// PREDICTIVE MODELS · FEDERAL CONTRACTING

Score the deal before you chase it.

Federal contracting intelligence built on 12+ million award transactions, 880,000 active SAM-registered entities, and FEC political contribution cross-references. Competitive intelligence lookups available now — predictive ML scoring in calibrated-evaluation beta. API-first, explainable, priced for small-to-mid-sized capture teams.

// FLAGSHIP · v1

Federal Contractor Win Score

Given a vendor (UEI) and an opportunity (NAICS + agency + set-aside + size), we return a competitive-intelligence ranking against eligible peers — plus the top drivers behind the score. Currently in calibrated-evaluation beta: full-density information-retrieval methodology (Hit@1 / Hit@3 / MRR against the actual SAM-eligible vendor pool) is being validated before publishing accuracy claims.

Win Score v1 — XGBoost + SHAP

beta · limited seats

A gradient-boosted classifier trained on every federal prime-contract initial award from FY2024 through FY2026 (12.2M rows), with negatives sampled from NAICS-eligible SAM entities. Every prediction ships with the five largest feature contributions, so capture managers can audit the logic rather than trust a black box.

Training rows
12.2M
FY2024-FY2026 initial awards
Vendor pool
880K
SAM-active entities
Status
Beta
calibrated-eval in progress
Eval methodology
Hit@K / MRR
full SAM-eligible candidate pool
POST /v1/score/federal-win BETA · INTERNAL · external launch Q2 2026
# Request
{
  "vendor": { "uei": "KDXA3M7NFGF4" },
  "opportunity": {
    "naics_code": "541330",
    "awarding_agency_code": "097",       # DoD
    "product_or_service_code": "AC13",
    "type_of_set_aside_code": "SBA",
    "primary_place_of_performance_state_code": "TN",
    "potential_total_value_of_award": 4500000
  }
}

# Response
{
  "uei": "KDXA3M7NFGF4",
  "competitive_rank": 3,
  "of_eligible_vendors": 47,
  "raw_score": 0.72,
  # Calibrated probability suspended in beta. Use rank.
  "top_drivers": [
    { "feature": "uei_naics_win_count", "value": 23, "direction": "positive" },
    { "feature": "uei_agency_naics_share", "value": 0.14, "direction": "positive" },
    { "feature": "match_naics_primary", "value": 1, "direction": "positive" },
    { "feature": "set_aside_match", "value": 1, "direction": "positive" },
    { "feature": "match_state", "value": 0, "direction": "negative" }
  ],
  "model_version": "win_score_v1"
}
POST /v1/rank/federal-win — rank all eligible vendors for an opportunity BETA · INTERNAL · external launch Q2 2026
# Use case: contracting officer or BD researcher wants the top 20 vendors
# most likely to win a specific opportunity, filtered by state and SB status.
{
  "opportunity": { /* same shape as above */ },
  "top_k": 20,
  "filter_state": "TX",
  "filter_sb_eligible_only": true
}
How this is different from filter-stack "AI scoring." Most real-estate and GovCon tools in the SMB tier label weighted-keyword filter pipelines as "AI scoring." Our underlying technology is an actual trained gradient-boosted tree ensemble (XGBoost) with per-prediction SHAP feature attributions on top of 12.2M federal awards. What we will NOT do: publish a headline AUC number until our information-retrieval evaluation (full SAM-eligible candidate pool, Hit@1 / Hit@3 / MRR / NDCG, slice metrics by NAICS / agency / set-aside) is complete. Internal v1 binary-classifier AUC of 0.99 was identified as inflated by negative-sampling collapse during pre-launch audit; we are rebuilding evaluation rigor before quoting numbers. Beta access today gets you competitive-intelligence rankings + top-driver explanations; predictive scoring claims come back online when they survive proper scrutiny.
// FLAGSHIP · v1 · NEW

NPI Provider Stability Score

Given a healthcare provider's NPI, we predict the probability they deactivate, retire, or change practice in the next 24 months. Built for provider-network adequacy planning, M&A and portfolio diligence, and health-plan directory management — network- and transaction-level analysis, not decisions about individual providers.

Provider Stability v1 — XGBoost on 9.4M NPIs

IN STOCK · 7.9M scoreable · $14,997 · by inquiry

Trained on the full NPPES Healthcare Provider Registry — 9.4 million records — with deactivation history as the supervised label and a 2-year forward prediction window. Features include practice tenure, taxonomy mix, organizational structure, geographic stability, license state, and structural signals like authorized-official presence and parent-organization linkage.

// Free demo offer
Send us up to 100 NPIs from a network or target list — we'll return a scored CSV within 24 business hours, no commitment.
Request Free Demo →
Custom List
$1,997
up to 10K NPIs
State Batch ex-CA
$4,997
per state
National Batch ex-CA
$14,997
7.9M NPIs scored
Subscription
$49,997 /yr
quarterly refresh
Training rows
9.4M
Full NPPES registry
AUC (hold-out)
0.750
2-year forward window
Lift @ decile
4.07x
vs random baseline
Recall @ decile
40.7%
of all deactivations
POST /v1/score/provider-stability BETA · INTERNAL · external launch Q2 2026 · batch CSV today
# Request — illustrative example, synthetic identifier (not a real provider)
{
  "provider": { "npi": "1000000004" }
}

# Response
{
  "subject_id": "1000000004",
  "probability": 0.0011,
  "model": "provider-stability_v1",
  "interpretation": "Probability of provider deactivation or practice exit in next 24 months",
  "top_drivers": [
    { "feature": "years_since_enum", "value": 18.86, "direction": "negative" },
    { "feature": "practice_state", "value": 852, "direction": "negative" },
    { "feature": "taxonomy_count", "value": 2, "direction": "negative" }
  ]
}
buyersProvider-network adequacy planners, health-plan directory teams, M&A and portfolio diligence
use caseQuantify network-level attrition risk for adequacy planning, valuation, and portfolio analysis
Methodology note. Initial v1 training included a `years_since_update` feature that artificially inflated AUC to 0.96 because the NPPES Last Update Date is bumped by deactivation events themselves (label leakage). We caught this in pre-launch validation and refit without that feature. The published v1 metrics (AUC 0.75, lift 4.07x) reflect the honest signal from stable, leakage-free features only — practice tenure, geography, organizational structure, and taxonomy. Every model release ships with this kind of methodology disclosure.
Not a consumer report. Provider Stability is a network- and portfolio-analytics product built from public records. It is not a "consumer report," and Empire Data Solutions is not a "consumer reporting agency," under the Fair Credit Reporting Act. Scores may not be used, in whole or in part, as a factor in any decision about an individual provider's employment, engagement, credentialing, insurance, credit, or other eligibility.
// FLAGSHIP · v1 · NEW

FMCSA Carrier Risk Score

Given a motor carrier's USDOT number, predict the probability they will have a reportable safety incident (crash) in the next 24 months. Built for trucking insurance underwriters, freight brokers, shippers, and 4PL logistics companies.

Carrier Risk v2 — XGBoost on 2.7M USDOTs · Census + BASIC + OOS

IN STOCK · 57 features · $14,997 · by inquiry
// Free demo offer
Send us up to 100 USDOT numbers — we'll return scored CSV within 24 business hours, no commitment.
Request Free Demo →
Custom List
$1,997
up to 10K USDOTs
State Batch ex-CA
$4,997
per state
National Batch ex-CA
$14,997
2.7M USDOTs scored
Subscription
$49,997 /yr
quarterly refresh

Trained on the full FMCSA Motor Carrier Census (4.4M registered USDOTs) joined to the public Crash File (985K crashes 2021-2026), SMS BASIC scores (5 BASICs across 705K carriers — Unsafe Driving, HOS Compliance, Driver Fitness, Controlled Substances, Vehicle Maintenance), and Out-of-Service order history (387K records). Time-cutoff design: features pre-2024-01-01, label from crashes 2024-01-01 to 2026-01-01. v2 is a substantial step up from v1 — exposure-adjusted features (annual VMT, fleet size) plus FMCSA's own inspection violation patterns dramatically improve top-decile precision.

Carrier registry
2.7M
After data quality filter
Features
57
Census + BASIC + crash + OOS
AUC (hold-out)
0.927
v1 was 0.726 — under audit
Lift @ decile
5.31x
14.9% base rate (real)
Methodology audit in progress. Carrier Risk uses the same negative-sampling family as Win Score, which we are now re-evaluating with full information-retrieval methodology. Until that completes, treat the 0.927 AUC as preliminary. The underlying signal — prior crash history, inspection volume, MCS-150 mileage, Census + SMS BASIC features — is real and lift@decile is unaffected by sampling mechanics, but the headline AUC may revise downward when measured against a realistic candidate-density evaluation.
POST /v1/score/carrier-risk BETA · INTERNAL · external launch Q2 2026 · batch CSV today
# Request — illustrative example, synthetic identifier (not a real carrier)
{
  "carrier": { "usdot": "9999999" }
}

# Response (illustrative — synthetic high-crash carrier profile)
{
  "subject_id": "9999999",
  "probability": 0.9994,
  "model": "carrier-risk_v1",
  "interpretation": "Probability of reportable safety incident (crash) in next 24 months",
  "top_drivers": [
    { "feature": "prior_crash_count", "value": 4499, "direction": "positive" },
    { "feature": "had_prior_crash", "value": 1, "direction": "positive" },
    { "feature": "prior_tow_aways", "value": 4261, "direction": "positive" },
    { "feature": "years_since_last_crash", "value": 0.005, "direction": "positive" }
  ]
}
buyersTrucking insurance underwriters, freight brokers, shippers, 3PL/4PL logistics
use casePre-bind carrier vetting, lane risk pricing, contract carrier qualification
marketTrucking insurance is a $50B+ industry — risk scoring is a paid feature in every quote
v2 features. Top predictors by gain: unsafe_driving_inspection_violations (the FMCSA Unsafe Driving BASIC), prior_crash_count, total_inspections, years_since_MCS150_update, power_units, annual_VMT. The model also encodes operating status, hazmat indicator, prior revocation flag, OOS order history, and 4 BASIC categories beyond Unsafe Driving. Hazmat Compliance and Crash Indicator BASICs are CSMS-internal at FMCSA and not publicly available — v3 will add inspection violation patterns at the violation-code level (13M public violation records exist).
// HOW IT WORKS · API BETA

Three calls. Real signal.

The multi-model FastAPI scoring service is operational on internal infrastructure today (3 models loaded, SHAP-style explanations, sub-100ms response). External launch with API keys + metered billing targeted for Q2 2026. Today's commercial deliverable: scored CSVs via batch tiers.

Step 01

Identify

Provide the 12-character UEI for your firm (or a competitor) plus the 6-digit NAICS of the opportunity. Opportunity metadata can come from our SAM.gov Opportunities feed or your own CRM.

Step 02

Score

POST to /v1/score/federal-win. Response in under 100 ms with a 0-1 win probability and the top five feature contributions signed positive or negative. (Endpoint planned Q3 2026.)

Step 03

Act

Use the probability to prioritize capture effort. Use the driver list to design a no-bid / bid-solo / bid-with-partner posture. Loop the lift back into your internal pipeline.

// AVAILABLE NOW · BATCH SCORING DELIVERY

Today: scored CSV deliveries. No API needed.

Each model ships above with four tiers (Custom List $1,997 · State Batch $4,997 · National Batch $14,997 · Quarterly Subscription $49,997/yr). One-time CSV deliveries, signed DPA, GPC-eligible at the Custom List tier. Federal buyers and most commercial buyers prefer this — no integration, no per-call billing, no rate-limit surprises.

// CALIFORNIA RESIDENT RECORDS: State and National batch deliveries currently exclude California-resident records pending CA Data Broker registration (CA Civ. Code §1798.99.80 et seq.). Custom List scoring is unaffected because the customer provides their own input list. Need CA-inclusive scope? Contact us for negotiated agreement once we complete CA registration. Records remain sourced from public-record agency filings, exempt from CCPA "personal information" definition under §1798.140(o)(2).
// METERED API · BETA — EXTERNAL LAUNCH Q2 2026

Hosted scoring API · operational internally

Join early-access list →

Multi-model FastAPI scoring service is operational on internal infrastructure today — 3 models loaded (Provider Stability, Carrier Risk v2, Federal Win Score), SHAP-style per-prediction feature contributions, sub-100ms response. External launch with API key authentication + Stripe metered billing + customer self-service portal targeted for Q2 2026 (June). Early-access subscribers get priority onboarding plus first month free at the Growth tier.

Starter Q2 2026
$299/ mo
  • 5,000 scores / mo included
  • API access · /v1/score · /v1/rank
  • Weekly model refresh
  • Email support (48h)
Notify me
Pro Q2 2026
$1,499/ mo
  • 250,000 scores / mo included
  • Dedicated vendor pool (your SAM registry segment)
  • Webhook push on new high-fit opportunities
  • Quarterly model review + custom features
Notify me
Enterprise Q2 2026
Custom
  • Unlimited scoring + SLA
  • On-prem / VPC deployment
  • Custom model training on your proprietary features
  • Dedicated ML engineer time
Talk
Roadmap honesty: The multi-model FastAPI scoring service is operational on internal infrastructure today (3 models loaded, SHAP per-prediction feature contributions, sub-100ms response). External launch with API key auth + metered Stripe billing + customer self-service portal targeted for Q2 2026 (June). Until then, all scoring is delivered as batch CSVs via the tiers above each model. Planned overage rates: $0.10 per score (Starter), $0.05 (Growth), $0.02 (Pro). Subject to change before launch. Early-access subscribers get priority onboarding plus first month free at the Growth tier.