How to build an ABM scoring model from your CRM (not generic intent)

Most ABM scoring models fail because they're built on someone else's data. Tools like 6sense score accounts using broad market models that tell you a company is "in-market" but can't tell you who to talk to or why this week matters more than last week. Sales gets a hot account list they can't operationalize.

Here's what actually works: building a scoring model from your own CRM data. Your historical conversion patterns, your personas, your signals, your time-to-convert metrics. This is how you identify which accounts and contacts are genuinely worth your team's time right now.

We'll walk through exactly how to do this, what CRM fields and objects you need, how to validate your model against real opportunities, and how to handle thin data in new segments. This is the blueprint we use at UserGems and what our AI Command Center automates for customers generating 47x ROI in pipeline and 11x in revenue.

Why your CRM holds better signals than third-party intent

Generic intent data tells you that someone at a company visited a category page or downloaded a whitepaper. That's a start, but it doesn't tell you if that person can actually buy, if your product fits their use case, or if they're three months or three years from a decision.

Your CRM already knows what actually precedes pipeline for your business. Which personas appear in your winning deals. Which signals correlate with faster conversion. Which account characteristics predict higher deal values. Which buying committee structures close versus stall.

UserGems connects to your CRM and historical data to build scoring models specific to your business. We analyze every closed-won opportunity to identify patterns: account firmographics, contact personas, engagement sequences, signal combinations, and time-to-convert metrics. Then we apply those learnings to score your current pipeline and target accounts.

The difference between account-level and contact-level intelligence matters here. 6sense scores accounts but can't tell you which specific person to reach out to or what their role is in the buying process. At enterprise especially, "same account" does not mean "same people." Your AE might be talking to IT while marketing serves ads to finance. That's not coordination.

UserGems scores contacts. We surface the specific person, the specific signal, and the specific reason to reach out. Then Gem-E, our AI agent, executes on it automatically by writing personalized emails, adding buyers to sequences, and syncing LinkedIn ad audiences so sales and marketing activate the same contacts at the same time.

What CRM fields and objects are needed to build scoring?

You need three categories of data to build a scoring model that actually reflects your conversion patterns.

Account-level fields: Industry and sub-industry, employee count and growth rate, revenue and funding stage, geographic location, technology stack, and account owner and segment assignment.

Contact-level fields: Job title and seniority, department and function, contact creation date, last activity date, email engagement history, and meeting attendance.

Opportunity-level fields: Stage history and progression, close date (won and lost), deal size and product mix, sales cycle length, win/loss reason, and personas involved in the deal.

Most teams already have 80% of this in Salesforce or HubSpot. The gaps usually show up in contact-level data, where job titles are inconsistent or buying committee members aren't fully tracked.

UserGems fills those gaps automatically. We enrich every contact record with verified job title, seniority, department, and function. We identify missing personas in your target accounts and surface them with contact information already validated. Our data maintains a 95% match rate and less than 5% bounce rate on email addresses, with contact and account data refreshed biweekly and monthly.

When your scoring model is built on clean, complete data, reps trust the recommendations. When it's built on stale or incomplete data, they ignore it.

How do you identify which personas appear in winning deals?

Start by pulling every closed-won opportunity from the past 12–18 months. Export the contact roles associated with each deal. Group them by job title, seniority, and department.

You're looking for patterns: which personas appear in 80%+ of your wins, which personas show up early versus late in the sales cycle, which personas correlate with faster close times, and which personas are present in your largest deals.

For most B2B companies, you'll see a core set of 3–5 personas that consistently appear in winning deals. A VP or Director who owns the problem. A Manager or IC who will use the product daily. An Executive who controls budget. Maybe a technical buyer who evaluates implementation.

Once you know your winning personas, you can score contacts based on how closely they match those profiles. A VP of Sales at a Series B SaaS company with 200 employees might score higher than a Sales Manager at a 10-person startup, if your historical data shows VPs at Series B companies convert faster and close larger deals.

UserGems automates this analysis. Our AI Command Center connects to your CRM, identifies which personas appear in your closed-won opportunities, and builds persona profiles that feed into contact scoring. When Gem-E surfaces a new contact, you see exactly why they match your winning patterns.

The transparency matters. When reps can see why a contact was surfaced and say "yeah, that tracks," they actually use it. 6sense's AI scoring is a black box. You get an account score but no visibility into why. Reps don't trust what they can't see.

Which signals tend to precede pipeline vs. noise?

Not all signals are created equal. Some reliably precede pipeline. Others just create noise.

Here's what we see across our customer base:

High-signal events:

• New executive hires in the function you sell into
• Funding announcements (Series A–C, depending on your ICP)
• Competitor research activity (intent data showing they're evaluating alternatives)
• Repeat website visits from multiple personas at the same account
• Event attendance or webinar registration from target accounts
• Job changes into buying roles (especially past champions moving to new companies)

Low-signal events:

• Single website visit with no follow-up
• Generic content downloads
• Broad intent topics that don't map to active buying
• LinkedIn profile views with no other engagement
• Cold email opens with no reply or click

The difference comes down to specificity and recency. A VP of Sales who just started a new role and is researching your competitor is a high-signal contact. Someone who downloaded a generic ebook six months ago is not.

UserGems monitors 600+ signals and applies machine learning to identify which combinations actually precede pipeline for your business. We track 42,000+ intent topics, including competitor research activity, so you know when a prospect is actively evaluating alternatives. We de-anonymize website visitors at the contact level, so you see who is visiting your site, how often, and how recently.

Then we layer in native UserGems signals: job changes, past champions, new hires and promotions into buying roles, account movements. These are the signals that consistently drive 10–15% conversion rates in our own ABM program and 47x pipeline ROI across our customer base.

You can adjust signal weights in the UserGems platform and preview changes before saving. No data science team required.

How do you handle thin historical data in a new segment?

Entering a new market or segment means you don't have 18 months of closed-won data to analyze. Your scoring model can't learn from patterns that don't exist yet.

Here's how to build a model when historical data is thin:

Start with leading indicators instead of lagging outcomes. If you're entering a new vertical but have strong data in an adjacent market, identify the account characteristics and personas that appear in both.

Use hiring velocity as a proxy for growth. Companies that are actively hiring in the function you sell into are scaling that function. If you sell sales enablement software, track accounts that are hiring multiple AEs, SDRs, or sales managers in a short window.

Layer in external signals. Funding announcements, leadership changes, and competitor research activity work across segments. A Series B company that just hired a new CRO and is researching your competitor is a strong signal regardless of industry.

Build a separate scoring model for the new segment. Don't dilute your core model by mixing in accounts that don't match your historical ICP. Create a distinct model, track conversion rates separately, and refine the weights as you close deals in the new market.

UserGems lets you build multiple scoring models and ICPs within the same platform. You can run one model for your core mid-market segment and another for enterprise, each optimized for the personas, signals, and conversion patterns specific to that segment.

As you close deals in the new segment, the AI Command Center learns from those outcomes and adjusts scoring automatically. The model gets smarter the longer it runs.

How do you validate scoring against real opportunities?

A scoring model is only useful if it actually predicts pipeline. Here's how to validate yours:

Pull your current open pipeline. Export every open opportunity and the contacts associated with each deal. Score those contacts using your new model. Do the high-scoring contacts map to your active opportunities? If not, your model needs adjustment.

Run a backtest on closed-won deals. Take your closed-won opportunities from the past 12 months and score the contacts involved. Did your model surface those contacts before the deal closed? If your highest-scoring contacts don't appear in your winning deals, your signal weights are off.

Track conversion rates by score band. Group your scored contacts into tiers and track how many convert to meetings, opportunities, and closed-won deals. Your top tier should convert at 3–5x the rate of your bottom tier. If conversion rates are flat across tiers, your model isn't differentiating signal from noise.

Monitor rep feedback. Ask your sales team: are the contacts we're surfacing actually worth reaching out to? If reps consistently ignore high-scoring contacts, either the data is wrong or the signals don't match their real-world experience. Fix it.

UserGems surfaces this validation data automatically. Our AI Command Center shows you which signals precede pipeline, how every scoring factor compares across customers versus non-customers, and which contact scores correlate with faster conversion. You get full visibility into why every account and contact is scored the way it is.

When you can see the math, you can trust the output. And when reps trust the output, they actually use it.

Building a scoring model that learns and compounds

Most ABM programs get rebuilt every quarter because the scoring model is static. Here's the better approach: build a model that learns from your outcomes and compounds over time.

Step 1: Pull 12–18 months of closed-won opportunities. Identify the personas, signals, and account characteristics that appear in your winning deals. Build your initial scoring model based on those patterns.

Step 2: Layer in real-time signals. Connect intent data, website visitor tracking, job change alerts, and any other signals your team monitors. Weight them based on how reliably they precede pipeline in your historical data.

Step 3: Validate against current pipeline. Score your open opportunities and active target accounts. Do the high-scoring contacts map to your real pipeline? Adjust weights as needed.

Step 4: Activate sales and marketing together. Use your scored contact list to drive coordinated outbound and ad targeting. High-scoring contacts get added to sequences automatically, with emails written by Gem-E using real buyer context.

Step 5: Monitor conversion rates and refine. Track how many high-scoring contacts convert to meetings, opportunities, and closed-won deals. Feed those outcomes back into your model. Increase the weight of signals that reliably precede pipeline.

This is how UserGems operates. Our AI Command Center continuously monitors accounts, updates scores weekly, and learns from your conversion data. Teams using this approach see 10–15% of targeted accounts convert to sales-accepted opportunities, with median ROI of 47x in pipeline and 11x in revenue.

Sendoso generated 47 opportunities and over $1M in pipeline within 30 days of launching Gem-E, with 20% reply rates. Accord now sources 50–60% of outbound meetings through UserGems and Gem-E. These results come from scoring models built on real CRM data, not generic market intent.

Why contact-level scoring beats account-level intent

6sense tells you that a company is in-market. UserGems tells you who at that company to talk to and why right now is the moment to act.

That distinction matters because enterprise buying committees include 6–11 people on average. Knowing that "Acme Corp" is researching your category doesn't tell you whether to reach out to the VP of Sales, the CRO, the RevOps lead, or the sales enablement manager. And it definitely doesn't tell you which of those people can actually move a deal forward.

Contact-level scoring solves this. We score every person at your target accounts based on their persona, their signals, and their fit with your historical conversion patterns. When a VP of Sales at a Series B company joins from a competitor, starts researching your category, and matches 90% of your closed-won persona characteristics, that's a contact worth reaching out to today.

Then Gem-E takes over. It writes a personalized email using CRM context, past interactions, and the specific signals that made this person worth reaching out to. It adds the contact to a sequence in Outreach or Salesloft. It queues a task directly in your rep's workflow.

At the same time, that contact syncs to a LinkedIn ad audience. Marketing serves them messaging aligned to their signal: new role, competitor research, past champion. The rep's outbound email and the LinkedIn ad hit the same person in the same week, referencing the same moment. That coordination is what turns a 1–3% conversion rate into 10–15%.

6sense can't do this because it operates at the account level. UserGems syncs both motions to the same contacts for the same reasons simultaneously. That's the difference between insight and action.

Ready to build a scoring model that actually predicts pipeline? Book a demo with the UserGems team to see the AI Command Center and Gem-E in action.