Clay is an online spreadsheet that queries more than 150 data providers and enriches every row with artificial intelligence. My review, after more than 20 outbound campaigns run at Tandem, fits in two sentences. It is the best tool I know for prospect research and enrichment. It gets expensive very fast when used without discipline.
This article does both jobs at once. It gives the usage verdict, with what I kept and what I dropped. And it walks through the full tutorial, from the first source to the message pushed into the sequencer, with real pricing taken from the vendor in October 2026.
What exactly is Clay?
Clay looks like a spreadsheet. The difference sits in the columns. Each column calls a data provider, an AI model or an API, and fills the cell with the result. You start from a list of companies or people, then stack columns until you get a qualified, scored prospect ready to contact.
The provider count changes with the page you read. Clay's pricing page says more than 150 data partners, its connector page says more than 200. Both numbers were taken the same day, in October 2026. The order of magnitude is what matters here.
The company carries real weight now. Clay raised roughly 100 million euros in September 2026, at a valuation around 6.3 billion euros, and the vendor claims more than 50,000 teams. That trajectory has a direct consequence for you, pricing and billing models move fast.
How much does Clay cost in 2026?
Clay has four tiers and bills in dollars. Converted at the October 2026 rate, the Launch tier starts around 165 euros a month and Growth around 440 euros a month. Annual commitment takes about 10 percent off. Every tier, including the free one, comes with unlimited seats.
| Tier | Monthly price | Data credits | Actions | What it unlocks |
|---|---|---|---|---|
| Free | 0 euros | 100 | 500 | 200 rows per table, waterfalls, email sequencer |
| Launch | approx. 165 euros | 2,500 | 15,000 | Phone numbers, job change tracking |
| Growth | approx. 440 euros | 6,000 | 40,000 | CRM sync, API calls, webhooks |
| Enterprise | Annual quote | 100,000 and up | 200,000 and up | Clay API, data warehouse syncs, SSO, role management |
Since March 2026 two internal currencies coexist and the distinction matters. A data credit pays for data bought from a provider, from roughly 4.5 cents each. An action pays for platform work, under one cent each. Unused data credits roll over, up to twice your monthly allowance. Actions reset every month.
The tutorial, building a Clay table in five steps
A Clay workflow always follows the same backbone. You activate a source, you enrich, you find contact details, you score, you push to your tools. Here is each step and what it consumes.
Step 1, activate a prospect source
You create a workbook, then choose where rows come from. The useful sources are filtered company search, people search, physical venues through Google Maps, file import, CRM pull and webhook. This step consumes no credits.
The webhook is the source teams underestimate most. You paste an address into your website form or newsletter signup, and every new contact lands in the table, ready to enrich. Outbound prospecting and inbound requests then share one pipeline.
Step 2, enrich every row with AI
You add a column, write the goal in plain language, and a model goes looking on the web. A simple example asks for each company's Facebook page. The model receives the name, the website and the address, searches, then fills the cell. You see its reasoning and its cost row by row.
The value goes far beyond finding a page. You generate variables that exist nowhere else, such as a summary of the business, two icebreakers for an HR lead and two more for a finance director. On our tables this work runs on the fastest model in the Claude range, which is enough for that kind of short synthesis.
Step 3, find emails and phones through waterfalls
The waterfall alone justifies the subscription. Clay tries several providers in the order you set, and stops at the first hit. You only pay for attempts that succeed, and you avoid buying three databases in parallel.
Provider order decides your hit rate. On French targets, put the providers strong in Europe first. FullEnrich claims 84 percent coverage across Europe, the Middle East and Africa against 77 percent in North America, vendor figures taken in October 2026. A US provider placed first will make you pay for misses before reaching the right one.
Step 4, score prospects with a model
Classic scoring adds points from fixed rules. More than 30 employees is worth 10 points, and so on. A model does better, because it reads the real nature of the business and matches it against your value proposition.
The scoring column receives every collected variable, then returns two things. A persona, which decides which message template goes out. And a score, which decides whether the prospect goes out at all. Off-target prospects stop here, before costing a sequence slot.
Step 5, push to the CRM and the sequencer
The last step sends qualified rows to your tools. Clay connects to Salesforce, HubSpot and the modern CRMs, as well as to multichannel sequencers. The personalised message is generated in an upstream column, so it travels with the contact, already written.
This is also where Clay meets its limits. The tool fills rows. Your offer and your sequence stay yours to decide. Copy remains the number one factor, and we cover it in our article on automating LinkedIn prospecting.
Three settings that cut the Clay bill
Most runaway Clay bills come from three settings left on default. Each takes a few minutes to fix.
- Plug in your own API keys. As soon as you connect your own account at a data provider or a model vendor, Clay stops charging data credits and you pay your provider directly. Actions remain, at under one cent each. On AI enrichment volumes, the gap runs into hundreds of euros a month.
- Gate the expensive columns. Add a condition column that ticks a box on a simple criterion, such as a minimum headcount. The following columns only fire on ticked rows. You stop paying full enrichment on prospects you were going to discard anyway.
- Agree to throw rows away early. A row that returns no result costs you neither data credits nor actions. Putting the most selective filters first therefore makes the tool do your triage for free.
Clay inside ChatGPT and Claude, what the connector changes
Clay now installs directly inside a conversational assistant. The vendor publishes a connector that works in ChatGPT, Codex and Claude, documented on its dedicated page, checked in October 2026. A rep asks in plain language for the finance and IT decision makers at a given account, and the answer comes back with verified contact details.
For a management team the value sits in control. Operations teams define CRM write rules, spend guardrails and reusable workflows. Reps trigger them in natural language without being able to bypass them. The governance model resembles the one we describe for putting AI agents into production.
The trade-off deserves stating. A connector that writes into your CRM from a conversation widens the exposure surface, and that needs framing like any agent rollout. We detailed the guardrails in our article on AI agent security in the enterprise.
Where your data lives, and what GDPR says
Clay publishes the full list of its subprocessors, which few vendors in this category do. Of the 34 names published in October 2026 in its trust centre, 31 are established in the United States. Mistral is the only one established in the European Union, Cohere is Canadian and Zenrows Spanish.
Clay is SOC 2 Type II certified, states GDPR and CCPA compliance, and carries ISO 27001 and ISO 42001 certifications. A signed data processing agreement is available on request from its security team. A transfer outside the European Union remains possible, provided you document it in your register.
On prospecting, the French rule deserves a reminder. The CNIL allows email prospecting towards professionals on legitimate interest grounds, as long as the approach relates to their role, they have been informed and they can object easily. Prior consent, on the other hand, is required towards consumers.
Should you buy Clay or build your own enrichment?
The question comes up on every engagement. A technical team looks at Clay, sees a spreadsheet calling APIs, and concludes it could do the same in house. Six months later the project half exists and nobody uses it.
I laid out that choice in detail in my LinkedIn post on the buy versus build trade-off, in July 2026. The market picture is clear. According to the fund Andreessen Horowitz, around 90 percent of 2025 customer support pilots ran on a third-party application rather than on in-house development. AI-native vendors iterate faster than any internal team on this scope.
Building still makes sense in two cases. Your business logic is your competitive edge and no vendor covers it. Or your data is too sensitive to leave. In every other case, buy the tool and spend your energy on the offer and the message. That is exactly the trade-off we apply on our autopilot deployments.
What Clay users actually say
Clay scores 4.6 out of 5 across 234 reviews published on G2, checked in October 2026, including 194 top ratings. The recurring praise covers integrations and the ease of chaining enrichments without leaving the interface.
The recurring complaint is the learning curve, cited in around fifteen reviews. My experience confirms it. The first two tables get built slowly, and you have to accept throwing them away. From the third one, the same skeleton duplicates in minutes.
The second complaint heard on the market concerns unpredictable bills. It held before the March 2026 rework, which separated data from platform work and made the reading simpler. If budget remains your first criterion, compare with the alternatives in our comparison of AI tools for sales teams.
Is Clay worth its price for a European SME?
Yes, on three conditions. You have real prospecting volume, at least a few hundred contacts a month. Someone on the team owns the tables, because a powerful tool without an owner produces invoices and nothing else. And your target is professional, which places you under legitimate interest.
No, in two situations. If you sell to fewer than fifty named accounts a year, manual research on each account stays more profitable. And if your offer or your message are not yet stable, Clay will mostly accelerate your unsubscribe rate.
What I keep after more than 20 campaigns are the provider waterfall and model-based scoring, which save a considerable amount of time. What I dropped are decorative enrichments, those columns added because they were available and which never changed a decision. For the logical next step, look at where those rows land, in our comparison of AI-powered CRMs, and the full field report in what I learned supporting sales teams.
