Last updated September 2026.
A Clay implementation in RevOps is not a tool rollout. It is a build order: one clean account object, then waterfall enrichment, then a score the sales team trusts, then routing and CRM writeback, and only then signals and outbound. Teams that run those phases in that order get a system they can debug. Teams that start with outbound get a credit bill and a CRM nobody believes.
This is the implementation view of Clay. If you are still deciding whether to buy it, or you need the current pricing, read Clay for GTM in 2026 instead. This page assumes the decision is made and the question is what to build first.
| Phase | What you wire | What it needs from you | Where it breaks |
|---|---|---|---|
| 1. Account object | One canonical account record, dedupe rules, domain as the key | An owner for CRM data, not a committee | Two records for the same company, created by two tools |
| 2. Waterfall enrichment | Firmographics, technographics, contacts, in provider order | Deciding which fields you will actually act on | Enriching every record instead of the ones you work |
| 3. Scoring | An ICP score with visible inputs | Sales agreeing to the rubric before it runs | A black-box score reps quietly ignore |
| 4. Routing and writeback | Assignment rules, CRM push, dedupe on write | Field-level rules on what may overwrite what | Enrichment overwriting a rep's manual correction |
| 5. Signals and outbound | Hiring, funding, tech-change triggers | A written definition of what counts as a signal | Every trigger firing, so none of them mean anything |
The rest of this page covers each phase, the orchestration question of what stays in Clay and what moves to n8n, and the five failure modes we see most often.
Where Clay sits as a company, before you build on it
The vendor question comes first, because a RevOps implementation is infrastructure you will still be running in three years.
Clay raised a $115 million Series D on 9 September 2026 at a $7.1 billion valuation, led by Wellington, with Sequoia, Andreessen Horowitz, CapitalG, Meritech, DST, StepStone, Boldstart, BoxGroup, Bloomberg Beta and Evolution participating. That is more than double the $3.1 billion valuation set at the Series C in August 2025 (Clay's own announcement, BetaKit, The SaaS News). Clay reports more than 17,000 customers, including Anthropic, Google, OpenAI, Stripe, ElevenLabs, Workday and Siemens, and says it is used by 80% of the Forbes AI 50.
Revenue figures for Clay are third-party estimates rather than published numbers, and the trackers disagree: Sacra put ARR at $150 million in May 2026, up from $108 million at the end of 2025, while other trackers quote figures above $300 million. Treat the funding and valuation as firm, because they come from the round itself, and the revenue numbers as estimates.
One change matters more than the valuation for anyone budgeting an implementation. Clay repriced in March 2026, splitting usage into data credits and actions, where an action is consumed per workflow step rather than per enrichment. A ten-step workflow charges you for ten steps. The plans and what they include are in our pricing section on the GTM post, and the per-row arithmetic for a live workflow, including which of the two pools runs out first, is in what Clay costs to run across a revenue engine. It is the single biggest reason to enrich narrowly, which is phase 2 below.
What the AI adoption numbers actually say about RevOps
Adoption is no longer the interesting question. In McKinsey's State of AI survey, 88% of respondents report regular AI use in at least one business function, up from 78% a year earlier, and 62% say their organizations are at least experimenting with AI agents. The number that should shape your plan is the third one: only about a third report that their organization has begun to scale its AI programs. Almost everyone has started. Very few have finished anything.
That gap is an implementation gap, not a technology gap, and it is the reason this page is organized as a build order rather than a list of capabilities.
On the sales side, Gartner's May 2026 research found that sales organizations providing AI-enabled next best actions were 2.6 times more likely to achieve commercial growth, and that organizations prioritizing AI upskilling for sellers were 2.4 times more likely to achieve strong revenue growth (Gartner). Note what that finding is and is not. It is a correlation across surveyed organizations, not a promise that installing a next-best-action feature produces growth.
Phase 1: one account object before anything else
Clay will happily enrich a mess. It will enrich both copies of the duplicate record, charge you for both, and push both to your CRM.
So the first phase has nothing to do with Clay. Decide what a company record is keyed on, almost always the root domain, and decide who owns the rule when two records disagree. Write down which system is allowed to create accounts. If your forms, your enrichment layer and your outbound tool can all create accounts independently, you do not have an account object, you have three of them.
The test for phase 1: pick ten accounts at random and ask two people to find the current owner, the last meaningful touch, and the employee count. If the answers differ, stop here. Enrichment on top of an ambiguous object multiplies the ambiguity.
Phase 2: waterfall enrichment, narrowly scoped
This is where Clay earns its place. Instead of one provider with one hit rate, you order providers and Clay walks down the list until a field comes back, so you pay for the successful lookup rather than the subscription.
Clay reaches over 150 data sources this way, and its research agent, Claygent, handles the questions no provider has a field for: what this company sells, whether they run a self-serve motion, which competitor is mentioned on their pricing page.
What to enrich, in order of how often it changes anything:
Firmographics. Employee count, country, industry, funding stage. This is the scoring input, so accuracy matters more than breadth.
Technographics. Stack detection tells you fit and timing at once. A company that just moved off the CRM you specialize in is a different conversation from one that never used it.
Contacts. Verified work emails and titles for the roles you actually sell to. Seniority mapping is worth more than volume here.
Research fields. One or two Claygent columns that answer a question a human would otherwise open six tabs for.
The discipline that separates a working implementation from an expensive one is scope. Enrich the records you will work this quarter, not the database. Under the action-based billing introduced in March 2026, every step in a workflow bills, so a wide table run across an untriaged list is the most common way teams burn a month of budget in a week. We wrote up the hybrid approach we use to keep credit consumption predictable when volume climbs.
The vendor case studies are worth reading with the right caveat. Clay reports that OpenAI doubled inbound enrichment coverage from roughly 40% to 80% using multi-provider waterfalls, and that Anthropic tripled its enrichment coverage, approaching full coverage on inbound, by enriching and scoring leads before they sync to Salesforce (Clay customer story, OpenAI, Clay customer story, Anthropic). These are vendor-published figures from the vendor's best accounts. They show what the waterfall approach can do, not what it will do on your list.
Phase 3: a score the sales team will defend
A score is a negotiation with sales dressed up as a formula. Build it as one.
Put the inputs on a page: the fields, the weights, and the threshold. Have the sales lead argue with it before it runs on anything. Then score a sample of one hundred accounts and read the top twenty and the bottom twenty out loud with them. If the top twenty are not accounts they want, the rubric is wrong, and finding that out on paper costs nothing.
Two rules that save rework. Score on fields you enrich reliably, because a weight on a field that is blank 40% of the time is a weight on luck. And keep the score visible: a rep who can see why an account scored 82 will argue with the rubric, which is useful. A rep who sees only the 82 will ignore it, which is not.
Our longer treatment of the rubric mechanics, thresholds and routing logic lives in the Clay enrichment and lead routing guide.
Phase 4: routing and CRM writeback without the overwrite problem
Routing is the easy half: fit and territory in, owner out, with a fallback owner so nothing sits unassigned.
Writeback is where implementations quietly break. Enrichment runs on a schedule. Reps correct records by hand. Without field-level rules, the next sync overwrites the correction, the rep notices once, and from then on they keep the real data in a spreadsheet. You have paid for enrichment and taught your team not to trust the CRM.
Write the rules field by field before the first sync. Which fields may enrichment overwrite, which may it only fill when empty, and which are rep-owned and never touched. Log what changed and why, so a disputed field has an audit trail instead of an argument.
This is also where dirty data costs are real rather than rhetorical. B2B contact data decays continuously: the most commonly cited benchmark, from MarketingSherpa's database decay research as used in HubSpot's database decay simulation, is about 2.1% per month, compounding to roughly 22.5% a year, and estimates from other studies run higher, into the 30% to 40% range depending on industry and seniority. Whichever end you believe, a list you enriched once and never refreshed is materially wrong within a year.
Phase 5: signals, defined before they are wired
Hiring posts, funding announcements, tech changes and leadership moves are all easy to detect and easy to over-subscribe.
Write the definition first. A signal is a change that alters what you would say to the account this month. A company posting one engineering role is not a signal. A company posting six support roles while their careers page adds a customer operations lead is a signal for an onboarding automation conversation.
Then cap the volume. If every trigger fires, reps learn that alerts mean nothing, and the whole layer becomes noise that costs credits.
What stays in Clay, what moves to n8n
Clay's agents now handle multi-step research, conditional enrichment and message drafting inside the table, so plenty of work that used to need external orchestration no longer does.
The dividing line we use is simple. Work that belongs to a list of records stays in Clay: enrichment, research, scoring, table-shaped transformations. Work that crosses systems, or that has to run on an event rather than a table refresh, moves to n8n: form submission to enrichment to CRM to Slack, with retries and error handling around each hop. n8n ships more than 500 official integration nodes plus a large community library, and the counts you see quoted vary depending on whether community packages are included.
A worked example of the crossing-systems case, which is the one most teams want first:
Form fill. A demo request arrives. The workflow starts on the event, not on a schedule.
Enrichment. The record goes to Clay's API and comes back with firmographics, contact detail and the research fields you defined in phase 2.
Scoring and routing. The rubric from phase 3 runs, the owner is assigned, and the threshold decides between a rep and a nurture track.
Writeback. The record pushes to the CRM under the field rules from phase 4, updating the existing account rather than creating a second one.
Alert. The owner gets a Slack message with the score, the inputs behind it, and the two research fields worth reading before they call.
The value is not the speed. It is that every step is inspectable when something goes wrong, which is what separates this from the CSV-and-hope pattern it replaces. That inspectability is also why we run this layer in n8n at Ziel Lab rather than inside a closed sequencer.
AI SDRs: what the layer above actually does
With enrichment, scoring and routing in place, autonomous outreach becomes possible: immediate qualification of inbound, drafted first touches that reference enriched specifics, follow-ups that do not get forgotten.
We are deliberately not quoting reply-rate multiples here. The numbers circulating for AI SDR performance come from vendor case studies and rarely disclose the baseline, the list quality or the sample size, and a reply-rate multiple on a badly targeted list is a statement about the list.
What is defensible is the shape of the change. The agent handles the first touch and the mechanical follow-up. The human takes the conversation the moment it becomes one. The handoff is where these implementations are won or lost: the rep needs the context in the CRM before they reply, not a link to a transcript somewhere else.
What we see break
Five failure modes, in the order we run into them in RevOps engagements.
Enrichment before the object. The most expensive mistake, because it is invisible for a month. Duplicate accounts get enriched separately, scores diverge for the same company, and the fix is a merge project after you have already paid to enrich both.
A score nobody agreed to. Built by RevOps in isolation, technically sound, quietly ignored by sales. The rubric conversation takes an afternoon and prevents a quarter of drift.
Overwrite wars. Enrichment and reps writing to the same fields with no precedence rules. The symptom to watch for is reps keeping a private spreadsheet. By then trust is already gone.
Enriching the database. Running a wide table over every record because the table makes it easy. Under action-based billing this is where budgets disappear, and most of those records were never going to be worked.
Signal inflation. Every available trigger switched on in week one. Alert volume rises, meaning falls, and within a month the Slack channel is muted.
None of these are Clay problems. They are sequencing problems, which is why the build order is the deliverable and the tool is not.
Where to start
If you are at phase 0, the useful first week is unglamorous: agree the account key, name the owner of CRM data, and pick the ten fields you will actually act on. Everything after that is faster than it looks.
If you already run Clay and it is not paying for itself, the diagnostic is usually phase 3 or phase 4 rather than the enrichment layer everyone suspects. Check whether sales uses the score, and check whether enrichment is overwriting rep corrections. Those two account for most of the implementations we are asked to rescue.
We build this layer for B2B revenue teams at Ziel Lab: Clay implementation and enrichment architecture, workflow orchestration in n8n, and the CRM side that has to hold it all. If you want a second opinion on an implementation that is already running, tell us what is breaking and we will tell you which phase it is.