Design a custom always-on buying signal — GTM Engineer take-home

GTM Engineer60 minutes

Context

Cinderlane is an invented product: a compliance-evidence collection tool for small clinical laboratories (5 to 50 staff) that must pass regular accreditation audits. Customers buy when an audit is coming or an audit went badly. Firmographic lists (lab size, state, specialty) exist, but every competitor can buy the same list, so the team wants one signal that is timely, sourceable from public or low-cost data, and hard for a competitor to copy. You will invent that signal and specify the full path from raw data to a drafted message. Nothing is sent.

Materials

Facts about the market, treat as true. (1) Labs are accredited on a two-year cycle, so audit timing is predictable if you know the last accreditation date. (2) Some accreditation bodies publish a public directory with lab name, accreditation status, and an expiry or renewal date. (3) Labs post for "Quality Manager", "Compliance Specialist", and "Laboratory Director" roles; a new quality hire often starts an audit-readiness project. (4) State health departments publish periodic inspection results and citations in downloadable tables for some states. (5) Labs that add a new test category must file notices; some of those notices are public. (6) Typical buyers: Laboratory Director, Quality Manager. (7) Data is messy: lab names vary across sources, one lab may appear under a parent company name, and public tables update on different schedules (monthly, quarterly, annually). Example directory row: "Redwater Diagnostics LLC, accredited, renewal due in 7 months, 2 locations". Example citation row: "Redwater Dx, state inspection, 3 deficiencies, documentation category, 41 days ago". Example posting: "Quality Manager, Redwater Diagnostics, posted 12 days ago".

Task

Invent one custom signal and specify it.

Name the signal in one line and say in two sentences why it indicates a purchase need now.

Sourcing: which of the provided sources you use, how you join them (including name matching), and what you do when a match is uncertain.

Freshness: how often each source updates, how often you refresh, and when a signal expires.

Scoring: how the signal becomes a score, including how you combine two or more sources and how you weight recency.

Defensibility: why a competitor could not trivially copy this, and the weakest point of your signal.

The first touch: write the message this signal powers, to a named role, using only facts the signal actually provides, and say what you would not claim.

Show your signal working on the three example rows.

Stay inside 60 minutes.

Deliverables

A one-line signal name and a two-sentence rationale
A sourcing and join specification, including the handling of uncertain matches
A freshness plan: refresh cadence per source and an expiry rule
A scoring rule with weights and a worked score for the example rows
A defensibility note including the weakest point of the signal
The first-touch message with the target role and a list of claims you would avoid

Rubric

Dimension	What good looks like	Evidence to look for	Levels, weak to excellent
Signal design	Invents a signal beyond firmographics that is timely, sourceable, and hard to copy, and argues each of those properties.	A named signal combining at least two sources; a written case for timing, sourcing, and defensibility; an honest weak point.	Weak Names an obvious signal such as recent funding or a firmographic list. Adequate A single-source signal with a vague rationale; no argument for timing or defensibility. Good A signal using one or two provided sources with a rationale for timing; defensibility thin. Very good A named signal combining two sources with arguments for timing, sourcing, and defensibility, plus an honest weak point. Excellent A combined signal with a corroborating event, an expiry story, a worked test on the rows, and a clear view of how a competitor might copy it.
Data hygiene	Specifies name matching, update schedules, and uncertain-match handling instead of assuming clean joins.	A join key or fuzzy-match rule with a confidence threshold; staleness handled per source.	Weak Assumes the sources join cleanly. Adequate Mentions name matching without a rule. Good A join rule for names, with no confidence threshold or staleness handling. Very good Fuzzy-match rule with a threshold, a plan for uncertain matches, and staleness handled per source. Excellent Match, parent-company, and multi-location handling specified, with a review sample and per-source freshness limits.
Workflow design	The signal runs on a schedule, refreshes sensibly, and expires when stale, without manual work.	Refresh cadence per source; expiry rule; a restartable flow.	Weak One-time pull; no refresh or expiry. Adequate Says it will refresh without a cadence or expiry rule. Good Refresh cadence stated for the signal as a whole; expiry vague. Very good Per-source refresh cadence, an expiry rule, and a restartable flow. Excellent Scheduled, restartable, and self-expiring, with alerting for a source that stops updating.
First-touch writing	The first touch uses only what the signal actually proves and does not overclaim or sound surveillant.	A message built on one fact; a stated list of claims avoided; a role-appropriate tone.	Weak Generic pitch; overclaims what the signal shows. Adequate Uses the signal but sounds surveillant or lists features. Good Built on one fact and role-appropriate; claims to avoid not stated. Very good One fact, the right role, and an explicit list of claims avoided. Excellent Message and avoided claims tied to what each source proves, in a tone the target role would respond to.
Clear write-up	The spec is clear enough that an engineer could build it and a manager could judge it.	Sections in order; examples used; terms consistent.	Weak Spec missing or too vague to build. Adequate Sections present but disordered; terms shift. Good Readable spec; an engineer would need follow-up questions. Very good Ordered sections and examples; an engineer could build it and a manager could judge it. Excellent Concise and exact, with a worked example for each section.
Working with AI	Uses an assistant to explore options and test the spec against examples, verifying its output.	Several signal options considered then narrowed; worked score checked for errors.	Weak Pastes one assistant answer unchanged. Adequate Asks the assistant for one idea and accepts it. Good Explores several options with an assistant and picks one with a reason. Very good Narrows several options, tests the spec on the example rows, and fixes errors in the assistant output. Excellent Uses the assistant to attack its own spec, verifies every number, and records what it changed.

Why unaided AI alone does not pass this

The sources, update cadences, and messy-data facts are given, so the work is committing to a specific signal and specifying it against that data. An obvious signal such as "recent funding" is easy to name; the judgment is in name matching across sources, staleness on mismatched schedules, and what the message can honestly claim, tested on the three example rows. A reviewer can verify the work by checking the worked scores against the example rows and by asking the candidate to walk through two choices, such as how they handle an uncertain match and why the signal expires when it does.

Notes for the reviewer

The best signals combine the renewal timing with a corroborating event (a quality hire or a documentation citation). Do not require a specific signal; score the argument and the specification. Candidates who invent data sources not in the materials should be asked to ground them. To verify, ask the candidate to walk through two choices, such as an uncertain match and an expiry rule. Not observed is informative, not punitive.

Adapt this in ChatGPT