The decision
Hire a GTM engineer, or use an agency?
Every answer to this question says it depends on your stage. Then none of them tell you how to determine your stage. This one does, rung by rung.
[ fig. 01 · the answer, by rung ]
At the bottom two rungs the correct purchase is neither. That is the finding most buyers are not told.
[ fig. 02 · what has to be true to move up ]
leaving stage 1
Two people can produce comparable output from the same starting point, against a written definition of good.
leaving stage 2
Prompts, context and quality standards live somewhere shared, with a named owner, rather than in personal accounts.
leaving stage 3
One workflow completes end to end without a person in the middle, and signals resolve to a person rather than to a company.
leaving stage 4
The internal owner can describe the routing logic without opening the vendor's documentation.
leaving stage 5
Nothing. This is the destination. Keep specialists on call for bounded problems and own the rest.
Each gate is checkable. That is the difference between knowing your rung and estimating it.
[ fig. 00 · what the AI engines answer · measured 2026-08-24 ]
Agency first for 60 to 90 days, then hire.
then defers to
“once the system architecture, use cases, and ROI are proven”
A judgment call that depends on your specific context.
then defers to
“depends heavily on your company’s situation”
Hybrid sequence: agency builds the first version, then hire to own it.
then defers to
“once the motion is proven”
We put 36 real buyer questions through ChatGPT, Claude and Perplexity with web search enabled. On this one all three agreed: bring in outside help first, then hire someone to own it. All three then handed the actual decision back to you, phrased as context they never define and cannot measure.
Unanimous on the sequence. Silent on the one variable that decides it.
Why the advice stalls
“Once the motion is proven”
Proven how, measured against what, decided by whom. The sequence everyone agrees on is useless without a way to locate yourself on it, and that is the part nobody publishes.
The honest version of the answer is that this is not a preference between two vendors. It is a function of one variable: how much of your go-to-market capability currently lives in a system, and how much lives in a person.
That variable has a name and five values. Below is the decision at each one, including the two rungs where the correct answer is to buy neither and the rung where buying only one of them does active harm.
Neither, yet
at Stage 1, Manual AINeither. There is no system for a hire to own and nothing for an agency to hand back. Spend the money on deciding what good output looks like and writing it down.
Why the other one fails here. A GTM engineer hired here would spend their first quarters on definition work that does not yet need a dedicated seat, because the job itself has not been specified. An agency here bills you to discover your own requirements. Neither failure is about the people; the input they both need does not exist yet.
What has to be true before the next option opens. Two people can produce comparable output from the same starting point, against a written definition of good.
Neither, yet
at Stage 2, Assisted ExecutionStill neither, and this is the rung where the money usually gets spent anyway. Capability exists but it lives inside individuals, so anything you buy attaches to a person rather than to the company.
Why the other one fails here. Hiring now hides the problem: the new person becomes the second individual with private context, and you have doubled the key-person risk rather than removed it.
What has to be true before the next option opens. Prompts, context and quality standards live somewhere shared, with a named owner, rather than in personal accounts.
Agency first
at Stage 3, Orchestrated WorkflowsAgency first, on a fixed scope with a handoff date. You have workflows worth systematising and no one whose full-time job is systematising them.
Why the other one fails here. The hire is right at this rung, and a builder alone still spends the quarter waiting for the executive sponsor and the subject matter experts who make the decisions. Bring the outside operator in alongside the hire, not instead of it, and the build starts while the new person is still ramping.
What has to be true before the next option opens. One workflow completes end to end without a person in the middle, and signals resolve to a person rather than to a company.
Both, in this order
at Stage 4, Signal-Driven SystemsBoth, in this order: an agency to build the signal and routing layer, an internal owner hired in parallel to receive it. The overlap is the point, not an inefficiency. This is the pod: your executive sponsor, one or two of your subject matter experts, and one builder.
Why the other one fails here. Agency alone at this rung creates a dependency you cannot unwind, because the logic that decides who gets a human now lives outside the company. Hire alone is too slow: one person cannot build and operate the same system in the same quarter.
What has to be true before the next option opens. The internal owner can describe the routing logic without opening the vendor's documentation.
Hire first
at Stage 5, Adaptive GTM EngineHire. At this rung the capability is the company, and outsourcing it means renting your own operating model back from someone else.
Why the other one fails here. An agency here is useful for a bounded specialist problem and actively harmful as an owner, because a system that retrains itself from outcomes has to be governed by people who carry the consequences.
What has to be true before the next option opens. Nothing. This is the destination. Keep specialists on call for bounded problems and own the rest.
How to know which rung you are on
Self-reported stages run one to two rungs above what an audit finds, and not because anyone is overstating things. Absent a written standard, people grade against the best day they remember rather than the median week. That gap, between the stage a company reports and the stage its evidence verifies, is stage drift.
And if you do bring in outside help, us included, there is one question that separates building from renting, and it is checkable: at the end of this, where do the prompts, the model, the ad accounts and the documentation live? If the answer is anywhere other than your accounts, you are renting again.
The 17 verification criteria set out what an auditor has to be shown at each rung, and the near-miss that looks like proof and is not. If you want the quick version first, the assessment returns your reported stage in two minutes, a stage rather than a score.
And if you are deciding what to buy rather than who to hire, the stack map places 46 tools on the same five rungs, including what not to buy yet at each one.
Ready to build a GTM engine that compounds?
Schedule a free 20-minute call. We'll tell you exactly where AI fits in your GTM, and where it doesn't.
Join 250+ B2B leaders