RoleMarketing Analyst CompanyClay, New York CandidateS. A. Basharat Ali Oct 2026

Clay's funnel is an assembly line. Here's where it leaks.

A data scientist's read of how Clay markets and sells, plus a working model of a funnel shaped like Clay's: a free plan, unlimited seats, usage metered in Actions, and growth from creators, agencies and community.

See the code → basharat.net sabasharat.ali@gmail.com

The assembly line, station by station

1
Signup12,890
2
Workspace6,390
14signups for the median enterprise
3
Company3,000
2,068personal emails to nurture; 1,571 resolvable
4
UsageActions, 14 days
17accounts over the bar in total, under it per workspace
5
Routingone state each
1,490signups land nowhere (unmapped form)
6
Paid19.4% of companies

Synthetic world built from public information, no real Clay data. Every number comes from python3 run.py.

the research

What Clay is doing in market

$7.1BSeries D, $115M led by Wellington, Sept 2026
~$200Mannualized revenue this quarter
60 / 40reported split, self-serve vs sales-led
50%+of pipeline traced to marketing
ecosystem

Creators, agencies, clubs

100+ LinkedIn creators with affiliate links and boosted posts, 125+ agencies building businesses on Clay, 50+ Clay clubs, a 30k-member Slack.

category

The GTM Engineer

Clay coined the job title, then built around it: a GTM Engineering course, a $1M scholarship fund, and Sculpt on October 8.

brand

Stunts and billboards

At INBOUND 2025: "Every artist has their medium, go-to-market has Clay," plus live activations built to get talked about.

paid

Clay Ads on Clay

Clay's own write-up reports cost per lead falling from $250 to $25 with its Ads product on its own audiences.

pricing

Free plan, Actions, unlimited seats

March 2026 moved to Free, Launch and Growth with Actions metering. Clay's memo called it intentionally revenue and profit negative near term.

product

Templates and agents

Claybooks and templates feed SEO and activation, and Sculpt is set to show what Clay calls agentic GTM.

The tension this role sits in: the channels that grow Clay fastest (creators, agencies, community, stunts) are the ones last-touch tracking can't see, and the March pricing change put the most pressure on the agencies who champion Clay. The posting asks for exactly this: keep the funnel healthy, find leaks and misroutes, and feed ground truth into attribution and goals.

play with it

Who gets credit depends on which ledger you read

Share of signups by channel

six findings

What a data scientist in the pod would catch

1
28.0% → 19.4%signup-to-paid, per signup vs per company

With unlimited seats, "lead" has three grains

62% of companies have 2+ workspaces. Counting signups inflates conversion, and 17 accounts cross the usage bar in total while staying under it in every workspace, so per-workspace routing never sees them.

What I'd do: report the funnel at signup, workspace and company grain side by side, and route on the company.
2
42.1% vs 29.8%paid rate, ranked by usage vs by the form

Usage predicts buying better than the form

14-day Actions
42.1%
"200+" on form
29.8%

The lists overlap on 502 of 1,097 companies; the form misses 595 of the heaviest users.

What I'd do: make Actions velocity the core of the product-qualified lead score, with the form as a tiebreaker.
3
1,490signups landing in no state at all

One unmapped form drops leads silently

Sales 4,121 + self-serve 5,449 + nurture 1,830 + nothing 1,490. Every missing one came from the template gallery form, 933 of them mid-market or enterprise.

What I'd do: a daily reconciliation, signups in = signups routed, with an alert the day a new form ships unmapped.
4
57.9%of signups called "direct" by last-touch

Last-touch can't see the ecosystem

Agencies, community and events get 0%. Self-report fixes most of it but answers only 55.5% of the time and over-credits creators by 6.5 points, because LinkedIn ads get remembered as "LinkedIn".

What I'd do: reconcile last-touch, self-report and affiliate links into one channel view, with how sure each number is.
5
+38.7% → +8.2%stunt lift, pre/post vs control metros

A stunt's naive lift is mostly the launch

The stunt ran in 4 of 16 metros the same two weeks a product launch lifted everyone. Difference-in-differences recovers +8.2% against a true 10.0%.

What I'd do: plan billboards and events with matched control metros, so the next big bet has a defensible return.
6
62.7%of heavy agencies will blow past 40,000 Actions

Which agencies hit the cap, and when you can tell

Day 5 call
76.8%
Day 10 call
83.2%
Day 20 call
92.7%
What I'd do: flag agencies by day 10 so customer success reaches them before the overage, not after.
What surprised me

I expected account-level routing to be the big win. It sent 158 more accounts to sales for 7 more paid companies in 60 days. The unmapped form and the usage score mattered far more.

The part I got wrong

My first version's self-report matched the truth almost exactly, because I'd written it as the truth plus missing answers. Adding real recall mix-ups made it useful but biased, which is the case for reconciling sources.

first 90 days

What I'd do in the pod

DAYS 1–30

Count the assembly line

  • Reconcile every signup to one state, daily; find unmapped forms and sources.
  • Agree funnel stage definitions with Analytics Engineering and GTM Engineering.
  • Report the funnel at signup, workspace and company grain.
DAYS 31–60

Route on behavior

  • Build the Actions-velocity score and test it against the form-based rule.
  • Resolve personal-email signups to companies through shared workspaces.
  • Alert on routing drops and stalled stages in Hex or Sigma.
DAYS 61–90

Give credit honestly

  • One channel view from last-touch, self-report and affiliate links, with confidence.
  • Control-metro designs for the next events and billboards.
  • Day-10 agency cap alerts for customer success; ground truth to Strategic Finance.

Run it yourself

Python standard library only. Under a second.

git clone https://github.com/sa-basharat-ali/plg-funnel-diagnostics
cd plg-funnel-diagnostics && python3 run.py
View the code →
sources

Where the research comes from