Clay says it has 250,000 accounts in Salesforce. Each rep gets a weekly top five.

Clay says it has 250,000 accounts in Salesforce and 850,000 users. Kyle Poyar reported them in an interview with Osman Sheikhnureldin, published on 8 November. Sheikhnureldin is an early employee and Clay's head of GTM engineering. Poyar's write-up draws the conclusion in one line: "Prioritizing the right accounts matters more than finding new ones."

That is an odd line to write about Clay. Clay sells the tools for finding accounts: enrichment from hundreds of data providers, AI research agents, lead lists. In its Series D announcement in September it said it has more than 17,000 customers and raised $115 million at a $7.1 billion valuation.

Yet the company known for finding treats its own problem as choosing.

One disclosure first. Poyar's newsletter, where the interview ran, states it is "proudly supported by Clay". The source is a Clay employee. Every number below is Clay's account of itself, and none of it has been checked by anyone outside the company. Read it as how Clay says it works.

The same goes for the framing. Clay sells choosing too: the issue opened with an ad for a Clay workshop on building scoring models in Clay. So the line about prioritizing also works as a pitch for scoring in Clay. The workflows below still show what that choosing looks like from the inside.

The narrowing

Sheikhnureldin walked Poyar through 17 automated workflows. The first few do one job: shrink 250,000 accounts down to a list a person can work.

1. Score every account. Each account gets a fit score from 0 to 100, written to a Salesforce field. Inputs include company size, B2B or B2C, whether the company sells product-led or sales-led, sales headcount, CRM, region and funding stage.
2. Gate the territory. An account enters a rep's territory only when its weighted fit score is above 60, it has no open deal and no rep owns it.
3. Cap the book. Sales-assist reps, on companies of 50 to 350 people, get 60 accounts each. Sales-led reps on 350 to 3,000 get 50. Reps on companies above 3,000 get 45.
4. Pick the week. The fit score is one of four inputs to a "propensity" score, which estimates how likely the account is to buy. The others are predicted contract value, intent signals and engagement with Clay. Every week it flags the five accounts each rep should work first.

So 250,000 accounts become 60 or fewer per rep, and five a week. Clay's machine is not built to add accounts. It is built to rule them out.

The free takeaway is the order of operations. Before you buy more leads, decide which of the ones you have deserve a person's time.

It started with a waitlist

This is not new for Clay. It goes back to 2022.

Varun Anand, a co-founder, told First Round Review about the early days. Clay kept a waitlist. Each morning Anand reviewed it in Clay. If you fit the ideal customer profile, he emailed you to book a call, and told you to bring a dataset or a problem to solve live.

Then Clay launched on Product Hunt in February 2022 and dropped the waitlist. Sign-ups flooded in. "It was hard to find signal versus noise," Anand said. In May 2022 Clay turned the waitlist back on. It stayed on for another 15 months, until Clay was in the millions of annual recurring revenue (ARR). He called it "one of the best decisions we made in the early days."

The growth story people tell about Clay is about reach: the Slack community, the agencies, the creator stacks on X. The decision Anand ranks among his best was a gate.

The role it coined

In 2023 Clay gave the people who build this machine a name. Anand wrote on LinkedIn that the internal, ops-focused GTM engineer is a role "we coined in '23". Poyar's piece says Clay popularised the term that year.

Ask Clay what that role should do first and you get the same answer. Everett Berry wrote it down on Clay's blog in April. The first GTM hire should be a GTM engineer, before the first account executive. Their first project is "the golden list": the accounts where, "if you got a magical intro tomorrow, you'd be thrilled."

So the job title Clay sells into is, by Clay's own description, a filtering job first.

The score took a year

The fit score sounds simple. Clay says it wasn't.

In a territory-planning post on 29 October, Clay's team says the above-60 rule "took us a year (and three different versions of scoring) to arrive at". First came letter grades. Then a machine-learning model that predicted contract value from past wins. Then the current weighted score, where two factors carry 49% of the weight: the tools a company runs, and whether it sells product-led or sales-led.

The most telling input is headcount. Sheikhnureldin looked at Clay's top customers by revenue and found they have 10% or more of their staff in sales or GTM. By Clay's own data, its best customers are companies with a big share of staff in sales or go-to-market. Our reading: the tool sold as automation does best where there are many people to automate around.

And nothing lands in a rep's book without a manager. Clay's territory post says each proposed account goes to a sales manager in Slack. They approve or decline. A decline needs a reason, and Clay looks for patterns in those reasons. Managers kept declining companies whose staff were mostly offshore, so Clay added that as an exclusion.

A human sends it

The rest of the 17 workflows prepare people. They don't replace them.

Outreach to a new executive or a former champion is drafted automatically. Reps edit and approve it before it goes. Pre-demo emails wait for the account executive's sign-off. For stalled deals, a rep can queue an email from Clay's head of sales. It lands in her Gmail drafts, and she edits and sends it herself.

Post-call follow-ups arrive in the rep's drafts, written in their voice. Clay builds that voice from what the rep has written and said without automation, pulled into its Snowflake warehouse. On the customer side, Clay estimates its QBR-deck workflow saves 90 hours per customer success manager each quarter. QBR means the quarterly business review deck a CSM prepares for each account.

Poyar's summary: there is "a human-in-the-loop component to nearly every workflow covered here."

What we can't see

The piece describes the machine. It doesn't score it: there are no conversion rates, no win rates for top-five accounts against the rest, and no figure for how many drafts reps throw away. One workflow, "best next steps", is still in beta by Sheikhnureldin's own account.

The two sources also differ on a detail. Poyar's write-up lists about ten fit-score inputs. Clay's territory post says the score is a weighted sum of eight factors. Both come from Clay.

So take the shape, not the numbers. Clay spent 2022 learning to turn sign-ups away. In 2023 it named the role that does the turning away. In 2026 its own team describes a machine that takes 250,000 accounts and hands each rep a weekly top five.