Snowflake
How AI Told Them Who Was Ready to Buy—Before Their Reps Did
The Results That Redefined ABM
High vs. low propensity accounts
Spent on low-intent accounts
Average deal size increase
Sales-qualified pipeline
The Challenge: Working Hard on the Wrong Accounts
Snowflake: The Cloud Data Platform Everyone Wants
Snowflake is one of the fastest-growing enterprise software companies in history. Their cloud data platform is used by thousands of companies worldwide.
The product is strong. The brand is recognized. The market is massive.
But their ABM team had a problem: They were working thousands of accounts every quarter—and most of them were never going to buy.
The ABM Budget Problem Nobody Talks About
Snowflake's ABM team was hardworking, talented, and well-funded. But they had no scientific method to prioritize.
💡 The Realization
The question wasn't "How do we work harder?"
It was "How do we know where to focus?"
The problem wasn't execution. It was prioritization.
38% of their budget was being spent on accounts that would never convert—not because the team was bad, but because they didn't know which accounts were ready.
The Solution: AI-Powered Propensity Models
Snowflake built a predictive AI model using Snowflake Cortex AI that scored every account by its likelihood to respond to outreach and book a meeting.
The model didn't replace the sales team. It told them who to call, when to call, and what to say.
The Meeting Propensity Model
The model combined four data signals into a single prioritization score for every account:
Which pages did they visit? How long did they stay? Did they download resources? Watch demos?
Signal: Someone who watched a 20-minute product demo is more ready than someone who bounced from the homepage.
What topics are they researching across the web? Are they reading about "cloud data warehouses" or "Snowflake alternatives"?
Signal: A company researching "migrate from on-prem to cloud data warehouse" is in-market.
Are decision-makers from that account engaging with Snowflake's LinkedIn content? Liking posts? Commenting? Sharing?
Signal: A VP of Data Engineering who likes 3 Snowflake posts in a week is paying attention.
What do past winning deals have in common? Company size? Industry? Tech stack? Previous interactions?
Signal: Accounts that look like past winners are more likely to convert.
The Propensity Score
Every account got a score from 0-100 based on these four signals.
- • High-Propensity Accounts (80-100): Maximum budget allocation, priority SDR outreach, personalized campaigns
- • Medium-Propensity Accounts (50-79): Standard ABM treatment, automated nurture sequences
- • Low-Propensity Accounts (0-49): Minimal budget, removed from active outreach, moved to long-term nurture
Budget and SDR capacity were reallocated toward high-score accounts—and pulled back from low-propensity ones.
The Results: 2.3x More Meetings, 150% Pipeline Growth
Sales-Qualified Pipeline Growth
By focusing on accounts that were actually ready to buy
Meeting Efficiency
High-propensity accounts converted at 2.3x
- 2.3x lift in meetings booked vs. low-propensity accounts
- SDRs spent time on accounts that actually responded
- Faster time from outreach to meeting
- Higher show-up rates for booked meetings
Budget Optimization
38% reduction in wasted spend
- 38% less budget spent on low-intent accounts
- Budget reallocated to high-propensity accounts
- Higher ROI on ABM campaigns
- More efficient use of SDR time
Deal Quality
Bigger deals from better targeting
- +80% increase in average deal value
- High-propensity accounts had larger budgets
- Better fit = faster sales cycles
- Higher win rates on qualified pipeline
The AI vs. Human Copywriting Test
A/B Test on LinkedIn Ads
Snowflake didn't just use AI for targeting. They tested it for copywriting too.
Human-Written Ad
Professional copywriters crafted ads based on best practices and brand guidelines.
AI-Generated Ad
AI analyzed thousands of high-performing ads and generated copy optimized for engagement.
AI didn't just match human copywriters—it beat them by 54%. Not because humans are bad at writing, but because AI can analyze patterns across thousands of ads that no human could process.
How the AI Model Actually Worked
1. Data Collection & Integration
Snowflake pulled data from multiple sources into their data warehouse:
- • Website analytics (Google Analytics, Snowplow)
- • CRM data (Salesforce)
- • Marketing automation (Marketo)
- • Intent data (Bombora)
- • LinkedIn engagement data
- • Historical deal outcomes
All data was centralized in Snowflake's own platform—making it easy to query and analyze.
2. Model Training with Snowflake Cortex AI
Using Snowflake Cortex AI, they trained a machine learning model on historical data:
- • Accounts that booked meetings (positive examples)
- • Accounts that didn't respond (negative examples)
- • All associated signals (website behavior, intent, LinkedIn, CRM)
The model learned: "What do accounts that book meetings have in common?"
Result: A propensity score that predicted meeting likelihood with high accuracy.
3. Real-Time Scoring & Prioritization
Every account in the CRM got scored in real-time as new data came in:
- • Account visits pricing page → Score increases
- • VP of Data likes LinkedIn post → Score increases
- • Intent data shows research on competitors → Score increases
- • No engagement for 30 days → Score decreases
SDRs saw updated scores in Salesforce every morning—telling them exactly who to prioritize.
4. Budget Reallocation Based on Score
Marketing and SDR resources were dynamically allocated based on propensity scores:
- • High-Propensity (80-100): Personalized outreach, premium ad placements, direct mail, executive engagement
- • Medium-Propensity (50-79): Standard ABM campaigns, automated email sequences
- • Low-Propensity (0-49): Minimal spend, long-term nurture only
This is where the 38% budget savings came from—pulling back from accounts that weren't ready.
The Key Takeaway
AI Didn't Replace the Sales Team—It Made Them Smarter
The 38% of budget previously spent on low-intent accounts wasn't wasted on bad execution.
It was wasted on the wrong accounts.
- ✅ Told SDRs who to call (high-propensity accounts)
- ✅ Told them when to call (when signals spiked)
- ✅ Told them what to say (AI-generated copy that converted 54% better)
Fixing that one input—prioritization—changed every downstream output:
- • More meetings booked (2.3x lift)
- • Bigger deals closed (+80% average deal value)
- • More pipeline generated (+150% growth)
The difference between companies that buy AI tools and companies that deploy them correctly is one well-designed model.
How I Apply AI-Powered Targeting Today
AI for MENA B2B Markets
Snowflake's model works globally—but needs localization for MENA
- Intent data sources differ (Bombora has limited MENA coverage)
- LinkedIn engagement is strong in Gulf markets
- Website behavior tracking works universally
- CRM data quality is often the biggest gap in MENA
Start with what you have: CRM data + website behavior. Add intent data as you scale.
You Don't Need Snowflake's Budget
The principles work at any scale
- Small companies: Start with simple lead scoring in HubSpot/Salesforce
- Mid-market: Add intent data (Bombora, 6sense, Demandbase)
- Enterprise: Build custom models like Snowflake
- The logic is the same: prioritize based on signals, not gut feel
Don't wait for perfect data. Start scoring accounts with the signals you have today.
The 38% Budget Waste Exists Everywhere
Most B2B companies are spending on the wrong accounts
- SDRs call accounts alphabetically or by territory
- Marketing sends the same campaigns to everyone
- No differentiation between ready-to-buy and not-ready
- Budget is spread evenly instead of concentrated on high-intent
Audit your ABM spend. How much is going to accounts that will never convert?
AI Copywriting Is Real
54% CTR lift isn't luck—it's pattern recognition
- AI analyzes thousands of high-performing ads
- It identifies patterns humans can't see
- It generates variations faster than any team
- A/B testing proves what works
Test AI-generated copy against your best human-written copy. Let data decide.
Propensity Models Need Four Signals
One signal isn't enough. Four signals create accuracy.
- Website behavior (what they do on your site)
- Intent data (what they research elsewhere)
- Social engagement (LinkedIn, Twitter)
- Historical patterns (what past winners looked like)
If you only have one signal, start there. Add more signals as you mature.
Ready to Stop Wasting Budget on the Wrong Accounts?
If you're a B2B company spending on ABM, outbound, or demand gen—and you're not using propensity scoring—you're probably wasting 30-40% of your budget.
Here's how to fix it:
- • How are you prioritizing accounts today?
- • What signals are you using (if any)?
- • How much budget goes to low-intent accounts?
- • Start with CRM data + website behavior
- • Add intent data if budget allows
- • Score accounts 0-100 based on signals
- • Pull back spend from low-intent accounts
- • Double down on high-propensity accounts
- • Measure lift in meetings, pipeline, and deal size
Want Similar Results for Your Business?
Let's discuss how Enterprise sales strategies can help you achieve breakthrough results
ask@mo-yf.me
Best for detailed inquiries
Free Consultation
Book 30-min call
Discuss your project
Connect professionally
Professional networking
✓ Free consultation • ✓ 24h response time • ✓ Arabic & English