Clay waterfall enrichment: build and check a recipe

To build a Clay waterfall enrichment workflow, define the field you need, put the proposed lookup steps in order, and test when each step should stop or continue. Check those conditions on a small sample before running a larger list. Keep the order and acceptance rules in fenbs so another person or AI can understand why the recipe works.

Clay documents how to preserve combinations of enrichment configurations as enrichment recipes↗ (opens in a new tab). That page explains saving and reusing configurations; it does not specify waterfall provider selection or automatic stopping behavior. Confirm those settings in your own table before treating the sequence as ready. This guide focuses on the decisions and tests that make that configuration reviewable.

What should the waterfall return?

Start with one output, such as an official company website. Write what a usable answer looks like before choosing any lookup step. A filled cell is not enough: a website for a similarly named business would pass a simple nonempty check and still be wrong.

For an illustrative company-domain test, require a match to the business name and location you supplied. Treat an uncertain match as something a person must review. Those are your acceptance rules, not a claim that Clay applies them automatically. Translate them into supported settings only after checking the configuration available to you.

• Required input: identify which fields a lookup needs before it can make a meaningful attempt.
• Acceptable output: describe the evidence that makes a returned value usable.
• Continue condition: explain when the next step should get a chance.
• Review condition: name the ambiguity that must stop the result from moving into other work.

The broader AI workflow library guide covers choosing reusable workflows. Here the question is narrower: can this particular enrichment sequence produce the field you need with a result you can check?

How do you choose the order of the steps?

List the candidate steps on paper first. Put the most appropriate lookup for your usual input near the beginning, then explain what the next step adds when the earlier answer is missing or unacceptable. Avoid adding more steps simply because they are available.

For each transition, write a plain sentence: “Continue if there is no accepted company domain.” That sentence exposes an important distinction between missing data, an unsuccessful request, and a result that fails your review. Decide how each case should behave before configuring a larger sequence.

Next, compare the proposed order with the configuration actually present in your Clay table. Check which inputs each enrichment uses and which output the following step reads. Do not assume that putting columns next to each other proves they run conditionally or that later work stops after a good answer.

How do you test a small sample?

Choose a few examples whose expected answers you can verify independently. Include an easy match, an ambiguous name, incomplete input, and a case where leaving the output empty is correct. Use approved test data rather than a valuable list whose results you cannot easily inspect.

1. Write the expected outcome for each sample before starting the enrichment.
2. Run the configured sequence on that sample and inspect the result of each step.
3. Check whether later steps ran when the acceptance rules said they should stop.
4. Check whether a missing or rejected answer reached the intended next step.
5. Record mismatches, change one condition, and repeat the same sample before expanding the run.

Track correctness separately from coverage. A sequence that fills more rows may also produce more incorrect matches. Keep a count of accepted, rejected, and unresolved results for your sample, without presenting that small test as proof of performance on every future list.

Save those distinctions with the recipe through fenbs. The next assistant then knows which results passed your check and which cases still need attention.

What should you check when results look wrong?

Work backward from the first wrong result. Was the input incomplete, was the returned value for another company, or did the next step read the wrong field? A final cell hides these differences. Reviewing the intermediate results makes it easier to change the right part of the sequence.

Also distinguish an intentional blank from a failure. If your rule rejects uncertain matches, an empty final result may be correct. Record that outcome as unresolved rather than quietly loosening the acceptance rule until every row contains something.

If data arrives through an event-driven connection, test the incoming fields separately from the enrichment order. The Clay webhooks guide covers that event contract and its failure checks. Keeping those checks separate helps you identify whether the problem begins before the first lookup.

How do you save a working Clay recipe?

Clay’s recipe documentation limits saving enrichment recipes to Enterprise customers. If that feature is available to you, use the documented sequence below. A recipe preserves enrichment configurations; do not treat it as a complete backup of table data or external credentials.

1. Hold Control or Command and select the columns to include.
2. Right-click the selected columns and choose Save as template.
3. Supply the requested name, description, and inputs, then select Create template.
4. To reuse the recipe, open Add enrichment, choose Templates, and select it.

Give the saved configuration a name that states its purpose. Keep the accepted input, stopping rules, and sample results alongside it in your own written instructions. After reuse, test a known example again to check that the copied configuration has the inputs it needs.

Clay’s Templates and Claybooks↗ (opens in a new tab) offer starting points and guided examples. They are distinct from the documented entitlement to save Enterprise recipes. For a general structure for your written instructions, see AI workflow templates.

How to keep your tested recipe with fenbs

Keep what worked so you can carry on from exactly where you left off, on any device and in another AI. You need the recipe’s purpose and checks, not a copy of every enriched business record.

1. Create a free account.
2. Write what the recipe should find, the input it needs, and the order you tested.
3. Save the rules for accepting an answer and continuing to the next step, with a simple example.
4. Write what passed, what failed, and the next check before a larger run.
5. Ask your connected AI to read this record before helping with the next change and save what changed before you stop.

fenbs does not run Clay enrichments or replace Clay’s saved configuration. If you only need configuration reuse inside Clay and have access to saved recipes, that may be enough. Use fenbs when the explanation and findings need to follow you between AI apps. More AI workflow best practices and the AI at work guides can help you make those checks repeatable.