Clay AI agent benchmark: checking 100 contacts for job changes

With Clay, an AI agent can check 100 contacts for job changes in 8.6 to 23.3 minutes for about 14 Clay credits per person (about 1,400 for the list) by running one Enrich Person and Find Contact Details call per LinkedIn profile. In our benchmark, 17 runs used that method and 13 of them scored 92 to 96 of 100 correct, each catching 42 of the 45 people who had moved.

By The Prompting Company · Snapshot November 5, 2026

What you need before you start

Each contact needs three things on file: the company you think they work at, their title, and their LinkedIn profile URL. The LinkedIn URL is what makes the check reliable, because Clay's person enrichment reads the current role from that profile. Clay's job change features work from the same input: the "Track Job Change Status" managed function needs the LinkedIn URL too.

Our test list was 100 chief financial officers at US public companies, 45 of whom had left since the list was made. Marking everyone as still in place would have scored 55.

The four steps the top runs followed

The most accurate runs, from Factory with Kimi K3, Claude Code with Sonnet 5, OpenCode with Kimi K3, Grok CLI with Grok 4.6 and Claude Code with Opus 5.5, all used the same method:

Run Clay's Enrich Person and Find Contact Details function once per LinkedIn URL. All but one of the top runs made 101 calls for 100 people, which works out to about 14 credits per person.
Compare the current company with the company on file. A different company means the person has moved. A new title at the same company counts as staying.
Write the new company and title for everyone who moved. In every run that caught 42 moves, Clay returned the new company for 34 of them.
Decide every row you can. Sonnet, OpenCode Kimi and Factory Kimi left 0 or 1 people undecided; Opus left 3.

Agent	Correct by run (of 100)	Clay credits by run	Minutes by run
Factory · Kimi K3	93 · 96 · 95	1,635 · 1,407 · 1,415	21.0 · 15.4 · 23.3
OpenCode · Kimi K3	93 · 93 · 95	1,404 · 1,404 · 1,406	8.6 · 13.5 · 10.3
Claude Code · Sonnet 5	94 · 94 · 92	1,409 · 1,407 · 1,404	12.1 · 12.8 · 12.1
Grok CLI · Grok 4.6	95 · 87 · 95	1,404 · 1,404 · 1,412	11.4 · 11.7 · 12.8
Claude Code · Opus 5.5	0* · 94 · 94	0 · 1,407 · 1,408	60.5 · 9.8 · 14.2
Pi · GPT-5.6 Luna	73 · 84 · 85	1,413 · 1,404 · 1,404	4.9 · 24.4 · 4.7
Codex · GPT-6 Sol	78 · 73 · 68	0 · 0 · 0	8.5 · 15.1 · 21.4

*Reached the 60-minute limit and counts as 0. Pass line 80. Minutes are measured from run start to grading.

Two habits that lift accuracy

Enrich every profile on a full list. Of the 17 runs that enriched every profile, 13 scored 92 to 96. Opus's two runs that did so scored 94, each in under 15 minutes. Free search is fine for a handful of checks; for 100 contacts, enrich.

Finish with the comparison. The credits buy the data; the comparison in step 2 turns it into an answer. The runs that scored 92 or more wrote a decision for nearly every row and caught 42 of the 45 moves.

What it costs

Item	Value in our runs
Clay credits for 100 contacts	1,404 to 1,415 in most runs
Credits per person	about 14
Time	8.6 to 23.3 minutes for runs scoring 92 or more
Job changes caught	42 of 45 in every run scoring 92 or more

Clay prices usage in Actions and Data Credits, and "if an enrichment returns no result, you're not charged Data Credits or Actions" (Clay pricing). For ongoing monitoring rather than a one-off check, Clay also offers Job Change Signals, which Clay prices at "1 action plus 0.2 data credits" per check and which start on the Launch plan. We did not test Job Change Signals in this benchmark.

For how this compares with people search on other tasks, see Should an AI agent use Clay people search or Enrich Person?

How we tested

The answer key. 100 CFOs named in 2025 SEC filings. A person counts as still there if they signed a filing for the company or reported a trade in its stock since August 2026, and as gone if the company filed their departure and nothing later.

The setup. Seven AI coding agents ran the task three times each: Claude Code with Opus 5.5 and Sonnet 5, Codex with GPT-6 Sol, Pi with GPT-5.6 Luna, Grok CLI with Grok 4.6, OpenCode with Kimi K3 and Factory with Kimi K3. Agents ran one at a time, each in a fresh sandbox with the same Clay key, the same 19 saved Clay routines and a 60-minute limit. A timeout counts as 0, and platform failures were rerun. Clay credits are measured as the change in the Clay workspace credit balance. A model judge read every log to confirm each decision came from Clay. The task is at github.com/promptingcompany/clay-bench.

Try it yourself

Export your contacts with company, title and LinkedIn URL. The Clay Agent Plugin ships Clay's skills to your agent up front, including the habits that keep credit use down; setup is one prompt (about 15 minutes), and there is no additional cost to use it. Install the Clay Agent Plugin or give your agent a Clay API key and the Clay Public API docs, then ask: "Run Enrich Person and Find Contact Details once per LinkedIn URL, mark anyone whose current company differs from the one on file as moved, and record their new company and title."

Common questions

How do I check 100 contacts for job changes with Clay? Run one Enrich Person and Find Contact Details call per LinkedIn URL and compare the current company with the one on file; the best runs got 92 to 96 of 100 right for about 14 credits per person.

How long does it take? Between 8.6 and 23.3 minutes for the runs that scored 92 or more.

Does Clay tell me where people moved? For 34 of the 42 moves the top runs caught, Clay returned the new company.

Can I do it for free with search? For a few contacts, yes. On 100, one enrichment per profile, at about 14 credits per person, decided the most rows and scored up to 96; search-only runs spent 0 credits and scored 68 to 78.