Developing Lead Scoring Models

Your lead scoring is broken. Here's the model that predicts revenue with 87% accuracy. Most B2B companies score leads like it's 2015. ┣ Downloaded whitepaper: +10 points ┣ Attended webinar: +15 points ┗ Opened email: +5 points Meanwhile, 73% of these "hot" leads never convert. Here's what we discovered after analyzing 10,000+ B2B leads: The leads scoring highest in traditional systems aren't buyers. They're information collectors. They download everything. Open every email. Click every link. But when sales calls? ↳ "Just doing research." ↳ "Not ready yet." ↳ "Send me more info." The leads that DO convert show completely different signals: They don't just visit your pricing page. They spend 8 minutes there, come back twice more that week, then search "[competitor] vs [your company]." They're not reading blog posts. They're calculating ROI and researching implementation. Activity doesn't equal intent. And that's where most scoring models fall apart. We rebuilt lead scoring from the ground up. Instead of rewarding every action equally, we weighted four factors based on what actually predicts revenue: ┣ Intent signals (40%) - someone searching "implementation" is closer to buying than someone downloading an ebook ┣ Behavioral depth (30%) - how someone engages tells you more than what they engage with ┣ Firmographic fit (20%) - perfect ICP match or bust ┗ Engagement quality (10%) - quality of interaction matters The framework is simple. The impact isn't. We map every lead to one of four tiers: ┣ 90-100 points → Sales gets them same-day ┣ 70-89 points → Automated nurture + retargeting ┣ 50-69 points → Educational content track ┗ Below 50 → Long-term relationship building No more dumping mediocre leads on sales and wondering why they don't follow up. Results after 6 months: ┣ Sales acceptance rate: +156% ┣ Sales cycle length: -41% ┗ Lead-to-customer rate: +73% The biggest shift wasn't the scoring model. It was the mindset. 🛑 Stop measuring marketing by MQL volume. ✔️ Start measuring it by how many MQLs sales actually wants to talk to. Your automation platform will happily score 500 leads as "hot" this month. But if sales only accepts 50, you don't have a volume problem. You have a scoring problem. Traditional scoring optimizes for activity. And fills your pipeline with noise. Revenue-predictive scoring optimizes for intent and fills it with buyers. If you'd like help with assessing your current lead scoring logic, comment "SCORING" and I'll get in touch to schedule a FREE consultation.

I replaced my client's 3-person SDR team and saved 100+ hours monthly by automating lead research and scoring with Clay. We created a process that automatically researches, enriches, and scores leads based on 6 key data points. In this post, I'll show you exactly how we built this system that anyone can implement. 1. Industry targeting: Instead of settling for broad categories like "Software" or "Technology," given by LinkedIn or major data providers, we set up an AI enrichment in Clay that reads websites and LinkedIn data to output specific niches like "HealthTech," "Martech," etc., making targeting much more precise. 2. Seniority filtering: We went beyond basic titles like Director or VP. Using Clay's AI enrichment, we analyze complete LinkedIn profiles to categorize prospects into Tier 1, 2, or 3 based on actual decision-making authority. You could feed the AI model their complete LinkedIn profile like their work experience, summary, or any other data available. 3. Persona identification: For complex segmentation, we set up Clay to identify hyper-specific personas. For example, we could identify "sales leaders managing 10+ SDRs in cybersecurity companies,". 4. Headcount qualification: Clay provides accurate headcount data from company LinkedIn profiles. We use this in the lead-scoring process to prioritize accounts within the client's sweet spot. 5. Intent signals tracking: Clay's AI Agent or native integrations can get critical signals like: - Job changes/Champion movements - Recent relevant posts - Hiring activity - Expansion/funding events - Tech stack changes - Event/conference participation 6. Lead scoring: To score leads with 100% accuracy, we use all the data points above and assign scores: - We pick scoring criteria based on the client's ICP (industry, headcount, seniority) - Set up simple comparisons (ranges for company size, exact matches for industries) - Assign points based on importance (right industry = 10 points, Tier 1 decision-maker = 10 points) - Clay adds everything up automatically This gives instant clarity on which leads deserve attention first. 7. CRM integration & data enrichment: Clay pushes everything directly to the CRM: - All enriched data flows straight to HubSpot or Salesforce - Custom variables map additional research findings to correct fields - Leads get tagged by priority score - The sales team only works on qualified, high-scoring prospects - Everything stays updated automatically with scheduled runs We also set up Clay to pull existing contacts from their CRM: - Dedupe them automatically - Re-enrich and score them based on fresh data - Push back with updated priorities - Let the team focus only on prospects most likely to convert This system now handles the same workload that previously took 3 people, while also delivering higher quality leads that convert better.

🚀 𝐈𝐧𝐭𝐞𝐠𝐫𝐚𝐭𝐢𝐧𝐠 𝐏𝐫𝐞𝐝𝐢𝐜𝐭𝐢𝐯𝐞 𝐋𝐞𝐚𝐝 𝐒𝐜𝐨𝐫𝐢𝐧𝐠 𝐰𝐢𝐭𝐡 𝐆𝐀𝟒 𝐃𝐚𝐭𝐚 𝐢𝐧 𝐁𝐢𝐠𝐐𝐮𝐞𝐫𝐲: 𝐓𝐫𝐚𝐧𝐬𝐟𝐨𝐫𝐦𝐢𝐧𝐠 𝐌𝐚𝐫𝐤𝐞𝐭𝐢𝐧𝐠 𝐏𝐫𝐢𝐨𝐫𝐢𝐭𝐢𝐞𝐬 Aligning marketing and sales teams is key to growth. Predictive lead scoring with BigQuery ML and GA4 helps prioritize high-value leads, ensuring the sales team focuses on top conversion prospects. 🤔 What is Predictive Lead Scoring? Why Does It Matter? Predictive lead scoring leverages machine learning, historical data, and behavioral signals to assess conversion likelihood. Using GA4 BigQuery ML, you can create a tailored model that helps sales teams to: ✔️ Prioritize effectively by focusing on high-probability leads. ✔️ Save time by minimizing effort on unqualified leads. ✔️ Improve collaboration between marketing and sales, with clear data-backed insights. ⚙️ Step-by-Step Guide to Building a Predictive Lead Scoring Model: 1. Extract Lead Data from GA4: Start by querying GA4 data to identify meaningful user interactions such as form submissions, page views, and engagement metrics. Combine these signals with CRM data (if available) for a holistic view. 2. Prepare Data for Machine Learning: Clean and preprocess the data to include features like ✔️ Engagement signals (page views, session duration). ✔️ Conversion-related events (e.g., form submissions, purchases). ✔️ Demographics and geography (from geo parameters). 3. Train the Predictive Model with BigQuery ML: Use a binary classification model (e.g., logistic regression or boosted trees) to predict the likelihood of conversion. 4. Score New Leads in Real-Time: Once trained, use the model to assign predictive scores to incoming leads. 5. Visualize and Share Insights: Use tools like Google Looker Studio to create dashboards showing lead scores, enabling sales teams to focus on high-value leads. 📈 Business Applications of Predictive Lead Scoring 💡 Prioritize High-Value Leads 💡 Optimize Marketing Strategies 💡 Improve Sales and Marketing Alignment 🚀 Pro Tip: Continuously Update the Model - Predictive lead scoring models improve with time and data. Regularly retrain the model using updated GA4 and CRM data to reflect changing user behavior, market conditions, and campaign strategies. 🔍 Real-World Example: For a SaaS business, implementing predictive lead scoring using BigQuery ML led to: 💡 A 25% increase in conversion rates by focusing on high-value leads. 💡 A 15% reduction in sales cycle time, allowing teams to close deals faster. 💡 Better marketing ROI by identifying and amplifying successful lead acquisition channels. 🚀 Final Thoughts: Predictive lead scoring with GA4 and BigQuery ML enhances lead prioritization and fosters collaboration between marketing and sales. Embrace data-driven insights to align priorities, boost efficiency, and drive growth. #DigitalAnalytics #BigQuery #GA4 #LeadScoring #PredictiveAnalytics #MachineLearning #SQLForMarketing #MarketingOptimization

Your ICP is not a persona slide. It's a lot of things. But the first thing it is? A scoring system. Can't score a company 0 to 100 on fit? Then you don't have an ICP. You have an opinion. Here's how to build one today. Step 1. Score your best customers. Open your CRM. Top 20 accounts. Not biggest logos. Best behavior. Rate each one, 1 to 5: Revenue. Velocity. Time to impact. Feature depth. How easy they are to work with. Multiply. Sort. Your top 20% just showed you what ideal looks like. Step 2. Find the pattern. What do those top accounts have in common? Firmographics. Industry, size, geo. Technographics. What tools they run. Signals. What happened before they bought. 5 to 8 attributes that keep repeating. That's your scoring criteria. Step 3. Weight it. Not everything matters equally. Industry match might be 25 points. Revenue range 20. Tech stack 15. Signals 15. Here's what most people miss. Different customer types need different weights. A TripAdvisor rating predicts buying behavior for a small restaurant. Means nothing for a PE-backed chain. Multiple segments? Multiple weight models. Score out of 100. Step 4. Tier your list. Tier 1 (80+): Looks like your best customers. Tier 2 (50 to 79): Good fit. Some gaps. Tier 3 (below 50): Not now. What you do with each tier is a different post. This one is about the score. Now the hard part. The smaller you are, the narrower tier 1 should be. At €1M ARR you don't need 5.000 tier 1 accounts. You need 50. But at that stage you have less data. Maybe 15 customers, not 500. Your model is more hypothesis than proof. That's fine. Start with 10. Iterate every quarter. Step 5. Validate across the whole journey. Your scoring model is a hypothesis. Here's how you prove it. Map these cycles per tier: MQL to SQL time. SQL to Win time. Win to Onboard time. Time to first impact. Time to full impact. Those are your actual validation cycles. If tier 1 accounts move faster, onboard smoother, and reach full impact sooner, your model works. If not, adjust the weights. Check every quarter. Homework: pull your top 10 customers. Score them. What do the top 5 have in common that the bottom 5 don't? That's your scoring model v1. ← Previous: https://lnkd.in/e49kzxXS Next → https://lnkd.in/eHXJunHT

About 2-3 months back, I found out that one of my client’s page had around 570 people visiting the pricing page, but barely 45 booked a demo. Not necessarily a bad stat but that means more than 500 high-intent prospects just 'vanished' 🫤 . That didn’t make sense to me because people don’t randomly stumble on pricing pages. So in a few back-and-forth with the team, I finally traced the issue to their current lead scoring model: ❌ The system treated all engagement as equal, and couldn’t distinguish explorers from buyers. ➡️ To give you an idea: A prospect who hit the pricing page five times in one week had the same score as someone who opened a webinar email two months ago. It’s like giving the same grade to someone who Googled “how to buy a house” and someone who showed up to tour the same property three times. 😏 While the RevOps team worked to fix the scoring system, I went back to work with sales and CS to track patterns from their closed-won deals. 💡The goal here was to understand what high-intent behavior looked like right before conversion. Here’s what we uncovered: 🚨 Tier 1 Buying Signals These were signals from buyers who were actively in decision-making mode: ‣ 3+ pricing page visits in 10–14 days ‣ Clicked into “Compare us vs. Competitor” pages ‣ Spent >5 mins on implementation/onboarding content 🧠 Tier 2 Signals These weren’t as hot, but showed growing interest: ‣ Multiple team members from the same domain viewing pages ‣ Return visits to demo replays ‣ Reading case studies specific to their industry ‣ Checking out integration documentation (esp. Salesforce, Okta, HubSpot) Took that and built content triggers that matched those behaviors. Here’s what that looks like: 1️⃣ Pricing Page Repeat Visitors → Triggered content: ”Hidden Costs to Watch Out for When Buying [Category] Software” ‣ We offered insight they could use to build a business case. So we broke down implementation costs, estimated onboarding time, required internal resources, timeline to ROI. 📌 This helped our champion sell internally, and framed the pricing conversation around value, not cost. 2️⃣ Competitor Comparison Viewers → Triggered: “Why [Customer] Switched from [Competitor] After 18 Months” ‣ We didn’t downplay the competitor’s product or try to push hard on ours. We simply shared what didn’t work for that customer, why the switch made sense for them, and what changed after they moved over. 📌 It gave buyers a quick to view their own struggles, and a story they could relate to. And our whole shebang worked. Demo conversions from high-intent behaviors are up 3x and the average deal value from these flows is 41% higher than our baseline. One thing to note is, we didn’t put these content pieces into a nurture sequence. Instead, they were triggered within 1–2 hours of the signal. I’m big on timing 🙃. I’ll be replicating this approach across the board, and see if anything changes. You can try it and let me know what you think.

Is your lead scoring still stuck in the pre-AI era? Traditional lead scoring gives you a number: "This lead is a 7 out of 10." or a "Medium Fit". Clean. Deterministic. Easy to route and prioritize. But here's what I keep running into with clients: SDRs look at that "7" and have no idea what it actually means. The score works for sorting, but it fails at decision-making. -- The observation: Most scoring models combine database filters (headcount, industry) with some AI-generated attributes (intent signals, "strength of social media presence," engagement propensity). You get a weighted score. But the rationale for the score is abstracted away. Your SDR sees a 4 and a 7, knows they should call the 7 first, but has zero context for how to approach either conversation. What if lead scoring needs two layers instead of one? ↳ Quantitative score (the "7/10") - for routing and prioritization ↳ Qualitative context (the "why") - for understanding and action Keep the first layer mostly deterministic - company size, technographics, behavioral signals, AI-generated attributes, whatever your model weights. The second layer is where AI actually helps. Not by making the score "better," but by explaining it with real data: Example context block: Score: 7/10 Recent activity: - CRO posted on LinkedIn yesterday about "evaluating new sales tools" - Engineering lead attended our webinar 2 weeks ago Company signals: - Series B raised 6 months ago - Hiring 3 SDR roles in past 30 days Timing context: - Q4 budget cycle likely starts in 2 weeks - No demo requests but high research activity Override signals: - Engagement spike suggests urgency despite mid-tier score - Multi-department interest (sales + eng) suggests internal testing -- The shift this enables: 1. Agency - SDRs and agents can override when context reveals the score misses something 2. Transparency - Everyone sees the same reasoning 3. Better judgment calls - That 6-score lead who just posted about their pain point might be more valuable than the 7 who downloaded something 3 months ago -- Future state thinking: This context layer doesn't have to be static. Imagine the context is updated periodically and by real-time events. And then you give an agent decision rights based on context thresholds: "If a lead's engagement score spikes in a short period of time and they exhibit key buying signals, send personalized outreach." The agent isn't making the scoring decision. It's acting on the combination of deterministic score + contextual signals that suggest the timing is right. -- As we move to an era of abundant intelligence, we don't have to abstract away all the details and tokens. We have AI for that now. Ironically, we can now architect flows that feel less rigid and more human by removing humans from the process. Anyone else experimenting with this? What am I missing?

Hot take: Lead scoring kinda sucks. I just finished deep research into lead scoring effectiveness. 98% of marketing-qualified leads never result in closed business. And only 35% of salespeople have confidence in their companies lead scoring accuracy. Zendesk tested 800 leads: → 400 "high-score" MQLs  → 400 random leads Conversion difference? ZERO. 98% of MQLs never close. 65% of reps ignore lead scores. But here's what actually works. Scoring your TAM. And here’s how you can build this in Clay. Step 1: Define Your ICP Criteria Pull your top 20 closed-won accounts. Find the patterns: • Revenue: $10M-$100M • Employees: 50-500 • Industry: SaaS, Tech, FinTech • Location: US/Canada • Tech Stack: Uses Salesforce • Growth: Funded or 20%+ headcount growth Step 2: Build Your Scoring Model Simple binary scoring (1 = match, 0 = no match): Criteria → Points → Weight • Revenue match → 1 point × 2 = 2.0 • Employee match → 1 point × 1.5 = 1.5 • Industry match → 1 point × 2 = 2.0 • Location match → 1 point × 1 = 1.0 • Tech stack match → 1 point × 1.5 = 1.5 • Growth signals → 1 point × 2 = 2.0 Total possible: 10 points Step 3: Score Your Entire TAM in Clay Import 5,000-50,000 accounts. Example A - Perfect Fit (10/10): • $50M revenue ✓ (2.0 points) • 200 employees ✓ (1.5 points) • SaaS company ✓ (2.0 points) • US-based ✓ (1.0 points) • Has Salesforce ✓ (1.5 points) • Series B funding ✓ (2.0 points) Example B - Partial Fit (5/10): • $200M revenue ✗ (0 points) • 300 employees ✓ (1.5 points) • SaaS company ✓ (2.0 points) • UK-based ✗ (0 points) • Has Salesforce ✓ (1.5 points) • No growth signals ✗ (0 points) Step 4: Assign Tiers & Take Action • Tier 1 (8-10 points): Dedicated SDR, personalized outreach  • Tier 2 (5-7 points): Coordinated campaigns  • Tier 3 (3-4 points): Marketing automation only  • Tier 4 (0-2 points): Exclude from outbound Step 5: Layer Intent Data Add a 30% weighted Intent Score: • Website visits • Competitor research • LinkedIn content • Topic consumption Final Priority Score = (Fit × 70%) + (Intent × 30%) Most lead scoring waits for someone to download a whitepaper. TAM scoring identifies your best accounts on Day 1. Comment "TAM" and I'll send you the full report. ✌️ P.S. Even HubSpot (who sells lead scoring) admitted their own system didn't work and built something else. Mark Roberge, former CRO at HubSpot, said: "At HubSpot, we tried the lead scoring approach, but ran into [problems]. We evolved to implement an alternative approach." 

Your sales team keeps asking: "Who should I call first?" And leadership keeps answering: "... all of them?" This is the daily chaos that lead scoring solves. Here's the truth most teams miss: Lead scoring isn't about complicated algorithms. It's about answering one simple question: "Would a sales rep actually want to call this person right now?" The framework is straightforward: 1. Track what they DO (behavior signals intent) Downloaded your pricing guide? That's different than reading a blog post Visited your demo page three times? They're telling you something 2. Evaluate who they ARE (fit determines conversion potential) A VP makes buying decisions. A student is usually researching Wrong role = wasted calls, no matter how engaged they seem 3. Watch for RED FLAGS (protect your team's time) No activity in 60 days? They've moved on Unsubscribed from emails? Clear message Then create simple buckets: → Cold (under 20): Keep nurturing → Warm (21-39): Monitor closely → Hot (40+): Sales calls now The biggest mistake? Setting this up once and forgetting about it. Your buyers evolve. Your scoring needs to evolve with them. Every quarter, ask your sales team two questions: 1. Which high-scoring leads actually closed? 2. Which ones were a complete waste of time? Adjust accordingly. Lead scoring replaces guessing with a clear order of operations. It stops arguments between sales and marketing. It protects everyone's most valuable resource: time. One number tells the whole story. What's the one action that tells you a lead is actually ready to buy vs. just browsing? (besides the coveted meeting booked! 😜 ) ________ ♻️ Repost to help others + Join 25k + people receiving tips via social and my free email newsletter, sign up here: https://lnkd.in/eRXtjQ_C