A GTM engineer interview exercise: audit a broken inbound workflow

Give a GTM engineering candidate a small broken workflow, its business rules and the same tools they would be allowed to use at work. Ask them to diagnose a failure, demonstrate a repair and explain how they know it worked. That gives you concrete decisions to discuss in the interview.

Our downloadable exercise contains synthetic inbound events, rules and definitions, an answer key and a reviewer scorecard. It is a starting point you can adapt and calibrate. GTM Vibes has not validated it as a predictor of job performance, and no cited employer supplied or endorsed it.

The answer key is public. Treat this as an open-book discussion of reasoning and adaptation; memorizing its totals is not evidence of engineering ability.

First, decide which job you are assessing

An exercise should resemble the work in your role. Sanity’s Senior GTM Engineer posting separates core business-system ownership from the applications and workflows its systems engineers build around those systems. LangChain’s Agent Systems posting emphasizes production health, evaluations, cost, latency and business metrics. We checked these employer descriptions on October 22, 2026.

Those mandates call for different depth. An inbound audit is relevant to workflow reliability and operations. It cannot assess a full-stack application’s design, a marketer’s copy judgment or an entire senior engineering role on its own.

Complete this brief before inviting candidates:

Field	Example for an internal GTM systems role
Business outcome	Requests reach the right team and unfinished handoffs are visible.
Systems owned	Intake integration and routing service; RevOps owns territory policy.
Work to demonstrate	Trace an event, apply policy, repair one failure and verify the result.
Operating responsibility	Own failures in the integration and a documented recovery path.
Skills required on arrival	Debugging, basic data analysis and explaining system behavior.
Skills that can be learned	The company’s particular CRM and workflow product.
What this exercise cannot assess	Production performance at scale or long-term stakeholder collaboration.

Replace the examples with your actual boundaries. If you cannot identify the policy owner or who maintains the finished system, resolve that before making it the candidate’s problem.

Send a clear candidate brief

The following is a suggested brief, not a rule used by the cited employers:

You have 60 minutes for this fictional inbound workflow. Read the supplied policy, identify the most consequential failures, and choose one to repair in a local simulation. Return your decisions, a short audit, and a demonstration of the repaired case plus a replay or negative case. Use a spreadsheet, SQL, code or a workflow tool. AI assistance is allowed; be ready to explain and verify what you used. Do not contact the fictional addresses or connect this exercise to a real CRM. Stop at the time limit and tell us what remains unfinished.

Tell candidates in advance what tools and access you will provide, how AI use is handled, what you will assess and how to request an alternative format or accommodation. Use the same baseline conditions. If a paid tool is essential to the job sample, provide it; do not make the candidate purchase a subscription. Offer compensation for a take-home assignment and keep the work fictional.

Sixty minutes is a proposed starting point. Pilot the exercise with people familiar with the target work before setting the time limit or using scores in hiring. This fixture requires no paid tool, CRM account or model subscription.

Observe a few decisions closely

The candidate walkthrough explains the dataset and calculations. During the discussion, use it to follow the candidate’s decisions rather than asking them to reproduce every line.

Ask the candidate	Evidence to look for
Which failure would you address first, and why?	A stated consequence and tradeoff. There can be several defensible priorities.
Why does R14 get suppressed even though it is an existing customer?	They can apply the supplied rule order and flag that a real support policy may differ.
What does R08 count as after its CRM write fails?	It remains an eligible request and an unresolved handoff; the failure does not remove it from the denominator.
What happens if R01 arrives again?	One business action for the same event, plus evidence that replay is handled.
What would you inspect before retrying a timed-out write?	Recognition that the original write may have succeeded despite the missing response.
Is the 12-minute median a good result?	The five responded events exclude four unresolved routes; neither group proves revenue impact.

If the candidate has seen the public answer key, change one business rule explicitly. For example, add a named queue for Canadian demo requests after a territory change. Ask what outputs and checks must change, and which should remain stable. Do not disguise an undisclosed policy as a test of intuition.

Score observed evidence

The downloadable scorecard covers six dimensions: diagnosis, business judgment, verification, measurement, operating handoff, and communication with tool use. Each has suggested anchors from 0 to 2 and space for evidence or a follow-up.

For verification, for example:

• 0: A changed flow is shown, with no rerun.
• 1: A passing normal case is demonstrated.
• 2: The repaired failure, a negative case and a replay are demonstrated with reproducible results.

Record what the candidate did: “Replayed EV001 twice; showed one assignment in output.” Avoid substituting an impression such as “strong systems thinker” for evidence. Leave a dimension blank if the format did not give the candidate a chance to demonstrate it, then follow up.

There is no validated pass score. Have reviewers score the same pilot independently, compare disagreements and revise ambiguous anchors. Decide how the exercise contributes to the broader hiring decision before reviewing applicants. A missing capability in one short work sample deserves a specific follow-up, not an invented claim about the person’s potential.

Make the AI policy explicit

Anthropic’s account of redesigning its performance-engineering take-home describes assessments losing signal as models improved. That is evidence about its particular engineering test, not proof that this GTM exercise can distinguish people from AI or predict hiring success.

The useful lesson for this exercise is to inspect decisions and verification. If AI tools are allowed, ask the candidate to identify an assisted step, one output they checked and any suggestion they rejected. Judge the explanation and observed work. The amount of generated code is not the outcome you are trying to hire for.

End with an operating handoff: who receives failure alerts, who can pause the workflow, what can be retried, and what requires a policy decision. That conversation helps both sides see the work the job actually contains.

Compare your brief with current GTM engineering roles and our GTM engineering versus RevOps guide. When the mandate is clear, submit your role to GTM Vibes with the specific AI responsibility and its source evidence.