluation. The first move is measuring it against what actually closed, which is cheap and settles the matter. Expect them to point out that if it turns out to be accurate, retraining is wasted work and the real problem is that forty features produce a number no rep can argue with.

  Weak answer

  A weak candidate picks one of the three and defends it, usually retraining because the model is old, without checking whether its age has actually cost anything.

• 15 You need to turn raw signup events into a qualified-account flag. It is read by the CRM for routing, by a dashboard the CEO checks, and by an outbound sequence. Your warehouse is dbt on Snowflake, your sync is Fivetran and Census, and RevOps cannot write SQL. Where does the logic live? Whether one definition beats the team who can edit it

  Follow up with

  • Who needs to be able to change it in six months?
  • Where do you find out it went wrong?

  Strong answer

  A strong candidate puts it in the warehouse, because three consumers of one definition is the entire argument for defining it once, and dbt brings version history and tests with it. Then they deal with the cost the question hands them, which is that RevOps cannot change it. Expect a concrete answer to that: a threshold in a table RevOps can edit, or a stated agreement about who they ask and how quickly it happens.

  Weak answer

  A weak candidate puts it wherever is quickest, usually the sync or the CRM, and does not reckon with three consumers drifting apart. Or they choose the warehouse and never mention that the team who owns the definition cannot touch it.

• 16 Your attribution pipeline runs at 6am. Ad platform data for the previous day is complete about 40% of the time; the rest of the time a few percent of spend lands up to three days late. The CMO reads the dashboard at 9am and has quoted its numbers in a board deck. Do you fix this in the pipeline or in the consumers? Whether a published number changes how they handle lateness

  Follow up with

  • What does that do to a number someone has already reported?
  • How late is late enough to stop waiting?

  Strong answer

  A strong candidate takes the board deck seriously, because it turns a freshness problem into a credibility one: a number that quietly changes after it has been quoted is worse than a number that arrives late. They will handle it in the pipeline, restating the last few days on every run, and they will want the dashboard to show which days are still settling. Expect a specific window rather than a principle.

  Weak answer

  A weak candidate pushes it to the consumers and leaves every dashboard to solve it separately, or says the ad platforms should deliver on time. That the CMO has already quoted the numbers is in the question and does not change their answer.

• 17 A 400 million row events table takes six hours and about $200 of warehouse credits to rebuild from scratch. It is loaded incrementally on an updated_at column. Upstream, records are occasionally hard-deleted, and edits are sometimes backdated. When would you run a full refresh? Whether they name what incremental structurally cannot see

  Follow up with

  • How would you know your incremental load has been quietly missing rows?
  • What would you have to know to stop doing full refreshes?

  Strong answer

  A strong candidate names what the incremental load cannot catch, and the question lists both cases: a hard delete leaves no updated_at to find, and a backdated edit lands behind the watermark. So a full refresh is not the expensive option, it is the only way to learn that the incremental path is still right. Expect a cadence tied to how much drift matters, and a cheaper check between refreshes, usually row counts or checksums by period.

  Weak answer

  A weak candidate answers on cost and says full refreshes are for when something has gone visibly wrong. The hard deletes and backdated edits are stated in the question and never get connected to what incremental misses.

• 18 The CRO wants attribution for a channel you cannot cleanly measure. What do you give them? Whether they ask what decision the number drives

  Follow up with

  • What would you refuse to put a number on?
  • How do you present the uncertainty without it being ignored?

  Strong answer

  A strong candidate asks what the number is for before producing one. Tell them a six-figure budget decision hangs on it and they will argue for standing up a holdout. Tell them it is going on a slide and they will give you a directional read with its assumptions written down. Either way they tell you plainly what the number cannot be used for.

  Weak answer

  A weak candidate either refuses because the channel cannot be measured, or produces a clean last-touch figure and lets it get quoted as fact.

• 19 Pipeline created jumped 40% overnight on the dashboard the CRO reads. Two things shipped yesterday: a change to how opportunity stages map to the funnel, and a new inbound campaign. How do you work out whether the metric moved or the pipeline broke? Whether a definition change is their first suspect

  Follow up with

  • What is the first query you run?
  • What would you have built beforehand to answer this in a minute?

  Strong answer

  A strong candidate goes to the stage mapping first, because a definitional change is the likeliest cause of an overnight step change and it is sitting in the question. They will check whether the jump appears only in new records or restates history, which separates the two candidates cleanly: a campaign cannot move yesterday's numbers, a remapping can. Expect them to look at row counts before they look at logic.

  Weak answer

  A weak candidate starts with the campaign because it is the better story, or explains why the business might have changed. Both things shipped yesterday and only one of them can rewrite historical rows.

• 20 You want to know whether a new onboarding step improves activation. You get about 400 signups a month and activation sits near 30%. Detecting a five-point lift at that volume would take most of a year. The change is a week of work and easy to roll back. What do you run? Whether reversibility lowers the evidence they demand

  Follow up with

  • What would you accept as evidence here that you would not accept at scale?
  • How would you avoid fooling yourself?

  Strong answer

  A strong candidate does the arithmetic the question invites, concludes a year-long test is not a real option, and says so plainly. Because the change is cheap and reversible, they will ship it and watch rather than wait. Expect them to name what would make them roll it back, and to be honest that they are trading a clean causal answer for a decision they can make this month.

  Weak answer

  A weak candidate runs the underpowered test anyway and reads whatever comes back at six weeks, or refuses to ship without a result. The change being easy to roll back is stated in the question and does not enter the decision.

• 21 You are about to rebuild a three-step onboarding flow. Today you have one event, fired when someone finishes all three steps. You have two weeks before the rebuild ships. What do you add now? Whether they instrument to locate a drop, not just see it

  Follow up with

  • What is hard to add once the change is already live?
  • How long do you wait before making the change?
  • What would you capture that you do not expect to need?

  Strong answer

  A strong candidate instruments entry and exit on each step, so a drop can be located rather than merely noticed, and does it now because two weeks buys a baseline that cannot be reconstructed afterwards. Expect them to name what is hard to add later, such as which variant someone saw or what they did instead of continuing. A good one will also say two weeks at this volume is a rough baseline rather than a clean one.

  Weak answer

  A weak candidate adds more completion events, or waits and instruments the new flow, which leaves nothing to compare against. The two-week window is in the question and goes unused.

• 22 Marketing wants to send a nurture campaign to a segment that includes accounts on your suppression list, some excluded for legal reasons and some because sales asked. Lists reach the sending tool three ways: a CRM sync, a reverse ETL job, and CSVs that people upload by hand. Where do you enforce suppression? Whether they find the path that bypasses the rule

  Follow up with

  • Who can override it, and does anyone find out?
  • What happens when a list is uploaded straight into the sending tool?

  Strong answer

  A strong candidate puts the check at the send itself, because the question names three ways in and a rule enforced on any one of them leaves the other two open. The hand-uploaded CSV is the one that matters, since it bypasses everything upstream. Expect them to separate the legal exclusions, which should be impossible to override, from the sales ones, which should be overridable deliberately and visibly.

  Weak answer

  A weak candidate filters the segment for this campaign, or enforces the rule in the CRM sync and leaves the CSV path open. They also treat the legal and sales exclusions as one list when the question distinguishes them.

• 23 If you had full control, what stack would you build internal tools on? Whether the stack is argued from maintenance or is a list of tools

  Follow up with

  • Who else on the team can change it after you build it?
  • What happens to it when you leave?

  Strong answer

  A strong candidate asks who uses it, how many of them there are, and who maintains it once they have built it, before naming a single tool. The choice then gets argued from how fast it can change and who can change it. Expect them to name the point where their own pick stops paying, which is where the tool should become a real service or should not exist at all.

  Weak answer

  A weak candidate names a stack and stops. The tools are current, and the answer contains nothing about who else can edit them.

• 24 A rep needs a permission that would let them delete records. What do you actually grant? Whether they ask what the rep is actually trying to do

  Follow up with

  • How would you know afterwards that they used it?
  • What do you do when they need it again next week?

  Strong answer

  A strong candidate asks what the rep is actually trying to do and how often, because the request is for a permission and the need is for an outcome. Then they look for the narrower thing that solves it, whether that is a scoped role, a soft delete, an approval step, or somebody else doing it, and they leave a record of what was used.

  Weak answer

  A weak candidate grants it because the rep is blocked, or refuses because it is dangerous. Either way the underlying request goes unexamined.

No question carries that tag.