Clay Credit Architecture: Your Own API Keys, Waterfall Order, and Cost per Qualified Lead

The Comparison Table Worth an Afternoon

provider role   clay-credit    direct API     winner at
                cost/call      cost/call      your volume
email finder A  ~$0.01-0.02    ~$0.003-0.006  direct
email finder B  ~$0.015        ~$0.005        direct
verification    ~$0.008        ~$0.002        direct
AI column       ~$0.005-0.02   ~$0.001-0.003  direct (own
                                              LLM key)
niche enrich    occasional     acct overhead  clay credits

METHOD: your credit price x credits per call
vs provider's price at your realistic tier.
Own keys for anything above ~20% of volume.

• •The AI column is the sneaky one: LLM-powered columns through credits cost multiples of calling the model with your own key via an HTTP column, and AI columns tend to run on every row. Own key here first.
• •Rate limits become your problem with own keys, so set Clay's concurrency sensibly and prefer providers whose limits fit your batch sizes; a stalled table costs time that partially offsets the savings if you ignore this.
• •Keep one credit-funded fallback in each waterfall even after going direct: when your key hits a limit mid-run, the fallback keeps the table moving.

Waterfall Order Is Arithmetic, Not Loyalty

For each provider, measure hit rate on a few hundred rows of your actual ICP, divide price per call by it, and sort ascending. The result frequently upends assumptions: premium providers earn their price on hard segments (senior enterprise, EU) and waste it on easy ones (US tech, where the cheap finder gets most of them anyway); geography flips orders entirely. Segment your waterfalls when the data says so, one order for US SMB plays, another for EU enterprise, and recompute quarterly because provider coverage shifts. This is the same logic as the verification waterfall applied one step upstream.

Cost per Qualified Lead: The Only Portable Number

Rows are not the product; qualified, verified, deduped leads are, and meetings are the point. Track spend at three levels: per qualified lead (total pipeline spend ÷ rows passing ICP + verification + dedupe), per positive reply, and per meeting, joined from campaign outcomes. Those numbers let you compare Clay against a data vendor, a scraper, or a done-for-you list honestly, and they surface the real optimizations: sometimes the cheapest improvement is not a cheaper provider but a tighter ICP filter that stops you enriching people who were never going to qualify.