You have a GTM engineer finalist. The resume says they built a Clay-to-HubSpot enrichment and routing pipeline that lifted qualified meetings. The portfolio screenshot looks right. The take-home came back clean. And you still have no idea whether this person architected that pipeline, maintained it, or sat in the standup while an agency contractor built it and handed over a Loom.
That gap is the whole reason reference checks still exist for this role. Interviews test fluency — can you talk about waterfall enrichment, webhook retries, dedupe logic on HubSpot company records. Fluency is cheap now. Ownership is not, and ownership is exactly what gets inflated. What follows is the call script we run on GTM engineer finalists: who to call, what to ask word-for-word, and which answers should stop an offer before you send it.
Ownership is the claim that fails most often
When job seekers admit to embellishing, the top two things they stretch are skill expertise and the scope of prior roles — 59% cop to inflating role scope. Those are not generic resume sins. For a GTM engineer, skill expertise is tool proficiency and scope is build ownership. The two most-inflated claims in hiring are the exact two claims this role is evaluated on.
The reason it keeps happening is incentive math. Most candidates assume verification is selective — 56% believe employers only “sometimes” check resume details. If you think nobody is going to call, stretching “supported” into “built” is a free option.
It is not free, because verification works when it actually happens. Employment verification flags a discrepancy in 15.3% of completed checks — roughly one in seven — and globally it is the check most likely to surface something, reported by 72% of screening respondents in APAC. One in seven is not a rounding error on a role where a bad hire means six months of broken routing and a CRM nobody trusts.
The interview tests whether they can describe the pipeline. The reference call tests whether they can be blamed when it breaks.
Why artefacts stopped being proof
The standard counter-move is “skip references, just do a work sample.” That worked in 2022. Four in ten candidates now use AI somewhere in the application process, and a Clay table, an n8n flow diagram or a HubSpot workflow spec is precisely the kind of artefact a model produces convincingly. A take-home no longer distinguishes the builder from the narrator. A former manager describing a 3am webhook failure does.
Step zero: confirm the reference is a real person with real authority
Before the five questions, establish that you are talking to who you think you are talking to. More than a quarter of surveyed job seekers — 27% — admit to listing fake references. And a small but real share go further: 6% admit to interview fraud, posing as someone else or having someone pose for them. A friend playing VP of RevOps on a 20-minute call is the cheap version of the same trick.
Three things, thirty seconds:
- Reach the reference, don’t accept the reach. You find the number or the LinkedIn, not the candidate. Candidate-supplied mobile numbers with no corporate email are the softest spot in the process.
- Confirm the reporting line out loud. “Just so I have it right — you were Sam’s direct manager from March 2023, or did they report into someone else on the RevOps side?”
- Ask who else touched the system. “Who else was in the HubSpot portal with admin rights while Sam was there?” This single question sets up every answer that follows, because it tells you the cast list before the candidate’s version does.
Then move. The script is five questions and it runs in about eighteen minutes.
The five questions, verbatim
Ask them in this order. The sequence matters — each one narrows the room the previous answer left open.
1. “Walk me through how the Clay-to-HubSpot pipeline came to exist. Who scoped it, who wrote it, and who signed off?”
Open-ended and deliberately unflattering to the candidate. You are not asking “was Sam good at this.” You are asking for a cast list and a chronology. Builders get named spontaneously. If the candidate’s name does not surface in the first forty seconds of a question about who built the thing, note it and keep going.
2. “When that pipeline broke, who got the message?”
The single highest-yield question in the script. Ownership is a function of who gets paged, not who gets credit. Watchers cannot answer this with a name and a channel. Owners answer instantly — “Sam, in #revops-alerts, usually before I noticed.”
3. “What did Sam build that you’d have to rebuild if they left tomorrow — and what did a vendor or contractor build?”
You are giving the reference explicit permission to split the work, which most references want to do and will not do unprompted. Phrasing it as rebuild cost forces specificity: a reference who genuinely relied on the person can tell you which three systems they’d panic about.
4. “Tell me about a decision Sam made in that system that turned out to be wrong, and what it cost.”
Everyone who ships at the data layer has shipped a bad dedupe rule, a runaway enrichment spend, or a routing change that black-holed 200 leads for a weekend. People who have genuinely owned a stack have a scar story and tell it without flinching. People who watched have no scars, because watchers don’t get blamed. “I can’t think of anything, Sam was great” on this question is a finding, not a non-answer.
5. “If I gave Sam the same brief today with no team and no budget for contractors, what gets built and what doesn’t?”
The calibration question. You are asking the reference to draw the line around independent capability. The useful answers are specific and bounded — “the Clay sequencing and the HubSpot side, absolutely; the reverse-ETL off Snowflake, they’d need help, we had a data engineer for that.” Bounded answers from a reference are the strongest signal in the whole call, because they prove the reference is describing a real person rather than endorsing a resume.
Scoring what you hear
Run each answer against this. Two or more amber flags on the same system means the scope claim is inflated by at least one level.
| Question | Owner answer | Watcher answer |
|---|---|---|
| 1. How it came to exist | Names the candidate as scoper or builder unprompted; recalls the original problem and the rejected approach | Names an agency, a contractor, or “the RevOps team”; describes the outcome, not the build |
| 2. Who got paged | A person, a channel, a rough response time | ”We’d raise a ticket” / “it went to the vendor” |
| 3. Rebuild cost | Lists two or three specific systems they’d lose; can split candidate work from vendor work | Cannot separate the two; “honestly it all kind of ran together” |
| 4. A wrong decision | A concrete scar with a cost attached and what changed afterwards | ”Nothing comes to mind” — nobody blames an observer |
| 5. Solo scope | Bounded: clear yes on some layers, clear no on others | Unbounded: “Sam could do anything” |
Unbounded praise is the tell people miss. A reference who says the candidate could build anything is a reference who never watched them build something hard.
The answers that should stop an offer
Not every amber flag kills a hire — plenty of excellent GTM engineers have genuinely inherited systems and run them well, and “I maintained and extended it” is an honest, hireable answer. What should stop an offer is the gap between what the reference describes and what the resume claimed.
Hard stops:
- The reference names a contractor as the architect and the candidate claimed the build. This is the central case. Maintaining someone else’s Clay-to-HubSpot pipeline is a real job; writing “architected” is a scope inflation, and scope is the most-inflated claim on resumes for a reason.
- The reference cannot place the candidate in the system at all. “Sam was on the marketing side, I don’t think they had portal access” ends the process.
- Dates or reporting line don’t match the resume. Employment verification surfaces a discrepancy in 15.3% of checks; when the reference contradicts the timeline, you have landed in that bucket and should finish the verification before anything else.
- The reference turns out not to be the manager they were presented as. With 27% of job seekers admitting to listing fake references, an unverifiable reference is a result, not a logistics problem.
- Question 4 produces nothing, and question 2 produces a vendor. Those two together describe a person who has never been accountable for an outage. For a first GTM engineering hire with no one above them in the stack, that is disqualifying regardless of how the loop went.
What to do with an amber finalist
Don’t kill them — requalify them. Take the bounded answer from question 5 and rewrite the role brief against it. A candidate who genuinely owns Clay, HubSpot workflows and routing, but needs help on the warehouse side, is a strong hire for most teams and a weak one if your first 90 days involve a reverse-ETL build off Snowflake or Databricks. That is a scoping decision, not a character judgement — and it is the conversation you can only have after the reference call. Our GTM engineer hiring guide covers how to set those scope tiers before you open the search, which is the cheaper place to do it.
Where the script fits in the process
Run it after the final loop and before the offer, never after. References gathered post-offer are theatre — you have already decided, and the cost of acting on a bad finding is now a rescinded offer.
Budget 18 minutes per reference and run two: one direct manager, one peer who was in the same systems. The peer call is where you confirm question 3, because peers know exactly which parts of the stack the contractor built. Two calls, same day, is a days-not-weeks addition to your process — not a two-week drag on a candidate you want to close.
And write the answers down verbatim. Not a summary. When you are comparing two finalists three days later, “Sam, in the alerts channel, before I noticed” versus “we’d open a ticket with the agency” decides it in four seconds. Paraphrase loses that.
If you want the script run for you — against references you didn’t source and a scope tier we set before the search opens — that is a standard part of how we work on GTM engineer searches, alongside HubSpot and broader revenue operations roles where the same inflation pattern shows up. Spray-and-pray sourcing produces finalists who interview well and break in month three. A structured call on who actually got paged is the last screen that catches it.
