How to use Clay for GTM (and what 3,342 ads ask)

Clay is a spreadsheet where every column can call an API. You use it for go-to-market by running one loop repeatedly: source rows, enrich them through a chain of providers, then act on the ones that qualify. The tool is straightforward. What separates people who get value from it is the part that happens before the tool opens, which is deciding what qualified means.

The short answers

• The loop is source, enrich, act. Everything else is detail. Clay’s own guides describe a three-layer working loop, and the company says it coined the term GTM engineering in 2023.
• Define the qualifying condition in writing before you build anything. The commonest failure is a technically good pipeline pointed at an ideal customer profile nobody has written down.
• Chain your enrichment providers, do not pick one. A single provider leaves rows blank. A fallback chain fills them, and the coverage difference across a real list is usually large.
• Clay appears in most adverts for the role. GTME Pulse’s analysis of 3,342 postings found Clay named in 69 per cent of GTM engineer job advertisements, a CRM in 55 per cent, and Python or SQL in about 30 per cent.
• The tool gets you read. The code sets the band. In a separate survey of 228 practitioners, high-code engineers reported a median of $135,000 against $90,000 for low-code. Two datasets, two populations, so a comparison rather than a measured premium.
• The job counts disagree by about a factor of two. Clay reports roughly 100 listings a month; an independent count found 45 genuinely titled GTM engineer postings in a month, and that around 45 per cent of people presenting as GTM engineers were agencies or consultants.
• Sample fifty rows by hand before you connect anything that sends. Enrichment fails silently, returning plausible wrong values rather than errors.

What Clay actually is

A Clay table looks like a spreadsheet and behaves like a pipeline. Each row is a company or a person. Each column is one of three things:

1. Data pulled from a provider: a firmographic lookup, an email finder, a technology detector, a jobs feed.
2. A computed value: a formula over other columns, or a condition.
3. Something written by a language model: a classification, a summary, a drafted line, given the row as context.

The reason this shape caught on is that it makes an integration pipeline legible to people who do not write code. You can see the intermediate state of every record, which in a conventional data pipeline requires deliberate instrumentation. That legibility is the actual product.

It is worth knowing where the vocabulary comes from. Clay coined the term GTM engineering, publishes the guides that define it, and sells the tool at the centre of it. That does not make the category unreal, and the demand data below is independent of Clay, but the definitional material in this space is largely written by an interested party, and should be read as such.

The loop, step by step

This is the part that answers the question directly. Six steps, in order, and the order matters more than any individual step.

1. Write the qualifying condition down, outside the tool. One or two sentences that a salesperson would agree with. “A qualified account has between 50 and 500 employees, runs a CRM we integrate with, and has posted a revenue operations role in the last 90 days.” If you cannot write it, the pipeline has nothing to compute and no amount of enrichment will supply it.

2. Source a deliberately small list. Two hundred rows, not twenty thousand. Everything in the next four steps is cheaper to get wrong at two hundred, and credits are consumed per enrichment call.

3. Enrich only the fields the condition needs. This is the discipline people skip. It is tempting to fill every available column because the columns are there. Each one costs credits and adds a place for a silent failure. If the condition needs employee count, CRM in use and recent job postings, enrich those three.

4. Chain the providers. For each field, order two or three providers so that a miss falls through to the next rather than leaving the cell blank. Coverage from any single provider on a real list is materially below complete, and the gaps are not random. They skew toward smaller companies and non-US ones, which is often exactly the segment you were testing.

5. Compute the qualification as its own visible column. Not as a filter, not as a view. An explicit column that says qualified or not and, ideally, a second column saying why. When the output is wrong later, this column is how you find out where.

6. Validate fifty rows by hand, then connect the output. Open fifty rows and check the verdict against your own judgement. You are looking for the systematic error, not the one-off. Only when that passes should the table write to a CRM or a sequencer.

Most people who report that Clay did not work for them inverted steps five and six with step one. They built a working pipeline and discovered afterwards that it was qualifying accounts nobody wanted.

Where the demand actually is

The clearest independent picture of what employers want comes from job advertisements, because an advert is a costly signal about a role somebody intends to fill.

GTME Pulse analysed 3,342 GTM engineer postings:

Named in the advert	Share of postings
Clay	69%
A CRM	55%
Python or SQL	~30%

Set against pay data from a separate survey of 228 practitioners across more than thirty countries:

Population	Reported median
High-code GTM engineers	$135,000
Low-code GTM engineers	$90,000
Non-US respondents (all)	$75,000

Read those two tables together carefully, because the temptation is to read more into them than they support. Clay appears in more than twice as many adverts as Python or SQL. In the survey, the high-code group reports a median half again as high as the low-code group. It is tempting to conclude that adverts systematically underweight the skill that pays.

That conclusion is a comparison across two separate datasets measuring two different populations, advertised roles in one and self-reporting practitioners in the other, and it is not a measured effect. The honest statement is narrower: the tool named most often in adverts is not the attribute associated with the higher reported band, and both figures point the same way. Treat it as indicative and as a reason to look past the tool line in an advert, not as a quantified premium.

One more caveat on all of it. The pay figures are predominantly US, and the survey’s own non-US median of $75,000 is well below its headline. Do not budget a seat outside the US from a US median.

How many of these jobs exist

Worth knowing before anyone specialises, and the counts disagree.

Source	Count	What it counts
Clay	~100 listings a month	The vendor’s own count
Kyle Poyar	45 postings a month	Independent, strictly titled GTM engineer

Roughly a factor of two between them, and the difference is definitional rather than a dispute about facts: how strictly you require the title to match. The same independent analysis found that around 45 per cent of people presenting as GTM engineers were agencies or consultants rather than employees, which matters if you are reading LinkedIn as evidence of how many in-house seats exist. It also means that if you are hiring, a meaningful share of the inbound will be people selling a service.

For scale on the surrounding market, 15,384 marketing technology products were mapped in the 2025 State of Martech, up 9 per cent on 14,106 the year before. The tools multiply faster than the roles that operate them, which is the structural reason tool familiarity is a poor filter: it is a fast-depreciating asset.

Where Clay stops being the answer

Three limits, and knowing them is itself a sign of competence.

Cost at volume. Enrichment is credit-based, so cost scales with rows times columns. There is a crossover where the same work belongs in a warehouse with direct provider contracts. Somebody who has hit that crossover talks about it unprompted.

Silent data quality failure. An enrichment chain that cannot find an answer often returns a plausible one rather than an error. Employee counts go stale, a parent company’s data attaches to a subsidiary, a common name resolves to the wrong person. None of this raises an exception. It just quietly degrades the qualification, which is why hand-validating a sample is a permanent practice and not an onboarding step.

It automates a motion, it does not fix one. This is the important one. A well-built Clay pipeline aimed at a badly defined segment produces more wrong outreach, faster, at greater cost, and with more damage to the domain reputation that outbound depends on. Automation multiplies whatever the motion already was. If the motion does not work manually at ten accounts, it will not work automatically at ten thousand.

Before you build: the data protection question

If you are operating in the UK, enriching and storing information about named individuals is processing personal data, and that is true whether or not the information came from a public profile. Electronic marketing carries its own rules on top of the general ones. The Information Commissioner’s Office publishes guidance on consent and the lawful bases.

The practical point for this article is one of sequencing. Decide what basis you are relying on, what you will store and for how long, before the pipeline is built and running. Retrofitting that to a live system with records already in a CRM is substantially harder than designing it in. Take proper advice on your own circumstances; this page is not it.

What competence actually looks like

If you are hiring for this, or learning it, the observable differences are consistent.

Weak signal	Strong signal
Lists the tools they have used	Describes a qualification they defined, and how they knew it was right
Builds the pipeline first	Writes the condition first, builds the smallest test second
Enriches every available column	Enriches only what the condition needs, and can say why
Uses one provider	Chains providers and knows where coverage drops off
Reports what the pipeline produced	Reports what it got wrong, and how they found out
Automates whatever is asked for	Has refused to automate something, and can say why

The last row is the one that most reliably separates people. A GTM engineer who has never declined to build something has not yet been in the position of seeing automation make a bad motion worse.

Worth naming a structural problem around all of this: only 45 per cent of GTM engineers say their own company understands the role. Whichever way you go, the mandate needs to be specific, because the ambient definition will not do that work for you.

How to test it in twenty minutes

Give a real qualification problem, not a tool exercise. “Here is a segment we care about. Define what makes an account in it qualified, build the smallest thing that tests your definition, and tell me what you would check before trusting the output.”

Watch for three things. Do they ask what the accounts are for, before building? Do they validate a sample by hand? Can they name what the enrichment will probably get wrong for this particular segment?

None of that is visible on a CV, and all three are visible in twenty minutes of watching somebody work. It is also why a tool line in an advert is a weak filter: 69 per cent of adverts ask for Clay, which means naming Clay distinguishes almost nobody.

What this data does not support

Some limits worth stating plainly.

The advert analysis counts what employers wrote, which is not the same as what they went on to hire or what the job turned out to need. Job description language is copied between companies, so tool mentions are partly a fashion.

The practitioner survey is self-reported and self-selected, at 228 respondents across more than thirty countries. Its high-code and low-code medians are group summaries and cannot support a claim about what an individual would earn by learning to code.

The pay figures are not comparable with each other. $132,000 median advertised, $176,000 median total compensation on a $132,000 to $241,000 range and a $94,573 average US base measure advertised salary, total compensation and reported base respectively, on three different populations. Quoting any one of them as the GTM engineer salary is wrong, and the spread between them is mostly definitional rather than real disagreement.

And there is no published study we are aware of measuring whether Clay-built pipelines outperform the alternatives on revenue. The adoption is well evidenced. The return is not.

Start with the condition, not the tool

The mechanics of Clay take a week to learn and the documentation is good. Nothing in this article is an argument against learning it, and 69 per cent of adverts mean you should.

What takes longer is the judgement about what to point it at: which accounts are actually worth contacting, what makes one qualified, when the answer is to fix the motion rather than automate it, and when to refuse. That judgement is what the higher band is attached to, it is what survives the next tool, and it is the only part of this that a job advert cannot check for.

Related reading

• What is a GTM engineer, for the role definition, the full pay picture and how many of these jobs exist.
• GTM engineer versus RevOps manager, for which of the operations roles you actually need.
• The GTM engineer role, for how the role is advertised in the UK market, rung by rung.
• What a recording shows that an interview cannot, for watching someone work rather than reading their account of it.
• RevOps job descriptions that attract the wrong applicants, for why a tool list is a weak filter in an advert.
• Work sample tests versus interviews, for the evidence behind testing the work rather than the description of it.