The Short Version
The best Clay lead scoring models separate fit from intent, expose their logic through reason fields, and push actionable outputs downstream. The goal is not a clever formula. It is a system your team can inspect, debug, and trust enough to act on.
I have seen dozens of Clay scoring implementations. The ones that work share a pattern: they treat fit and intent as separate questions, they make the logic visible, and they focus on outputs that change behavior rather than calculations that look impressive in a demo.
The ones that fail share a different pattern. They pile every enrichment field into a single score, hide the weights in a formula nobody remembers, and produce numbers that reps learn to ignore because they cannot figure out why a lead scored high or low.
This guide covers how to build scoring that works. Not the fanciest approach, but the one that actually gets adopted.
Why You Need to Separate Fit from Intent
Every lead scoring model answers two different questions. Fit asks: "If this person engaged with us, would they be a good customer?" Intent asks: "Is this person actively looking right now?"
These questions have completely different implications for what you do next. A high-fit, low-intent account is perfect for proactive outbound. A low-fit, high-intent lead might be worth a quick call, but you should not build your quarter around it. When you fuse these into one number, you lose the ability to make that distinction.
I worked with a team that had a single composite score. A Series B fintech company with 200 employees and an SDR visiting their pricing page three times scored 78. A 50-person agency in an industry they never closed, with a founder who downloaded an ebook, also scored 78. The sales team had no way to tell them apart. They started ignoring the score entirely.
The fix was simple. We split the model into a fit score and an intent score. Now the rep sees: "Fit: 92 (Series B, 200 employees, Salesforce stack). Intent: 65 (3 pricing page visits this week)." They know exactly what they are looking at and why it matters.
If a rep cannot explain why a lead scored high after looking at the record for 30 seconds, your model is too opaque. Scoring should illuminate, not obscure.
Build the Model in Layers
A clean Clay scoring architecture has four distinct layers. Each one builds on the previous, and each one is inspectable on its own.
The Four-Layer Scoring Architecture
These are the fields that define whether someone matches your ideal customer profile. Industry, employee count, revenue band, geography, funding stage, tech stack. Use waterfall enrichment to maximize coverage across providers like Clearbit, Apollo, and PeopleDataLabs. Each field should have explicit thresholds, not hidden weights.
These are the behavioral signals that indicate timing. Website visits, pricing page clusters, demo requests, content downloads, webinar attendance, job changes, hiring patterns. Intent signals decay over time, so track when they occurred, not just whether they occurred.
Every score should be paired with a plain-language explanation. "High fit: Series B, 180 employees, uses Salesforce and HubSpot." "High intent: Visited pricing page 4x this week, VP Sales title." This is what makes the score actionable instead of mysterious.
Sync only what your downstream tools need. Fit score, intent score, priority tier, owner assignment, and the reason fields. Do not flood your CRM with every intermediate calculation. The output layer should be minimal and immediately useful.
This architecture makes debugging straightforward. When conversion drops, you can isolate whether the problem is bad fit logic, stale intent signals, or data quality issues upstream. Teams that skip this structure end up tweaking one mysterious number without ever finding the root cause.
Clay Implementation: Templates and Field Mappings
Clay gives you the tools to build this architecture cleanly. Here is how to structure it.
Organize with Column Groups
Create separate column groups for each layer. One group for raw enrichment data, one for fit inputs and calculations, one for intent signals, one for reason fields, and one for final outputs. This keeps the table navigable and makes it obvious where each piece of logic lives.
Use Formula Columns for Scoring Logic
Clay's formula columns let you build explicit, inspectable scoring rules. A fit score formula might look like this:
// Employee count scoring
IF(employees >= 100 AND employees <= 500, 25,
IF(employees > 500 AND employees <= 2000, 20,
IF(employees > 2000, 10, 5)))
// Plus industry scoring
+ IF(industry IN ["SaaS", "Fintech", "B2B Software"], 25, 10)
// Plus tech stack scoring
+ IF(tech_stack CONTAINS "Salesforce", 15, 0)
+ IF(tech_stack CONTAINS "HubSpot", 10, 0)The key is that every threshold is explicit. No hidden weights, no ML black box. When something seems off, anyone on the team can read the formula and understand what is happening.
Build the Reason Field with AI
Use Clay's AI column to generate human-readable explanations. Feed it the raw fit and intent data and prompt it to summarize: "In 2-3 sentences, explain why this lead scored [fit_score] on fit and [intent_score] on intent. Reference specific data points."
The output might be: "Strong fit: Series B fintech with 220 employees, running Salesforce and Segment. Moderate intent: VP Sales visited the pricing page twice last week but no demo request yet."
That is something a rep can read in five seconds and know exactly what they are dealing with.
Waterfall Enrichment for Coverage
Clay connects to 150+ data providers. No single provider has complete data for every record. Use waterfall enrichment to query providers in sequence until you get the fields you need. For employee count, you might try Clearbit first, then Apollo, then PeopleDataLabs. For tech stack, BuiltWith first, then HG Insights.
This maximizes coverage while controlling costs. Set confidence thresholds so you do not stop at low-quality results. A "verified" email is different from a "catch-all domain" email, and your waterfall should reflect that.
Track when each enrichment field was last updated. Company size from 18 months ago might be wildly inaccurate. Build refresh triggers for high-priority records and flag stale data in your scoring logic.
The Failure Modes That Kill Scoring Models
I have seen the same mistakes repeatedly. Here is what breaks Clay lead scoring and how to fix it.
| Failure Mode | What It Looks Like | The Fix |
|---|---|---|
| Too many inputs | The score fluctuates unpredictably. Reps do not trust it because small changes produce big swings. | Reduce to 8-12 fit criteria and 4-6 intent signals. Use only signals that correlate with actual pipeline outcomes. |
| No negative scoring | Scores only go up. Old leads with one engagement months ago sit at the top of the queue forever. | Implement score decay for intent (10% reduction per 30 days of inactivity). Add explicit negative points for disqualifying attributes like competitor domains. |
| No freshness controls | Old enrichment data keeps bad-fit leads looking good. A company that was 200 employees when you enriched them is now 50. | Track enrichment timestamps. Refresh priority records monthly. Flag stale data in the scoring logic. |
| Contact-only scoring | You score the contact without considering the account. A great title at a terrible company looks like a hot lead. | Always bring account-level fit into the model. A VP at a 10-person agency in the wrong industry should not outscore an SDR at your target account. |
| No messaging handoff | High scores still produce generic outreach because reps do not know what to say. | Attach reason fields, persona context, and relevant proof points. The score tells you who to contact; the context tells you what to say. |
The last point is where most teams stop short. Clay can solve the data assembly problem brilliantly. But a score without context still produces bad outreach. The rep needs to know not just that the lead is hot, but why they are hot and what that implies for the conversation.
Operationalize the Score
A lead score only matters if it changes what people do. The best scoring models do not stop at calculation. They drive routing, sequence entry, review queues, and rep prioritization.
Define Action Thresholds
Set explicit thresholds that trigger specific actions:
- Fit 80+ and Intent 70+: Immediate notification to AE, skip SDR qualification
- Fit 80+ and Intent 40-69: Proactive outbound sequence, AE notified for review
- Fit 60-79 and Intent 70+: SDR qualification call, assess fit gaps
- Fit below 60: Nurture sequence only, regardless of intent
The thresholds should map to actual behaviors your team will follow. Do not set thresholds you cannot enforce.
Sync to CRM with Purpose
Push the outputs to Salesforce or HubSpot in fields that are actually visible in rep workflows. A fit score buried in a custom field nobody looks at is worthless. Put the priority tier on the lead list view. Put the reason field on the record detail page. Make the score impossible to ignore.
Build Feedback Loops
Track what happens after the score. When a high-fit, high-intent lead goes closed-lost, find out why. Was the fit model wrong? Was the intent signal misleading? Was it a timing issue? Feed the outcomes back into the model.
The teams that calibrate monthly in the first 90 days, then quarterly after that, end up with scoring models that actually predict pipeline. The teams that build once and forget end up with models that drift into irrelevance.
Frequently Asked Questions
What should a Clay lead scoring template include?
A strong template separates fit signals from intent signals, uses formula columns for explicit thresholds, includes reason fields that explain why each lead scored high or low, and maps only the fields your CRM or sequencer needs downstream. Keep it to 8-12 fit criteria and 4-6 intent signals.
How do you build Clay lead scoring workflows for fit and intent?
Start by creating separate column groups for fit inputs (industry, size, tech stack, funding) and intent inputs (site visits, content downloads, hiring signals). Score each layer independently using formula columns. Combine them into a priority tier, generate reason fields with AI, and sync the outputs to your CRM.
What field mappings matter most in Clay lead scoring?
The highest-value mappings are fit score, intent score, priority tier, owner assignment, and plain-language reason fields. These give reps the context to act on a score rather than just see a number. Avoid syncing every intermediate calculation; it clutters the CRM and confuses the workflow.
Can Clay replace CRM lead scoring?
Clay is better suited for assembling data and calculating scores using enrichment from 100+ providers. The CRM remains the system of record. Most teams calculate in Clay, sync the useful outputs downstream, and let the CRM handle routing and workflow automation. It is a complementary relationship, not a replacement.
How many data points should a Clay lead score use?
Start with 8-12 fit criteria and 4-6 intent signals. More than that introduces noise and makes the score unpredictable. Fewer leaves gaps where poor-fit accounts slip through. The best models use the minimum inputs that correlate with actual pipeline outcomes, then calibrate based on closed-won and closed-lost analysis.
Conclusion
Lead scoring in Clay works because you control every input and can see exactly why each score lands where it does. No black box, no guessing what the algorithm values.
The key is keeping fit and intent separate. Fit tells you whether an account matches your ICP. Intent tells you whether they are active right now. Conflating them creates scores that are hard to interpret and harder to act on.
Build the model in layers: firmographic filters first, then behavioral signals, then a synthesis that explains the score in plain language. Calibrate against real outcomes - closed-won and closed-lost patterns will tell you which inputs actually matter. And connect the score to action: routing rules, sequence selection, prioritization. A score that sits in a spreadsheet is not a scoring model. It is a reporting metric.
