Account scoring with firmographic data

It's Monday. The SDR team has a list of 800 target accounts, a scoring column in Salesforce populated last quarter, and a CRO who wants to know why the top-scored accounts aren't converting faster than the bottom-scored ones. The scoring model isn't broken. It's scoring accounts the data layer can see, while the accounts that actually close sit in the gaps the model can't price. Account scoring is a ranking function, not a feature, and its output is only as honest as the data layer feeding it.

Account scoring is a data-driven method of ranking target accounts by their propensity to convert and their expected value once converted. The output is a numeric score (0-100 or a tier letter) attached to every account in the CRM, used to prioritize SDR and AE effort, ad budget, and ABM spend. Which scoring model fits your team depends on the ICP. Teams selling into LinkedIn-native enterprise and mid-market SaaS have a radically different data reality than teams selling into local-business operators, trades contractors, franchise decision-makers, multi-location restaurants, or independent healthcare groups. About 50% of decision-makers in those segments have no LinkedIn presence. One distinction shapes everything that follows: discovery (building the universe of businesses and decision-makers from scratch, especially the local operators LinkedIn doesn't index) is a different job than enrichment (filling in attributes on accounts you already know, which is what Clay, Apollo, and ZoomInfo do). Scoring models break when teams confuse the two. The rest of this article reads differently for the two ICPs, and the scoring-model recommendations diverge.

1. Account scoring vs. lead scoring

Lead scoring ranks individuals on MQL-readiness (title, behavior, role). Account scoring ranks companies on fit plus in-market-ness. They solve different routing problems. Lead scoring handles "which person in this account should the SDR call today." Account scoring handles "which accounts should the SDR work at all this quarter." Teams selling into committees (any ACV above ~$25K) need account scoring first, with lead scoring as a secondary layer inside scored accounts.

2. The three layers of an account score

Most articles name two layers (fit and intent). A working scoring model needs three: fit, in-market, and reachability. Skipping reachability is the most common cause of scoring models that look right on paper and fail in production.

2.1. Layer 1

Does the account match the ICP on static attributes? Company size, industry classification, annual revenue band, geography, ownership structure, funding stage, tech stack, franchise hierarchy or multi-unit operator status. Fit is a boolean-ish filter compressed into a score. Firmographic data is the primary input. This layer is why firmographic data quality dominates scoring-model quality for 80% of B2B motions.

2.2. Layer 2

Is the account actively evaluating solutions or approaching a buying trigger? Third-party intent (keyword-surge data, research activity on review sites). First-party intent (site visits, content engagement, demo requests from anyone at the account). Trigger events (funding rounds, exec hires, M&A, hiring velocity for relevant roles, tech-stack adds or churns). In-market scoring is where the transparent-signals-versus-black-box-score decision lives.

2.3. Layer 3

The layer most scoring models skip. A high-fit, in-market account is worth nothing if the decision-maker isn't in the data layer with verified direct-dial mobile and email. Reachability is a per-account metric: does your contact database return a decision-maker-level contact, is the mobile verified, is the email deliverable? Build this into the score as a multiplier, not a downstream filter. Teams that don't weight reachability end up with a scored list where the top 20% of accounts are unreachable on the channels the SDR team actually uses, and the score loses credibility with the SDR team inside one quarter.firmographic data.

Effective coverage (coverage × accuracy) is the right way to think about reachability inside the score. 60% coverage at 83% accuracy = ~50% effective coverage. 20% coverage at 70% accuracy = ~14% effective coverage. The SDR team works accounts that are both worth working and actually workable.

3. How firmographic data feeds the scoring model

3.1. The high-weight firmographic attributes

Company size (employee count plus revenue band) is the most predictive attribute in most B2B scoring models. Industry or vertical classification (NAICS or SIC for enterprise, trade classification for contractors and vertical-SaaS ICPs) determines which ICP cohort the account belongs to. Geography matters for territory routing and compliance-sensitive products. Funding stage and ownership structure proxy for buying velocity and budget cycle.

3.2. The attributes most scoring models under-weight

Franchise or PE hierarchy. One legal entity can be ten operating units with ten different decision-makers. Trade classification (HVAC, plumbing, electrical vs. the 287K "Contractor" gray zone). Licensing status (active vs. lapsed contractor licenses). Multi-unit operator counts. Each of these is a firmographic attribute horizontal providers under-source because they sit outside LinkedIn and corporate web. For vertical ICPs, these attributes should carry meaningful weight. For most scoring models today, they carry none because the data layer doesn't populate them.

3.3. Fit without accuracy is ranking noise

A scoring model that weights employee count 30% is only useful if employee counts are accurate. About 20-30% of firmographic records go stale within 12 months because companies raise rounds, get acquired, and shift headcount. Local-business records decay structurally faster because of higher closure rates, ownership transitions, and phone or email turnover. Effective coverage (coverage × accuracy) is the right way to think about scoring-model input quality.

4. Building the account scoring model

4.1. Start with closed-won pattern analysis

Pull the last 50-200 closed-won accounts. Tabulate firmographic attributes: size band, industry, funding stage, geography, technographic signals, franchise or multi-unit status, any vertical-specific attributes. The pattern that emerges is the fit-layer spec. Don't build the model from the ICP doc the marketing team wrote 18 months ago. Build it from the data the CRM already contains about deals that actually closed.

4.2. Weighting

Initial weights: 50% fit (firmographic match), 30% in-market (intent plus triggers), 20% reachability (effective coverage on the decision-maker). Re-weight quarterly based on closed-won attribution: which score component predicted conversion best in the last cohort? Models that never re-weight drift into irrelevance. Models re-weighted monthly overfit and thrash.

4.3. Transparent signals vs. black-box scores

Predictive scoring platforms (6sense, Demandbase) compress the three layers into one opaque number. Transparent signal stacks expose which underlying attributes drove the score, letting the SDR explain why they're calling. Black-box scores are fine as a default. Transparent signals are what lets the scoring model survive a contested QBR with the CRO.

4.4. Tiering the scored list

Three-tier framing works for almost every motion. Tier A (top 10-20%): named-account treatment, multi-channel orchestration, AE-led outreach. Tier B (next 30-40%): SDR sequences at full cadence. Tier C (remaining): nurture, ads, light-touch automation. Re-tier monthly. The point of tiering is resource allocation, not labeling. If every account gets the same treatment regardless of tier, the score isn't doing work.

In most well-calibrated scoring reviews, 60-80% of pipeline comes from the top 20% of scored accounts. When the ratio is closer to 40/60, the scoring model is under-prioritizing and the SDR team is working too flat a distribution. That's a re-weighting trigger, not a "throw out the model" signal.

5. Account scoring models

5.1. Rule-based scoring

Points assigned per attribute by human judgment. Employee count 500+ = 20 points. Series B+ funding = 15 points. SaaS vertical = 15 points. Intent-signal match = 20 points. Transparent, auditable, easy to defend. Weakness: weights drift from actual conversion patterns over time. Right choice for teams with under 12 months of closed-won data or small pipeline volume.

5.2. Predictive / ML scoring

Machine-learned models trained on closed-won and closed-lost history. Strong signal extraction when there's enough closed-won volume (typically 200+ per segment) and data cleanliness holds up. Weakness: opaque, overfits small samples, and inherits the data-layer coverage ceiling. A predictive model trained on LinkedIn-sourced firmographic data will predict well on LinkedIn-native ICPs and poorly on segments where the data layer is thin.

5.3. Hybrid scoring (rule-based floor + predictive overlay)

Rule-based scoring as the auditable backbone. A predictive overlay adjusts weights based on recent conversion data. This is where most mature B2B teams land. Balances transparency (the CRO can be shown which rules drove a score) with adaptive weighting (the predictive layer updates faster than a human can manually re-weight).

5.4. Intent-weighted scoring (6sense / Demandbase style)

In-market intent signals drive the primary score, with firmographic fit as a filter. Works well for teams with strong intent-data sources and LinkedIn-native ICPs. Structural limitation for local and SMB segments: intent data for local-business operators is thin (they're not researching on G2, not producing Bombora-indexed content-consumption signals), so intent-weighted scoring degrades for these segments.

6. The architectural question most scoring models skip

6.1. Where the scored universe actually comes from

Scoring models operate on an account universe. That universe is sourced from manual CRM upload, integration with a horizontal contact database (ZoomInfo, Apollo, Clay, Cognism, Lusha, all five), or discovery inside a vertical or discovery-first data layer. The scoring model inherits the universe's coverage ceiling. Most teams never audit this. They assume the CRM universe is "the market" and tune the model against it. The CRM is actually a biased sample of whichever data layer populated it.

6.2. The LinkedIn-dependency ceiling for non-LinkedIn-native scoring

ZoomInfo, Apollo, Clay, Cognism, and Lusha all share the same upstream architecture: LinkedIn scraping plus corporate web data. For LinkedIn-native enterprise and mid-market ICPs, this is a non-issue. Coverage is strong and the scoring model works. For local-business, trades, franchise, and independent operator ICPs, about 50% of decision-makers have no LinkedIn presence, so the horizontal data layer returns 10-20% decision-maker mobile coverage. Run a scoring model on that universe and the top-scored accounts become "the ones we happen to have data on," not "the ones most worth working." This is a source-architecture problem, not a vendor-quality problem. Switching between horizontal providers changes the invoice, not the coverage profile.

6.3. The two models of data for account scoring

Model 1 (LinkedIn-native): score accounts against a LinkedIn-dependent universe. Works for enterprise SaaS, corporate mid-market, tech-buyer motions. Model 2 (discovery-first): build the universe from non-LinkedIn sources (state licensing boards, permit filings, franchise disclosure documents, corporate filings, vertical registries) and then score. Works where Model 1 hits the coverage ceiling. DataLane is a Model 2 data layer: 17M+ US local-business locations indexed; 805K+ contractor license records with trade classifications that resolve the 287K "Contractor" gray zone; 60%+ decision-maker mobile coverage at an 80%+ accuracy floor (~83% in controlled head-to-head tests) on local and SMB segments. DataLane feeds the scored universe. The scoring model and the ABM platform keep working the way they always did.

7. Account scoring for local-business and franchise ICPs

7.1. Home services (contractors, trades, field service)

Attributes that drive the fit layer: trade classification (HVAC, plumbing, electrical, roofing, general, not the generic "Contractor" gray zone of 287K misclassified businesses), employee count, active licensing status by state (805K+ contractor license records across 50 states), service area, and permit velocity (trailing-12-month permit-filing volume as a growth-signal proxy). Horizontal firmographic providers under-source trade classification and licensing status because these sit in state licensing board records, not LinkedIn. A scoring model without these attributes lumps 287K generic "Contractors" into one bucket and misses the actual targeting cut.

7.2. Restaurants and multi-location foodservice

Attributes: POS and tech detection (Toast, Square, TouchBistro, Clover signals), franchise hierarchy (one franchisee running 12 units is a different scoring target than one single-unit operator), unit count, operator type (independent, franchise, multi-concept group), and cuisine category. About 50% LinkedIn absence on independent operators means horizontal providers can't populate operator-level decision-maker data reliably. Franchise hierarchy is the gap no horizontal provider resolves cleanly. They see one account and miss the operational structure.

7.3. Franchise operators and multi-unit businesses

Score at the operator level, not the brand level. A franchisee operating 5 locations under one brand is a different scoring target than a 50-location area developer. Franchise disclosure documents and state franchise registries carry operator-level data horizontal providers don't index. Scoring by brand name alone produces worthless rankings for multi-unit motions.

7.4. Independent healthcare groups

Less mature vertical from a data-availability standpoint. NPI registry covers individual providers but doesn't resolve group-level ownership cleanly. Scoring models for this segment are weaker across the board. Discovery-first data layers are starting to close the gap.

8. Where each scoring approach is the right choice

8.1. When rule-based scoring is the right choice

Teams with under 12 months of closed-won data, small pipeline volume, compliance-sensitive industries needing auditable scoring logic, or teams where the CRO needs to defend score rationale to finance. Rule-based scoring is the right choice for most teams under $10M ARR.

8.2. When predictive scoring platforms (6sense, Demandbase) are the right choice

Enterprise and mid-market teams with strong LinkedIn-native ICPs, meaningful closed-won volume (200+ per segment), deep intent-data budgets, and data-science capacity to validate model outputs. 6sense and Demandbase are the right choice for that profile. The scoring-model gap for local and SMB ICPs is downstream of the platform, not a platform failure. They're built for a different data reality.

8.3. When native CRM scoring (Salesforce einstein, HubSpot predictive) is the right choice

Teams already on Salesforce or HubSpot with moderate pipeline volume and no budget for a standalone predictive platform. Einstein Lead Scoring and HubSpot Predictive Lead Scoring are reasonable mid-tier options. Honest limitation: both inherit the CRM's data-layer coverage ceiling, same as any scoring system.

8.4. When a discovery-first data layer belongs under the scoring model

Teams selling into local-business operators, trades, franchise decision-makers, multi-location foodservice, or any ICP where about 50% of decision-makers have no LinkedIn presence. The scoring-model problem in these segments is a data-layer problem, not a model-design problem. A discovery-first data layer (DataLane) feeds the scored universe. The existing scoring model and ABM platform keep working.

9. Operationalizing the score

9.1. Routing

Tier A: AE-led, multi-channel orchestration, named-account treatment, ad spend, and personalized content. Tier B: SDR sequence at full cadence, phone-first with email downstream. For local-business-operator ICPs, cold calling the decision-maker's direct mobile is the highest-use channel. Reaching the owner directly on mobile avoids the gatekeeper on the business main line (the receptionist at a plumbing company, the hostess at a restaurant, the front desk at a dental office), which is where most local outbound dies. Tier C: nurture, light-touch automation, retargeting ads.

9.2. Decision-maker connect rate as the score-validation metric

Decision-maker connect rate (DM connect rate) is the operational check on whether the scoring model is working. DM connect rate is the rate at which a dial reaches the decision-maker directly, not a gatekeeper. Tier A accounts should convert dials to DM conversations at meaningfully higher rates than Tier C. When DM connect rate on scored mobile numbers doesn't vary by tier, the model is either under-weighting reachability or scoring against a universe the data layer can't actually reach. When citing any connect-rate number, always specify what's being dialed (business main line vs. verified owner mobile).

9.3. Re-scoring cadence

Monthly is the practical default. Weekly thrashes (scores swing on sample-size noise). Quarterly drifts (intent and trigger data goes stale). Re-score at month-end, re-tier accounts, and communicate tier changes to SDR and AE leads.

9.4. The SDR / ae feedback loop

SDRs working scored accounts have information the model doesn't. "This account is a parent company, the real buyer is at a subsidiary." "This contact left three months ago." "We already sold a product to their sister brand." Build a feedback mechanism (CRM flag, weekly review) so SDR intelligence updates the account record and re-scores downstream. Scoring models that don't ingest SDR feedback degrade as a function of SDR seniority. The best reps stop trusting the score first.

10. Common account scoring failure modes

10.1. Scoring a biased universe

The CRM universe was populated by whichever data provider was under contract 18 months ago. The scoring model tunes to that biased sample. "Top-scored accounts" becomes a circular definition: the accounts the model likes are the accounts the data provider happened to cover. Audit the universe before tuning the model.

10.2. Weighting attributes the data layer doesn't populate accurately

A scoring model weighting franchise hierarchy 15% is worthless if the data layer doesn't resolve franchise hierarchy. A scoring model weighting trade classification 20% is worthless if the data layer returns the 287K "Contractor" gray zone instead of specific trade designations. Weight-by-attribute and accuracy-by-attribute have to match.

10.3. Treating reachability as an afterthought

Scoring fit-and-intent, then filtering for reachable accounts downstream, discards the signal that the unreachable accounts were high-value. Build effective coverage into the score as a multiplier.

Frequently asked questions

Firmographic account scoring is most useful when the firmographic graph actually maps to the segment you sell into. For LinkedIn-indexed enterprise, the standard sources work. For local-business ICPs, firmographics rebuilt from licensing and registration sources produce different scores. The model is only as good as the upstream data layer.