Scoring Model v3.0

Every candidate account is scored across 22 independent signals grouped into four families, plus firmographic tier qualifiers. The score is the weighted sum of all signals that fire for that account. Geography contributes when available but is never required — national, virtual, and inaugural events produce meaningful results from non-geographic signals alone.

v3.0 adds firmographic/profile signals (employee size, revenue direction, event loyalty, engagement recency, national footprint) and uses firmographic tier qualifiers (LTV, employee band, win rate) to shape sponsorship level recommendations without inflating scores.

Geographic Signals (~25% of max when state is available)

When an event has no state (virtual/national), these signals contribute zero — but the model does not break. Non-geographic signals fill in.

Intelligence & Product-Fit Signals (~30%)

Relationship & Commercial Signals (~25%)

Firmographic & Profile Signals (~20%)

New in v3.0. These signals integrate CRM and ML-derived account attributes. Fields with low coverage (LTV, AvgSpend) are used as tier qualifiers only — they shape the recommended sponsorship level but never inflate or penalize the score. Only fields with broad coverage (>50% non-null) are used as scoring signals.

Quality Floor (Not a Hard Cap)

Instead of returning a fixed number of targets, the system computes a quality floor:

Every account scoring above the floor is returned. If 12 accounts qualify, you get 12. If 47 qualify, you get 47. The "Max Targets" dropdown optionally caps the output, but defaults to "All" so you see every quality target.

This eliminates two problems: (1) padding with low-quality filler to reach a fixed count, and (2) artificially restricting the pool when many accounts genuinely qualify.

Tier Assignment (Signal-Driven + Firmographic Qualifiers)

Sponsorship level recommendations are signal-driven, not quota-driven. An account qualifies for a tier based on the signals it possesses. v3.0 adds firmographic tier qualifiers that serve as confirming evidence:

Firmographic qualifiers only widen eligibility (additional OR conditions). They never override signal absence — an enterprise company with zero intelligence signals still gets Patron.

Explanation Enrichment (Shown, Not Scored)

The following fields appear in the explanation text for sales context but are not used in scoring. This avoids double-counting data already captured by other signals or penalizing prospects with sparse data:

Candidate Pool Sources

The candidate pool is assembled from seven sources. Geographic sources (3-4) contribute when a state is available; non-geographic sources (1, 2, 5-7) always contribute:

1. Churned sponsors — accounts that sponsored this series previously but not this year
2. Product family cross-sell — accounts buying other events in the same family (e.g., other EE events)
3. State awards — accounts with government contracts in the event state
4. Adjacent state awards — accounts with contracts in neighboring states
5. BI tag match — accounts whose inferred business tags match event session topics
6. ML-predicted high-value — accounts with predicted revenue > $50K regardless of geography
7. Open pipeline — accounts with active CRM opportunities (active commercial relationship)

Known Limitations

• Inaugural events: No historical sessions, no audience profile, no churned sponsors. The model falls back to product family + ML + pipeline + firmographic signals. Session topic matching only works if future sessions are loaded via the Sessions ETL.
• Virtual/national events: Geographic signals score zero but National Award Footprint fills in. Targets come from product-fit, commercial, and firmographic signals.
• Session data freshness: Future session data depends on the Sessions ETL having been run recently. If sessions aren't in Salesforce yet, session topic matching uses only historical data.
• BI tag coverage: ~4K accounts have BI tags. Accounts without BI profiles cannot be matched via session topics or sector match.
• Adjacent states: Only 20 states have defined adjacency. Others get no regional pattern signal.
• Firmographic sparsity: Employee count is unknown for ~44% of accounts, LTV/AvgSpend are populated for ~7%. These fields are designed as "bonus only" — missing data yields zero contribution, never a penalty.
• Revenue direction skew: ~97% of accounts are "Stable." Growing/Declining labels come from revenue trend analysis — the signal is valuable when present but fires rarely.

Risk-Aware Integration (v3.0)

v3.0 integrates firmographic and CRM data with deliberate guardrails to avoid the most common scoring model pitfalls:

This Report's Diagnostics