Specialism
Clay automation consultant
Nothing in the tool is hard to click. What is hard is a Clay table that runs every week without spending a month of credits in a morning. Teams usually bring in a consultant at the point where the budget is gone, three columns are all guessing at the same email address, and nothing has reached the CRM in a shape a rep would trust. That is the automation work described here.
The work
What does an engagement in your workspace actually produce?
Four things, in this order, because each one makes the next worth paying for: a list worth enriching, waterfalls that stop when they should, spend that is bounded by design rather than by the invoice, and a record that lands in the CRM without creating a duplicate.
Clay (opens in new window) is a GTM data platform — GTM meaning go-to-market, the sales, marketing and revenue-operations side of the business. It looks like a spreadsheet: each row is a company or a person, and each column is either a data provider being asked a question, an AI step doing research, or a write into another system. That familiarity is the trap. A spreadsheet costs nothing to recalculate; this one bills for it.
Sourcing and deduplication. Rows arrive from an import, a CRM view, a webhook, or an HTTP request from another workflow. The only free filtering happens here, before a single paid column runs, and it is the step teams skip. We define the ICP — the ideal customer profile, the written description of who is actually worth contacting — as a filter the table applies, then match incoming rows against what your CRM already holds on domain and a normalised company name. Enriching a contact you already own is the purest form of waste there is.
Enrichment waterfall design. An enrichment waterfall is an ordered chain of data providers for a single field: the first is asked, and only if it comes back empty is the second asked, then the third. The order is the whole design. It should be set by measured hit rate on your segment rather than by list price or by whichever provider a template shipped with, and the chain needs a floor — a depth past which the row is marked unresolved and handed to a person, instead of being asked of five more providers who will also not know.
Credit control. Every paid column gets a condition on both ends: run only when the inputs it needs exist, and only when the output it fills is still empty. That single discipline stops the two most common leaks — enrichment attempted on rows that could never resolve, and columns silently re-running over rows that were answered weeks ago. Costs and the credit behaviour of each provider are published by the vendor and change over time, so we scope against what your workspace shows on the day rather than against a number on a blog post.
AI research columns. Clay's agent column (Claygent) reads sources and returns a field, which is genuinely useful for questions no provider sells — whether a company runs a specific platform, whether a location is open, what a job posting implies. It is also the fastest way to write a confident guess into your CRM. We constrain it to a fixed output shape, require a source where one is claimed, and give it an explicit not-found value, so an empty answer is a legitimate result rather than an invented one.
Clay-to-CRM handoff. The table is a workbench, not a filing cabinet. The finished record has to reach the CRM or the sequencer once, in the right shape, with one system owning each field and a queue holding anything that fails. Where the write needs real retries and a log you can inspect afterwards, we route it through a workflow platform you also own rather than firing it and hoping.
| Stage | What happens in the table | The decision that controls the cost |
|---|---|---|
| Source | Rows arrive by import, CRM view, webhook or HTTP request | Whether the row is in the ICP at all — filtering here is free, enriching first is not |
| Deduplicate | The row is matched against records the CRM already holds | Match on domain and normalised name before any paid column executes |
| Company resolution | A domain resolves to a firmographic record | Which provider leads the waterfall for your segment, by hit rate rather than price |
| Person resolution | A named contact is identified at that company | Whether a title-based search may guess when no named person exists |
| Email discovery | The waterfall walks providers until one returns an address | How deep the chain may go before the row is marked unresolved |
| Validation | The address is checked before anything is ever sent to it | What happens to catch-all domains and role addresses: send, hold, or review |
| AI research | An agent column reads sources and returns a field | A fixed output shape and an explicit not-found, so a guess never becomes a CRM value |
| Write-back | The finished record is pushed to the CRM or the sequencer | Which system owns each field, and where a failed write is parked for retry |
An enriched list is still a list of people. The moment a table starts feeding a sequencer, you are contacting strangers at volume, and the rules that apply to a person apply to the automation. Consent handling, suppression and reliable opt-out processing are design decisions in the pipeline, not a policy document — the baseline is the FTC's CAN-SPAM compliance guide (opens in new window). Deliverability has its own published bar: authentication, a visible unsubscribe and a spam-complaint ceiling are all set out in Google's email sender guidelines (opens in new window), which is why validation sits in the pipeline before the first send rather than after the first bounce report.
Engagements
Three ways teams bring us in
Build the pipeline
One list, sourced and deduplicated, enriched through waterfalls with costs bounded, and written into your CRM in a shape reps will use.
Audit the spend
An existing workspace read column by column: what is re-running, what never resolves, and which providers are earning their place.
Document and hand over
The table logic written down and the conditions explained, so your own team extends it without calling us first.
Where the table stops and the rest of the stack starts
Enrichment is the front of a longer process, and most of the value leaks out at the seams rather than inside the table. What happens to a reply, how fast a person sees it, and whether the score means anything are covered by automated lead qualification, which is the layer this feeds. The wider funnel — capture, deduplication, routing and follow-up across every source rather than outbound alone — sits on sales and marketing automation.
On the destination side, a record only stays clean if the CRM has somewhere sensible to put it. If that CRM is HubSpot, the property model, lifecycle stages and enrollment triggers need designing before any of this is switched on, and our HubSpot automation consulting page covers that work. Where the write itself needs retries, monitoring and a failure queue you can inspect, we build the orchestration in a platform you own — n8n consulting describes how those workflows are structured and handed over.
When this platform is the wrong answer
When the list is small enough to research by hand, when your data already lives in a provider you pay for and the CRM can call it directly, or when the real problem is that nobody follows up on the leads you already have. Adding an enrichment layer on top of a broken follow-up process buys you better-formatted silence. We will say that on the call rather than after the first invoice.
Questions
Practical questions about enrichment work
Are you a Clay partner or a certified Clay expert?
No, and we will not imply a badge we do not hold. We are an independent consultancy that builds inside your own workspace, using the same tables, waterfalls and integrations you already pay for. If a listed partner is part of what you are buying, that is a fair reason to hire one instead, and we will say so.
Our credits run out halfway through the month. What usually causes that?
Almost always enrichment running on rows that were never worth enriching, and columns re-running on rows that already have an answer. The fixes are unglamorous: filter to your target list before any paid column executes, guard every column with a condition on its inputs and its output, and cap how deep a waterfall may go before a row is marked unresolved.
Can Clay replace our CRM?
No, and treating it as one is where the mess starts. It is a research and enrichment surface, not a system of record: it has no ownership model, no reliable history, and no single version of a contact your whole team works from. The finished record belongs in the CRM, written once, with the table as the workbench rather than the filing cabinet.
What happens when an AI research column returns something wrong?
It is designed to be able to fail out loud. We constrain the prompt to a fixed output shape, require a source where one is claimed, and give it an explicit not-found value so an empty answer is an answer rather than an invented one. Anything unresolved goes to a review queue instead of being written into a CRM field a rep will trust.
Free consultation
Tell us what is costing you the most time
Describe the process. You get a straight read on whether it is worth automating, roughly what it would take, and what we would tackle first — on a call that costs nothing.
- Under an hour, free, and no obligation
- Fixed price per stage, agreed before any work starts
- You own everything we build, documented and in your accounts
- If you do not need us, we will say so on that call
The form is the fastest way to reach us — we reply from a real person, usually within one business day.