Quick answer
Inbound lead scoring in Clay works by running cheap checks first, title, employee count, geography, and only letting a lead reach the expensive step, an AI agent reading the company's website, once it has cleared every cheap filter. Get that order right and the workflow qualifies a lead and alerts sales in under a minute, without burning credits on people who were never going to pass.
I've booked over 2000 meetings for B2B clients running cold email and LinkedIn outbound, and a growing share of that pipeline now starts with a lead scoring table like the one I'm walking through here, built in Clay for a batch of form fills from a recent event. I recorded a video showing the full build. This is the written version, built around the one decision in it that actually matters: not which AI model does the qualifying, but the order the steps run in.
What does a Clay inbound lead scoring workflow actually do?
It takes a raw signal, a form fill, a resource download, a demo request, and turns it into a routed decision without a human touching it first: enrich the person, check whether they're the right kind of buyer, check whether their company fits, research the company if it does, and hand sales a short dossier instead of a name and an email address. The leads that don't qualify don't get dropped either. They go into a slower nurture track instead of sitting in an inbox nobody checks.
The table I built for this started from ten names and emails collected at a small Clay meetup I ran in Riga. That sample size is tiny on purpose: it's small enough to walk through every step by hand and see exactly where leads fall out, and the logic scales to a few hundred a week without changing.
Step 1: capture the lead without a form
The entry point is a webhook. Whatever collects the lead, a landing page, a gated asset, a signup, sends the full name and email into Clay the moment someone submits it. Clay then splits the full name into a first name right away, since any LinkedIn message or email that follows uses the first name only. That's a free step and it costs nothing, but skipping it is the kind of small thing that makes an otherwise automated workflow look hand built when a message goes out addressed to "Jack Brown" instead of "Jack."
Step 2: find the LinkedIn profile with a waterfall
From a name and an email alone, the next step finds the person's LinkedIn profile using a waterfall: several data providers queried in sequence, where the second only runs if the first comes up empty, the third only runs if the second does, and so on. In the ten person sample, every provider combined found all ten profiles. The part worth noticing is the cost logic, not the hit rate: a waterfall only charges for a successful lookup, so if the first provider finds the profile, nothing downstream runs or gets billed for that row. I go deeper into how that provider ordering affects your bill in Clay's 2026 pricing overhaul.
Step 3: enrich the person, and check they still work there
Once a profile is found, a second enrichment step reads it for the fields that actually matter: full name, current title, current company, and a current-employer flag. That last field is easy to skip and shouldn't be. People change jobs constantly, and a lead scoring table that enriches someone's old employer wastes the next five steps on a company the lead doesn't work at anymore. The step also pulls the company's domain and LinkedIn page, since both feed the company-level research that comes later.
Step 4: qualify on title before anything else runs
This is the first real filter, and it runs on a title alone, nothing about the company yet. An AI step reads each title and decides whether it belongs to a senior decision maker or a senior sales role, the two profiles that matter for this particular offer. In the ten person sample, five cleared that bar and five didn't. The five who didn't stop here. Nothing about their company gets looked up, because a correct title match on the wrong person is still a lead that was never going to buy.
The criteria themselves are whatever matters for your own offer, decision maker, specific department, specific seniority. What makes this step worth building is where it sits: right after the cheapest possible enrichment, and before anything that touches a company record.
Step 5: gate the expensive step on employee count and geography
Only the people who passed the title check get their company enriched: name, employee count, country. Those two fields, headcount and geography, become a second filter before the workflow spends anything on deeper research. In the video the cut was a floor of 10 employees and a requirement that the company sit in Europe, a specific fit for that offer, not a universal rule. A company with no listed employee count or no country at all gets treated as disqualified by default, since there's nothing to qualify it against.
The mechanism that makes this work is a conditional run: a column set to execute only if employee count is at or above the floor and the company's country matches the target region. Leads that fail either condition simply don't trigger the next column. Nothing runs, nothing gets billed, and the lead moves straight to nurture instead of research.
Tip. Order your gates from cheapest to most expensive, not from most important to least. A free title check belongs before a paid company lookup even if the title matters less to you than the company does, because the point of the gate is protecting the expensive step downstream, not ranking which filter you care about most.
Step 6: let an AI agent read the company's website
Everyone who clears both gates reaches the most expensive step: an AI agent that visits the company's own website and comes back with a working profile of the business, three or four likely pain points, a few plausible customer segments, a handful of discovery questions tied to those pain points, and a draft outreach line built around something specific to that company rather than a generic compliment. It's a close cousin of the research step I cover in how I personalize cold email with Clay's ICP matching, where the same kind of site-reading step builds the opening line instead of the lead score. In both cases, the agent doing the reading is only worth the credits because something cheaper already cleared out the rows that would have wasted it.
Clay sits in what I'd call the enrichment and signal agent category of AI sales tooling, not the full-motion AI SDR category: it finds, scores and researches, it doesn't write a sequence or press send on its own. That's exactly why the gating matters here. A full AI SDR platform absorbs a bad lead into its own process. A workflow built from pieces like this one only works if you decide, deliberately, what reaches the expensive piece and what doesn't. I go through that split in more detail in the three categories of AI sales tools and how to choose between them.
Step 7: turn the research into a document a rep can use
Raw fields in a spreadsheet row aren't something a rep reads before a call. The last data step pushes the company name, the pain points, the customer segments, the discovery questions and the draft outreach line into a formatted document, one per qualified lead. It's a small step, but it's the difference between a rep actually reading the research before a call and a rep ignoring a column of text they'd have to scroll sideways to see.
Step 8: notify sales, and don't drop the rest
The last step sends a message to the sales team's channel: name, title, LinkedIn, company, and a link to the document. That's the entire interface a rep needs to decide who to call first. For the leads that didn't qualify, dropping them isn't the right move either. They go into a slower nurture sequence with more resources, since a person who isn't a fit today can still become one later, and the cost of keeping them in a low-touch track is close to nothing.
| Dimension | A rep doing this by hand | This Clay workflow |
|---|---|---|
| Time per lead | Several minutes of searching, reading and judging | Roughly 40 to 60 seconds, start to finish |
| Who decides fit | Whichever rep happens to pick it up, inconsistently | The same gates, applied the same way, every time |
| Cost on a disqualified lead | The rep's time is already spent before they find out | Close to nothing, the expensive step never runs |
| What happens to non-fits | Often nothing, they sit in an inbox | Routed into a nurture track automatically |
| What sales receives | A name and a guess | A name, a qualification reason and a research document |
Why order matters more than which AI you use
It's tempting to watch this workflow and focus on the AI agent step, since it's the most impressive one to watch run. It's also the wrong place to focus. The AI research step would work exactly as well, and cost exactly as much, if it ran first, against all ten leads instead of five. The entire saving, and most of the reliability, comes from making that expensive step the last one in the chain instead of the first, so it only ever touches leads that already proved they're worth the spend. The same logic holds on the outbound side of my own work: I run email and LinkedIn as one sequence for clients, and the gating that decides who enters that sequence in the first place matters more to the result than which channel sends the first touch.
Put differently: a smarter AI step bolted onto a badly ordered pipeline still wastes money on leads it should never have seen. A cheap, well ordered pipeline with an average AI step at the end still saves most of the spend a bad order would have burned.
What does qualifying a lead in 40 to 60 seconds change?
In the video, the whole chain, from the lead hitting the webhook to the Slack message landing with a rep, runs in roughly 40 to 60 seconds. That number matters less as a speed brag and more for what it replaces: a rep manually finding the company's LinkedIn page, finding the person, reading their title, deciding if it's relevant, then reading the company site to figure out whether it's even worth a call. That's easily ten or fifteen minutes of unglamorous research per lead, done inconsistently, usually after the lead has already gone cold waiting in a queue. Compressing that into under a minute isn't really about the AI being fast. It's about a rep never seeing a lead that hasn't already cleared the questions they'd have asked anyway.
Where this workflow breaks down
The title qualification step is only as good as the job titles people actually list on LinkedIn, and plenty of senior people use vague or outdated titles that an AI check can misread. The company research step depends on a usable website: a thin homepage or a site with no real content about what the company does gives the agent little to work with, and the research it returns gets generic fast. And the whole thing depends on your gates being set correctly for your own offer. A floor of 10 employees and a Europe-only filter made sense for the specific case in the video. Copy those numbers into a workflow for a different market or a different deal size and you'll disqualify leads that were actually a fine fit, or worse, let through ones that weren't.
Key takeaways
- Order cheap filters before expensive ones. Title and employee count checks cost almost nothing and should run before anything touches a company website.
- A conditional run, gated on the previous step's result, is what stops the expensive AI research step from running on leads that will fail anyway.
- Waterfalls only bill for a successful lookup, so provider order in the chain matters as much as which providers you pick.
- Disqualified leads aren't dead leads. Route them to nurture instead of dropping them.
- The speed payoff of a workflow like this comes from the gating, not from the AI step being inherently fast.
FAQ
What is inbound lead scoring in Clay?
A Clay table that takes a raw inbound signal, like a form fill, enriches the person and their company, checks both against your criteria, and routes the result: a qualified lead with research attached goes to sales, everyone else goes to nurture.
Why score leads before routing them to sales instead of sending everything?
Because a rep's time spent judging fit on an unqualified lead is the same cost whether the lead was any good or not. Scoring first means reps only see leads that already cleared the checks they'd have run manually anyway.
Does this workflow need Clay specifically, or does the logic carry over?
Clay is what chains the enrichment, the conditional gates and the AI research step into one table, but the underlying idea, cheap filters before expensive ones, works in any tool that lets you sequence steps and skip the ones a lead has already failed.
What happens to leads that don't qualify?
They don't get dropped. In this workflow, they go into a slower nurture track with more resources, since someone who isn't a fit today can still become one later and keeping them in a low-touch sequence costs almost nothing.
How much does a workflow like this cost to run?
It depends on list volume and how many paid lookups and AI steps run per row, and that's exactly what the gating in this workflow is built to control: the expensive AI research step only runs on leads that already passed the free and cheap checks first.