What is GTM engineering? Lessons from our 4 live Clay systems
GTM engineering means building outbound as a system rather than a campaign. A GTM engineer builds the machine. An SDR works the output of it. We run four of these systems on Clay ourselves, so here is what the role actually builds, what it costs to buy, and what our own audit caught.
By Ishan Vats · Founder of IV Consulting · runs 4 live Clay outbound systems · 150+ engagements
GTM engineering is the practice of building go to market as a system rather than as a campaign. Instead of buying a static list and sending to it on a schedule, a GTM engineer builds a machine that watches your market continuously: it detects buying signals such as a job change, a funding round or a tech stack change, enriches each one to a verified contact, scores the account once, then routes outreach across email, LinkedIn and voice in a fixed order, with every touch and reply written back to the CRM the same day. The role is closer to systems engineering than to sales. A GTM engineer builds the machine. An SDR works the output of it. The difference matters because one keeps producing after everyone goes home and the other stops when the person does. When we audited our own four live Clay systems in October 2026, every defect we found was an engineering bug, not a copy problem.
The definition
What does GTM engineering actually mean?
The term showed up because a job stopped working. For roughly fifteen years the outbound playbook was stable: buy a list, write a sequence, hire people to send it, add more people when you wanted more pipeline. That worked while inboxes were quiet and sending was expensive. Both of those things stopped being true.
What replaced it is not a better sequence. It is a different shape of work. GTM engineering treats go to market the way you would treat any other production system: something with inputs, triggers, state, error handling and a measurable cost per unit of output. The person who builds it is doing engineering, and the artefact they leave behind is infrastructure rather than activity.
The one sentence version
A GTM engineer answers one question in a way that a machine can act on: what caused this send. If the honest answer is "it was Tuesday and this account was next in the list", there is no system yet, however sophisticated the tooling looks.
The comparison
GTM engineer vs SDR vs RevOps: who does what?
These three get confused constantly, usually by people hiring for one and describing another. They are not seniority levels of the same job. They are different jobs.
| GTM engineer | SDR | RevOps | |
|---|---|---|---|
| Builds or runs? | Builds the machine | Works the output of it | Maintains the record of it |
| Output | A system that keeps detecting | Conversations this month | Clean reporting and process |
| Scales by | Adding signals and channels | Adding people | Adding rigour |
| Fails when | The trigger is weak | The list is cold | The data is entered by hand |
| Measured on | Cost per verified contact, cost per meeting | Meetings booked | Forecast accuracy |
| Leaves behind | Infrastructure | Activity | Governance |
The practical consequence is that hiring an SDR to fix a broken system gives you a person working a cold list faster. Hiring a GTM engineer to hit this quarter's number gives you infrastructure that pays back next quarter. Both are real mistakes and both are common, because the two roles are often written into the same job ad.
RevOps is the one most likely to already exist at your company. If it does, GTM engineering is not a replacement for it. The CRM work sits between them, which is why we treat the CRM as part of the build rather than as something handed over afterwards.
The buying decision
GTM engineer vs SDR agency vs Clay agency vs DIY: which should you pay for?
Short answer: for a 5 to 50 person team, pay a GTM or Clay agency to build inside your own accounts, not an SDR agency, and hire a GTM engineer only once three or more campaigns are live. The four options differ less on price than on cost shape and on what you still own when the contract ends.
| In-house GTM engineer | SDR agency | Clay / GTM agency | DIY in Clay | |
|---|---|---|---|---|
| Cost shape | Salary plus every tool seat, ongoing | Monthly retainer, often tied to meetings booked | One-time setup, then a monthly fee per live campaign or a handover | Tool subscriptions and Clay credits, plus your own hours |
| What you own at the end | The system, if they documented it before leaving | The meetings. The lists and sequences usually stay theirs | Everything, if it was built in your Clay, sending and CRM accounts | Everything, including the bugs |
| Who fixes it when it breaks | One person | Not your problem, and not visible to you | The agency, on a written scope | You, usually after the numbers drop |
| Best fit | Teams already running 3+ live campaigns | Teams that need meetings this quarter and accept renting them | 5 to 50 person teams that want the machine without the hire | A founder with time, a narrow ICP and patience for credit maths |
| Main failure mode | Hired before there is a system to run | Volume on a cold list, your domain takes the hit | Built in the agency's own workspace, so you rent it forever | Silent defects nobody audits |
Our verdict
For a team of 5 to 50 people, do not start with an SDR agency, and do not start by hiring a GTM engineer. The SDR agency sells you the output and keeps the machine, which is the opposite of what this whole discipline is for. The in-house hire is right eventually, but hired first they spend their first quarter building what a specialist builds in weeks, alone, with nobody to review it. Start with a partner who builds inside your own accounts, then decide whether to run it yourself or keep them on monthly. That is our own model, so weigh the bias, but the test is the same for anyone you hire: ask whose Clay workspace the tables will live in. If the answer is not yours, you are renting.
The same logic applies to packaged AI SDR tools, which sell the machine back to you on an annual contract; we priced Artisan and 11x against a Clay + Smartlead build per meeting.
DIY is a fair choice if you have the hours. Just budget for the part nobody puts in the tutorial, which is auditing what you built. The next two sections show why.
The build
What does a GTM engineer actually build?
Five layers. Miss any one of them and the other four produce activity rather than pipeline.
1. Detection
A defined set of buying signals, each with a source and a schedule. Not a list of accounts, a list of reasons to write. The hard part is cutting this list rather than extending it.
2. Enrichment
A provider waterfall that runs until the contact is verified rather than guessed, with scraping filling whatever has no API. This is where cost per verified contact becomes a number instead of a feeling. We priced that number for Clay, FullEnrich and an n8n waterfall at three volumes.
3. Scoring
One composite score per account, built from trigger type, trigger age, fit and contactability. One score, not five, because five scores means nobody can say why an account was worked.
4. Routing and suppression
The score picks the channel order and the entry point. Suppression is shared state across email, LinkedIn and voice, so a reply on one channel stops the queue on the other two. On LinkedIn the queue should end at a human pressing send; the HeyReach ban shows why.
5. The record
Every touch, reply, no show and pass written back to the CRM the same day. This is the layer teams skip, and skipping it is why so few can tell you what a meeting actually costs them. A system you cannot measure is a system you cannot improve.
Notice what is not on that list: copy. Copy matters, but it is downstream of all five layers, and better copy on a bad trigger is still a message with no reason to exist. That ordering is the single biggest difference between how a GTM engineer and a sales team approach the same problem.
First-hand
What did auditing our own 4 Clay systems actually find?
On 13 October 2026 we audited every outbound system IV runs on Clay: two blog prospecting systems, an MCP prospecting system and a VC partner system, 15 Sequencer campaigns in all. This is what a GTM engineer's week actually looks like.
We are publishing the zero on purpose. The campaigns were one to two weeks old and replies are the only metric we count, so the honest read is "not working yet", which is exactly when you audit. Here is what turned up, and notice that not one item is about copy:
- Authentication. One sending domain published two DMARC records, one at p=reject and one at p=none, left over from an old newsletter tool. Gmail's headers showed no DMARC result at all. Fixed the same day with a single p=reject record; test sends from all 12 inboxes now land in Primary with SPF, DKIM and DMARC passing.
- Capacity. Three inboxes at 20 a day is a 60 email ceiling, and 14 live campaigns were sharing it. The cap was hit every day, about 42 researched leads sat in queue, and the feeds kept paying to research new ones. The fix was 9 more inboxes on three fresh domains, warming for a week before they join a shared pool of 12 across every campaign, rather than more campaigns on the same three. If you are buying inboxes for the same reason, we compared the real cost per cold email inbox across six providers.
- Stale data in the copy. The "hot" tier was defined as companies whose blog had gone quiet, but about 37% of them (26 of 71) had posted in September or October. The data was right when scored and wrong by the time it rendered into an email.
- Wrong inbox. About 12 of 327 founders were being emailed at a previous employer or university address. The VC workflow had a guard that checks the email domain matches the company. The other two workflows did not.
- Timezones. About 15% of one system's leads sat outside the US and UK but were sent on New York hours.
The Claygent credit tip, and its catch
Credits, not actions, are what run out first on Clay's Launch plan. On one blog prospecting run of about 85 companies, 72% of data credits went to just three Claygent research steps. The tempting fix is to point Claygent at your own Anthropic key, which costs 0 Clay credits. We did exactly that on the VC system, and Sonnet with web search came to roughly $30 for about 33 funds, around $1 per fund, several times our own estimate. We moved those agents back to Clay's managed models, and the 100 fund test came to about 530 Clay credits in total. Zero credits is not zero cost. The cheaper lever is moving deterministic web checks (sitemaps, RSS, tag lookups) off Clay entirely and keeping Claygent as the fallback.
Every defect above is cheap to fix and expensive to miss, and most start with the wrong trigger list or a data layer nobody checks. See how we map the buying signals for your ICP, then build the data layer, scoring and CRM writeback in your own accounts, with an audit like this one before anything scales.
The cut list
What GTM engineering is not
Not buying more tools
An estate of ten tools with no shared state is ten subscriptions and one problem. The tools are the cheapest part of this and the least decisive.
Not lead generation
A lead gen agency sells you the output and keeps the machine. GTM engineering builds the machine and leaves it running in your accounts.
Not AI personalisation
Generating a bespoke first line for an account with no reason to be contacted produces a well written message that should not have been sent.
Not a growth hack
It is slower to start than buying a list and it compounds instead of decaying. If you need meetings this Friday, this is the wrong thing to start.
The test
Do you actually need GTM engineering?
Two tests, and they point in opposite directions.
You probably do if
Your reply rates have fallen while your sending volume has not. Nobody can name the signal behind the last booked meeting. Two reps have contacted the same account in the same week. The CRM is updated on Fridays, from memory. You do not know your cost per verified contact. Any two of those together mean you are running a list rather than a system.
You probably do not if
Your market is small enough to work by hand, under a few hundred accounts you could genuinely name. Or your buyers do not produce public signals, which happens more often than the category admits. Or you need pipeline this month, in which case build the system and work a list at the same time, rather than waiting on the build. If you are in the first group, the engineer vs agency vs DIY table above is the next decision.
Those numbers came from one signal: venture funds announcing a new vehicle, detected on a schedule, enriched to the partner covering that sector, written to while the capital was still new. Not a bigger list. A reason to write, and a system that noticed it in time. That is the whole argument, and it is what our GTM and CRM setup is built to produce.
FAQ
Questions people ask about GTM engineering
What is GTM engineering in simple terms?+
What is the difference between a GTM engineer and an SDR?+
Should we hire a GTM engineer or a Clay agency?+
Do we need a GTM engineer or is this something our RevOps team does?+
What tools does GTM engineering use?+
How long does it take to build a GTM system?+
Does running Claygent on your own API key save Clay credits?+
Ishan Vats
Founder, IV Consulting · AI & automation consultant
I build the GTM and AI systems teams run on, from signal detection through to CRM writeback. 150+ engagements over 10+ years. The audit numbers on this page come from the four Clay systems IV runs for its own outbound, not from a vendor deck.
See how we build it →Keep reading
Related guides and work
AI and automation for sales and marketing teams
Where the automation actually pays back once the signals are landing.
Read the guide →
Scraping the signals nobody sells you
The build for the sources that have no API and never will.
See the stack →
Answering a fast signal in minutes, not days
What it takes to respond while the window is still open.
Read the build →Want to know which signals you are already missing?
Book a free 30 minute call. We will look at who you sell to, name the signals firing on those accounts that nobody at your company is watching, and tell you what the window is on each one. If the honest answer is that your market does not produce enough signal to build on, we will say that instead.
Book the signal audit →Free 30 minute call. You leave with the signal list either way.