← Guides

GTM engineer: what the role ships, and what a hire costs

We counted what 19 explainers say the job is, then worked out what a hire costs and how long it takes.

A GTM engineer, or go-to-market engineer, is a software engineer whose product is the company’s own revenue systems: the programs that find buyers, reach them and qualify them. That is the job as we practice it, and it is older than its title. Clay coined the name in 2023, and by Clay’s count about 100 listings a month now carry it, as TechCrunch wrote on October 2, 2026 in announcing the company’s session at TechCrunch Disrupt.

Fewer companies are hiring one than the talk suggests. When Kyle Poyar of Growth Unhinged counted GTM engineer postings in Sumble’s job data in September 2025 he found “128 from the past three months”, against 14 RevOps postings for each one. Clay’s partner directory, he noted, listed more than 120 agencies, and by one estimate his article cites, 45% of those with the title are agencies or consultants. On those counts and that estimate, the job is rented about as often as it is hired for.

Yet you would not learn that from the pages AI engines cite. We asked five engines about the role on October 28, 2026, in the run our guide to generative engine optimization describes, and the pages they cited most, from Norwest, ZoomInfo and Clay, treat hiring as the only route. They also describe the job mostly by its tools, which says little about what a company gets. One other page, cited as often as Clay’s, sells fractional work. We do this work ourselves and sell one way of buying it, which is all the more reason to check our arithmetic.

Where the title came from

The name is three years old. “Clay coined the role in 2023”, TechCrunch wrote on October 2, 2026, adding that “The company says roughly 100 GTM engineering job listings are now appearing each month”. Clay’s own guide, dated November 3, 2026, says the same. The monthly figure is the company’s own count of a job it named, and TechCrunch reports it as such.

The pay and growth figures that travel furthest come from one study. Henley Wing Chiu of Bloomberry, a firm that sells data on what companies buy and hire for, analyzed about 1,000 postings and last updated the write-up on January 25, 2026. Its most-quoted figure is “an increase of 205% new GTM jobs year-over-year”, which compares the first nine months of 2025 with the first nine of 2024 and counts, in the study’s words, “combining both GTM Engineers + Revops”. The method is thinly described, and most pages that repeat the number leave RevOps out of it.

The same study takes up the obvious question, whether this is RevOps under a new name, and comes close to saying yes. It found that “9 out of 10 responsibilities that shows up in GTM Engineer roles appears in RevOps Engineer postings too” and called the two jobs “essentially the same”. We would not go quite so far. The job is real, though most of it used to be called RevOps.

What our count of the explainers shows

The explainers describe a sales job. On November 6, 2026 we rendered the 19 pages that explain the role, 12 that rank on Google and 7 that are among those AI answers cite most, and counted the mentions in their text. Sales, SDRs or reps come up 370 times, Clay 171, enrichment 165, outbound or sequencing 132 and the CRM 123. Content is mentioned 11 times, on 6 of the 19 pages, and SEO or AI answers once. A count of mentions measures emphasis and not much else, so read the bars as proportions.

PROTOCOL What the GTM engineer explainers talk about Mentions in the text of 19 pages, counted November 6, 2026 Sales, SDRs orreps 370 Clay 171 Enrichment 165 Outbound orsequencing 132 The CRM 123 Marketing 122 Scoring or routing 109 Reporting ordashboards 42 Content 11 SEO or AI answers 1
Source: our count of mentions in the extracted body text of 19 explainers, 12 that rank on Google and 7 among those AI answers cite most, November 6, 2026.

Norwest gives the common definition. The venture firm’s explainer was the page most cited when we put two questions about the role to the five engines. “GTM engineers typically focus on five areas”, Braden Koch wrote in it on June 15, 2026: pipeline automation and enrichment, CRM architecture and data integrity, outbound infrastructure, GTM analytics and attribution, and AI-native GTM. A reader of these pages would take the job for outbound and CRM plumbing, with content and search hardly in it. The job is also older than its name: the author of this page built GTM systems before “GTM engineer” was a job title.

What a GTM engineer ships

In our practice a GTM engineer writes software, a definition not everyone shares: Clay’s guide, dated June 2, 2026, says the job is done in no-code tools, and Norwest notes that “Many of the strongest candidates don’t come from traditional engineering backgrounds”. What sets the job apart from the rest of engineering is that each system it ships answers a commercial question, on which the engineer is expected to have a view. The thing handed over is a GTM system, one that produces qualified signups, replies or booked calls. A pipeline, an AI workflow or an application is a part of one, and the code is only how it gets built. The table takes six such systems apart as we would scope them.

GTM systemThe data it collectsWhat gets builtThe business decision behind it
Search and AI answersSearch Console queries and keyword demand; what ChatGPT, Claude, Perplexity and Gemini answer to buyers’ questions, and whom they citeA drafting workflow that writes from the product’s code, docs and engineer calls; a check of every claim before a page leaves; pages staged as drafts or pull requests; a report every Monday of rankings, AI citations and signups from searchWhich questions are worth owning, and what the company is cleared to claim
Target accountsThe company’s best customers, studied; the market scored against them; hiring, new leaders, funding, site visits and engagement on each accountA scored list in tiers, with the buying committee on each account; a monthly change report; a sync to the CRM; a daily ranked queue with the reason for each nameWhich accounts and which people come first
OutboundResearch on each account on the listA message for each account in the company’s voice; a queue where a person approves every one; sending set up on warmed domains; replies sortedWho is worth writing to this quarter, and what to say
Paid acquisitionSpend and clicks by campaign, joined to which signups turned out to be qualifiedSignup quality by source and page; a question about campaigns answered in Slack with the chartWhere the next dollar goes, and which campaigns stop
Inbound lead responseThe prospect’s website, read the moment the form is submittedA fit check; a form that books a qualified lead straight onto a calendar; the reason for every decision written to the CRM; rules the team can changeWho gets a salesperson’s time
SocialThe conversations prospects are having on LinkedIn, X and RedditA morning brief of the few worth joining, who is in each thread, and an angle to addWhere a person from the company should show up, and when to stay out

The base they share

All six stand on the company’s context: its positioning, voice, product and customer knowledge, the definition of a qualified lead and the claims the company is cleared to make. The context lives as files in a repository the company owns, with checks that run on each change and a server that lets the team’s AI tools read it. Beside it sits a data foundation of CRM, analytics, search and call data, kept in a warehouse if the job calls for one, joined and defined in code so that the same question gets the same answer twice. At Aptible we built the connection that lets the team use its company data in Claude Code. The rest is operations, the unglamorous part that keeps a system running: schedules, monitoring, a written runbook and training, so that it works next Monday without the person who built it.

How a system runs

Outbound shows the whole chain. A scheduled job researches each account on the list, and a drafting step writes a message for each in the company’s voice, reading the context to learn what may be claimed. The drafts land in a queue, itself a small application, where a person approves each message before it is sent from warmed domains. Before any of this comes the decision about who is worth writing to this quarter, which the engineer is expected to argue.

Search runs the same way. The drafting workflow writes from the product’s code, docs and engineer calls, a check runs over each claim, and the page is staged as a pull request for an editor to approve. At Aptible, by product marketer Kelsey Ellis’s account, a technical piece now takes an afternoon to draft.

Ordinary engineering, commercial judgment

None of the engineering is unusual. It is code in a repository, tests that run on each change, scheduled jobs, data pipelines, and a web application where a person needs a surface. The commercial half is what makes this a separate job, since an engineer with no view on which accounts come first can build an excellent scoring model for the wrong list.

The first thing shipped in any engagement is a decision, written down before any code: one job, what done means, and who reviews it. A role described by its stack can stay busy for a quarter without making one, whereas a role described by what it ships owes a working workflow that someone runs each week, by a date.

The context comes before the tools

Each system in the table reads from the same files, which is the argument for writing those files first. Whoever builds the systems, whether an employee, a contractor or us, should begin with the written context: positioning, voice, product and customer knowledge, and how a qualified lead is defined. It belongs in a repository the company owns. Skip it and each new workflow has to explain the product and the buyer from scratch, which it will do a little differently each time.

A build that starts from the tools ages quickly, as the explainers themselves show. Clay is named on 14 of the 19. Of the 9 dated before June 2026, none names Claude Code, Codex or Cursor as a tool the engineer works in, though one from 2025 lists the app builders Lovable and Bolt, and none names MCP, the standard through which AI tools read outside systems. Of the 10 dated since, 2 name one of those three coding agents and 2 name MCP. We counted by reading each mention, because a pattern match also catches Cursor the company and the sites’ own menus. Two pages in ten is a small shift on few pages, but the tools the newest pages name are absent from the oldest.

A company’s knowledge of its product and its buyer changes more slowly than its tools do. A workflow that takes its positioning and its claims from files the company owns can be rebuilt in the next tool, because the judgment was not stored in the old one. Generating is something the models already do; whether the output is any good is decided upstream of generation, by what they are given to read. A guide to company context, which we are writing, takes the subject further.

What a hire costs

Only one figure for pay has a study behind it, and it is for pay as advertised. Across “GTM job postings that included public salary ranges”, Bloomberry reports, “the median comes in at $127,500”. That is what employers who publish a range say they will pay, which isn’t the same as what people in the job earn. Official statistics offer no check, since the Bureau of Labor Statistics has no such occupation. The nearest it tracks is sales engineers, a different job, whose median wage was $124,900 in May 2025.

An employee costs more than a salary. The bureau’s Employer Costs for Employee Compensation release for June 2026 puts wages and salaries at 70% of an employer’s total cost for a worker, averaged across private industry. Divide the advertised median by that share and the hire comes to about $182,000 a year. The arithmetic is simple enough that at least one vendor’s page we read does it too, and it leaves out recruiting, equity and tools.

A hire also takes time to make. Ashby’s 2026 Talent Trends Report, published on April 28, 2026, gives a median of 30 days to hire for business roles and 40 for technical ones in mid-2025, and of 56 and 76 days from opening a job to a first fill. A GTM engineer could be counted as either, and no source we found says which, so we give both. After the start date comes ramp, for which we found no independent figure for this role.

As for what the money buys, Bloomberry’s postings ask for an average of just over four years’ experience, and “SQL and Python each appeared in 38% of GTM Engineer job postings”. Someone then has to manage that person and review what gets built. A company that has not had an engineer inside its go-to-market team before should count the managing as part of the cost.

The alternative we sell

A full-time hire is one route to these systems. Agencies and fractional engineers are others, and Poyar’s figure of more than 120 agencies in one partner directory suggests how common they are. We sell a different one, the 2 week sprint, so read its column as a seller’s. The agency and fractional routes have no column, because the cost and timing figures we found for them are sellers’ quotes and one aggregator’s ranges, none with a method we could check.

Full-time hire2 week sprint
What you getAn engineer on your payroll, building the systems and keeping them runningOne working workflow on your data, your company context, training and a written runbook
Time to a first working workflowA median 30 to 40 days to hire and 56 to 76 to a first fill (Ashby), then ramp; no independent figure for ramp in this roleDay 10: handed over with training, a runbook and a review
Cost and its basis$127,500 median advertised pay (Bloomberry); about $182,000 a year once benefits are counted (pay divided by BLS’s wage share); recruiting, equity and tools on topStarting at $7,500 per sprint, paid at kickoff, the price agreed before kickoff; AI and data tool costs stay on your accounts
Who reviews what goes outThe engineer’s manager, which assumes someone on the team can read the workSomeone on your team, within a day; a person approves anything a customer sees
What you own afterwardThe systems and the accounts; the written context only if someone is asked to write itThe workflow, your company context in your own repo, the training and the runbook
If it does not workYou part with an employee and open the search againMoney back if it doesn’t do what was agreed on day 1, provided access and reviews arrived on time
Best whenThe systems will need changing every week for years, and someone on staff can judge the changesOne recurring job, a growth gap you can’t staff, a reviewer within a day

Whichever you choose, the written context belongs in your own repository. A company that owns its tables and logins, but not the positioning, voice and claims its systems read from, owns the plumbing and rents the judgment.

When hiring is the right call

Hire when the systems will need changing every week for years and somebody already on the team can judge an engineer’s output. Norwest, in the explainer quoted earlier, puts the inflection around Series A to Series B. With both conditions met we would hire and not run a sprint, because an employee sees every failure and every request, whereas a visiting team sees two weeks of them. The risk in hiring is a system that only its builder can explain, so make the written context and the runbook part of the job from the first month.

A sprint has limits of its own. It is a poor fit for a strategy deck with nothing running, for a website chatbot, and for volume content nobody reviews, and it depends on a reviewer who answers within a day. Without one, any builder’s date slips. Which systems a company needs at all is a question of GTM strategy, the subject of another guide we are writing.

Questions for whoever builds it

Put the same questions to anyone who would build this for you, whether an employee, a contractor, an agency or us. Ask first what is written down as done, and by what date. An offer letter and a statement of work can each be vague on both, and vagueness favors whoever is being paid. Day 1 is the cheapest time to argue about it.

Ask next what happens if the thing doesn’t do what was agreed: whether money comes back, and on what condition. Find out, too, whether the context is yours, in your own repository, or lives in the builder’s head and the builder’s tools. A system whose judgment leaves with its builder has to be bought again.

Last, ask who checks what reaches a customer, since a system can send a message, publish a page or score a lead with nobody looking. On the answer depends whether automation helps a company or embarrasses it. We think a person should approve anything a customer sees, in our work as in anyone’s, and that the answer should be a name.

Any of these routes can leave a company with a well-described system that nobody runs, and any of them can deliver one that works, provided what done means was written down before the building began. The route matters less than whether a working workflow exists on a date, with a person’s name on what it sends.

Start with one sprint.

An AI engineer and a GTM operator join your team. Each sprint ships one workflow your team runs every week, on your own data.

How sprints work →

Is this your company’s website?

We take a quick look, then you pick a time.

Enable JavaScript and reload this page to continue.