The last time I briefed a VA for a prospect list, I waited 24 hours, received a CSV back with about 15 percent bad emails, spent another two hours cleaning it, and still launched a campaign that bounced more than I'd like to admit. I don't do that anymore. Clay replaced the whole process, and the output is consistently cleaner.
This is the exact workflow. Three steps, one export, and a filter pass before anything goes to Instantly.ai. Nothing complicated, nothing proprietary. You can replicate it in an afternoon.
What Clay Is, and What It Isn't
Clay is a data enrichment platform. You feed it a source (a LinkedIn Sales Navigator search, an Apollo export, a list of company domains, or a raw CSV), it pulls contact and company data from dozens of providers in a priority order, and it can write personalized first lines using AI before you hand the list to your sequencing tool.
It does not send emails. It does not manage replies. It does not replace your CRM. Clay enriches, qualifies, and personalizes. That's the lane. Knowing exactly what it doesn't do is as important as knowing what it does, because trying to stretch it beyond enrichment is where teams waste time and credits.
Step 1: Build Your Source List
The source defines everything that follows. Clay connects directly to LinkedIn Sales Navigator (import via a saved search URL), Apollo.io (API connection), and accepts CSV imports from anything else. You can also use Clay's built-in company search, which draws from its own data layer and is useful when you're starting from a domain list rather than people.
Tight filters are not optional. Broad inputs produce noisy outputs, and noisy outputs waste enrichment credits. Before I start any Clay table, I define: specific job titles I actually sell to or want to reach, company headcount range (for the campaigns I run, roughly 20 to 500 employees), the industries I can serve, and the geographies that are realistic. That usually produces a source universe of 500 to 2,000 contacts for a given campaign, which is the right size to enrich before evaluating.
If you're pulling from LinkedIn Sales Navigator, export the search URL, not individual contacts. Clay reads the URL dynamically and can refresh the list on a schedule. This is how you build a living table that updates as new people match your criteria, rather than a static snapshot that goes stale.
Step 2: Run the Enrichment Waterfall
This is where Clay earns its place in the stack. Instead of calling one data provider per contact and paying regardless of whether data comes back, Clay's waterfall logic tries providers in a sequence you define: Provider A first, fall back to Provider B if nothing is found, then Provider C. You're only charged for successful data hits, not for every attempt.
For each contact I typically enrich:
- Work email (verified): this is the non-negotiable enrichment. Sending to an unverified email list damages your domain reputation before the first reply. Clay integrates with ZeroBounce, NeverBounce, and its own verification layer.
- Current job title and company: job titles from LinkedIn can lag reality by several months. Cross-check with a second provider when accuracy matters.
- LinkedIn URL: useful for the personalization step and for any manual research before a follow-up call.
- Company size and industry: confirms the contact still fits your ICP after enrichment (people change companies).
- Recent company news: optional, but a funding announcement or product launch gives the AI step a specific angle to reference.
After enrichment, always spot-check 10 to 20 rows manually. Providers return stale data more often than their accuracy claims suggest. A title that's eight months out of date still passes verification. A few minutes of human review at this stage prevents a lot of awkward personalisation in the next one.
Step 3: Write First Lines with AI, Then Read Them
Clay has a built-in AI formula column that connects to OpenAI or Claude and runs a prompt against every row in your table. You write the prompt once; it executes at scale. The setup takes about 10 minutes. The prompt quality determines the output quality far more than the model choice does.
A weak prompt produces generic results. Something like "write a personalized opener for {{first_name}} at {{company}}" returns copy that sounds personalized but reads identically for every contact with the same title. A stronger prompt gives the model a specific angle:
"Write a one-sentence email opener for {{first_name}}, who is a {{title}} at {{company}}. Angle: this role typically deals with [specific challenge relevant to your offer]. Keep it under 18 words, no buzzwords, conversational. Do not start with 'I'."
The result is noticeably better. Not perfect, review is still required, but the variance between contacts is real and the openers don't all read from the same template.
Review at least 20 outputs before export. A poorly framed prompt produces the same weak line 500 times without you realizing it. Scale amplifies both good prompts and bad ones in equal measure. A few minutes of output review before you push to a sequencer is cheap insurance.
Before You Export, Filter It Down
Before anything leaves Clay for Instantly.ai or Smartlead, run a filter pass. Remove:
- Any contact with no verified work email
- Anyone your domain has emailed in the last six months (check your sequencer's suppression list)
- Anyone whose company size or industry doesn't match your ICP after enrichment revealed their actual details
- For account-based work: more than two or three contacts per company in a single campaign wave
This filter step typically removes 15 to 30 percent of the raw enriched list, and that's expected. The contacts leaving are the ones who would bounce, mark you as spam, or receive a message that's clearly wrong for their situation. Removing them is not waste; it's precision.
Clay has a native export to Instantly.ai and to Smartlead. No CSV download and re-upload required. Set up the integration once, point it at the right campaign, and the clean list lands in your sequencer with all the enriched fields mapped. From there, your sending domain setup, inbox rotation, and warmup history are what determine whether those emails actually reach the inbox.
Where Clay Stops and the Other Tools Start
Clay builds the list. Instantly.ai or Smartlead handles sequencing and delivery. Pipedrive, HubSpot, or GoHighLevel manages the responses and tracks the relationship forward. These three lanes are not interchangeable, and trying to stretch one tool into the others' territory shows up immediately in the data quality.
The first time you set this up it takes three to four hours: sourcing, enrichment column configuration, the AI prompt tuning, and the export integration. After the initial setup, running a new campaign off the same Clay table template takes 30 to 45 minutes. The VA equivalent, when I was doing it that way, cost a day of elapsed time and still required cleanup. The Clay version costs credits and about half an hour of my attention.
That's the tradeoff. The credits are real money. The time is yours. Most teams running consistent outbound will find the math straightforward after the first campaign.
Sources
Clay.com (platform documentation and enrichment provider list). Apollo.io for source list building. Instantly.ai for sequencing and native Clay export. ZeroBounce and NeverBounce for standalone email verification.
If you're building an outbound workflow and want to compare notes on what's working, reach out here.
Comments