Auto-Classify Every Submission: InsightUW's 6-Step Industry Classification Pipeline

How InsightUW runs a deterministic, auditable classification pipeline on every submission — from D&B lookup to confidence-scored actuarial class — without manual intervention.

The Problem

Industry classification is not a single lookup. It is a chain of dependent steps: resolve the company, find the industry code, map it to an internal class, assign a hazard group, calculate a rate factor, and record the result with enough metadata to audit it later.

When underwriters perform these steps manually, they skip steps, use different sources, and arrive at different conclusions. One underwriter checks D&B; another uses the broker's stated SIC code. One maps to the internal class table; another uses last year's classification for a similar account. The result is inconsistent classifications that corrupt rating accuracy and portfolio analytics.

The challenge is not just automation — it is orchestration. Each step depends on the output of the previous step, and the pipeline must handle partial failures gracefully. If D&B returns no match, the pipeline should still attempt NAICS-based classification at a lower confidence level rather than failing entirely.

The InsightUW Approach

InsightUW runs a 6-step classification pipeline on every submission at intake. The pipeline is sequential, fault-tolerant, and produces a confidence score that determines whether the classification is auto-validated or routed for manual review.

Step 1: D&B Lookup

The pipeline begins by querying Dun & Bradstreet for the insured company (see Blog 36). If a DUNS number is provided on the submission, it is used for an exact lookup. Otherwise, the insured name is matched using exact and fuzzy strategies with a 70% similarity threshold.

Output: DUNS number, industry description, primary NAICS, primary SIC, annual sales, employee count, year started, risk score.

Failure mode: If D&B returns no match, the pipeline continues to Step 2 using any NAICS code provided by the broker on the application.

Step 2: NAICS Resolution

The 6-digit NAICS code from D&B (or broker input) is validated against the NAICS reference table. This confirms the code exists, retrieves its official description, and flags any deprecated codes.

Output: Validated NAICS code and description.

Failure mode: If no NAICS code is available from any source, the pipeline skips to Step 6 and records the classification as incomplete with 0% confidence.

Step 3: SIC Resolution

The validated NAICS code is cross-referenced to its corresponding 4-digit SIC code. InsightUW maintains a NAICS-to-SIC mapping table derived from Census Bureau concordance files.

Output: SIC code and description.

Step 4: Actuarial Class Mapping

The NAICS code, SIC code, and submission LOB are used to look up the Actuarial Class Map table (see Blog 37). This returns the internal class code, actuarial segment, hazard group, rate factor, ISO GL class, and NCCI WC class.

Output: Complete actuarial classification including hazard group and rate factor.

Failure mode: If no mapping exists for the NAICS/SIC/LOB combination, the classification is flagged as "unmapped" and routed for manual review.

Step 5: Confidence Scoring

The pipeline calculates a confidence score based on which data sources contributed to the classification:

The confidence score determines the validation workflow:

• 92%+: Auto-validated. The classification is accepted and flows to rating.
• 65-91%: Pending validation. The underwriter sees a "Validate" button and reviews the classification before it flows to rating.
• 0%: Manual classification required. The underwriter must classify from scratch.

Step 6: Record and Audit

The complete classification result is written to the submission record and logged to the audit trail:

Validation Workflow

Classifications move through three states:

When an underwriter overrides a classification, the original classification is preserved in the audit trail alongside the override reason. This creates a feedback loop for improving mapping accuracy.

Seed Data Results

InsightUW's seed dataset includes 59 submissions that have been auto-classified through the pipeline:

Cyber Submission: NovaPay Technologies

A broker submits a Cyber Liability application for "NovaPay Technologies," a payment processing fintech. Here is the pipeline execution:

Total pipeline time: 0.9 seconds.

What the underwriter sees: When the Cyber underwriter opens the submission, the classification panel shows: NAICS 522320 (Financial Transactions) mapped to FI-PAY-01 (Financial - Payments), Hazard Group I, Rate Factor 0.85x, Confidence 92%, Status: Validated. No action required — the classification is already confirmed and the rate factor is ready for the rating engine.

What This Means for Underwriters

1. Fully automated classification — 59 of 59 seed submissions were classified without manual intervention, with 42 auto-validated at 92% confidence
2. Graceful degradation — When D&B returns no match, the pipeline falls back to broker-provided NAICS codes rather than failing, ensuring every submission gets at least a partial classification
3. Confidence transparency — The confidence tier system tells underwriters exactly how reliable the classification is and whether it needs review
4. Sub-second execution — The full 6-step pipeline completes in under 1 second, meaning the classification is ready before the underwriter opens the submission
5. Complete audit trail — Every classification records its source, confidence, validation status, and timestamp, supporting regulatory review and actuarial analysis

What's Next

The classification pipeline produces data — but underwriters need to see it. Blog 39 covers the Classification Chain Visualization that gives underwriters an interactive visual map from D&B lookup to rate factor, with clickable nodes and validation controls.

InsightUW is an AI-powered underwriting workstation for P&C carriers. Request a demo to see the auto-classification pipeline in action.