Skip to main content
12 min read

How Enrichment Works

ABM.dev uses a sophisticated multi-source enrichment engine that gathers evidence from multiple providers, synthesizes insights with AI, and delivers 90 standardized fields with confidence scores.

Overview

When you submit an enrichment request, ABM.dev runs a sophisticated pipeline: analyzing source availability, gathering evidence in parallel from multiple providers, synthesizing insights with AI, and outputting 90 standardized canonical fields with per-field confidence scores.

Multi-Source

4+ providers queried in parallel for comprehensive coverage

AI Synthesis

Intelligent merging with narrative generation and persona matching

90 Fields

Standardized canonical fields with per-field confidence

CRM Sync

Automatic field mapping and writeback to HubSpot, Salesforce

The Enrichment Pipeline

Every enrichment request flows through a 6-stage pipeline designed for maximum accuracy and efficiency:

1

Input Normalization

Your input data (email, name, company, LinkedIn URL) is normalized and validated. Domain names are extracted from emails, names are parsed into components, and LinkedIn URLs are standardized.

[email protected] → domain: acme.com, likely name: Jane Smith
2

Source Portfolio Analysis

Before querying any sources, we analyze which providers can contribute based on your input. This produces a portfolio score predicting enrichment quality.

TierScoreMeaning
Excellent≥0.85All key sources available (LinkedIn + email + company)
Very Good≥0.70Most sources available, high-quality expected
Moderate≥0.50Core sources available, solid enrichment
Poor<0.50Limited sources, may need more input data
3

Parallel Evidence Gathering

All available sources are queried simultaneously to minimize latency. Each source returns evidence with its own confidence level.

LinkedInHunter.ioPerplexityTavily

See Data Sources for details on each provider.

4

AI Synthesis

Evidence from all sources is merged using AI models that resolve conflicts, generate narrative fields, and match buyer personas. This stage:

  • Resolves conflicting data using source reliability and recency
  • Generates summaries, highlights, and outreach angles
  • Matches against your buyer personas with confidence scoring
  • Calculates ICP fit scores for companies

AI Models

By default, synthesis uses Claude Sonnet for generation and Claude Haiku for auditing. See Advanced Configuration to customize models.
5

Projection & Validation

Synthesized data is projected into 90 canonical fields. Each field receives:

  • Confidence score (0-100) — how reliable is this value?
  • Source attribution — which sources contributed?
  • Freshness timestamp — when was data last verified?

See Canonical Fields for the complete field reference.

6

Field Mapping & Writeback

If CRM integration is enabled, canonical fields are transformed and written to your CRM using configurable field mappings. Transformations can format, concat, split, or convert values.

canonical: title → hubspot: jobtitle → "VP of Engineering"

See Field Mapping for transformation options.

Understanding Confidence Scores

Confidence scores help you decide how to use enriched data in your workflows.

0.9+

High Confidence

Multiple sources agree. Safe for automated workflows.

0.7-0.9

Medium Confidence

Good for enrichment, consider human review for critical use cases.

<0.7

Low Confidence

Limited source agreement. Recommend manual verification.

Best Practice

Set confidence thresholds in your integration logic. For example, only auto-update CRM fields when confidence is above 0.85.

Data Sources

ABM.dev aggregates data from multiple providers, each with different strengths:

Source TypeBest ForData Freshness
Social NetworksCurrent job title, profile photo, connectionsReal-time
Company DatabasesCompany size, funding, industryWeekly updates
Email VerificationEmail validity, deliverabilityReal-time
TechnographicsTech stack, tools usedMonthly scans

Continue Learning