Multiple Names for the Same Company
The organization connects institutional clients with subject matter experts across industries for research and advisory engagements.
A reliable understanding of each expert’s employment history is important to that model. The business needs to know which company an expert actually worked for, even when employer information appears under different names or formats.
That requires more than storing company names. It requires a consistent company identity that can connect multiple variations back to the same underlying organization.
Company information across the enterprise contained many variations of the same employer identity.
One organization could appear under its full corporate name, an abbreviation, a business-unit name, or another commonly used variation. Without consolidation, those records could be treated as separate companies even when they referred to the same underlying organization.
Fragmented Employer Identities
Different company-name variants made it difficult to consistently organize experts by the organizations where they had worked.
Without a unified company identity, the same employer could effectively appear multiple times across the dataset.
Unstructured Company Data
Much of the relevant company information existed in large blocks of unstructured text rather than clean, standardized fields.
This made conventional field-by-field matching difficult and reduced the effectiveness of purely deterministic approaches.
Rules-Based Matching Was Not Enough
The relevant company information could not always be reduced to standardized attributes.
As a result, exact-match and traditional rules-based approaches alone were not suitable for the core entity-resolution challenge.
Preserve Existing Data Authority
The enterprise already had a company-data source it considered authoritative.
The objective was not to deduplicate or replace that trusted source. Instead, the organization needed to accurately connect enriched company records back to the correct authoritative company identity.
AI-Based Company Matching in Databricks
LakeFusion established a company-mastering process within the enterprise’s Databricks environment.
The relevant company information had already been enriched upstream and was available in Databricks. Rather than adding another enrichment layer, LakeFusion focused on resolving those existing records into consistent company identities.
LLM-Based Entity Resolution
Because much of the source information existed as unstructured text, LakeFusion applied LLM-based entity resolution directly against the available company information.
This enabled the enterprise to address matching scenarios that could not be handled effectively through deterministic rules alone.
Asymmetric Company Mastering
The mastering model was intentionally asymmetric.
The existing authoritative company source remained unchanged. Instead, enriched records were matched against the appropriate trusted company identity.
This preserved the enterprise’s established source-of-truth model while improving how fragmented company information connected to it.
How LakeFusion Implemented the Company Master
The implementation combined source-data assessment, matching-strategy design, golden-record creation, and continuous refinement.
Data Profiling and Assessment
LakeFusion first assessed the structure of the company data.
The team identified that much of the information needed for matching existed in large unstructured text blocks rather than cleanly separated attributes.
That finding ruled out a purely deterministic matching strategy.
Matching Strategy Design
The teams evaluated two approaches:
- parse unstructured text into structured fields before matching;
- match directly against the existing text.
The direct-matching approach was selected for the initial implementation.
This allowed the enterprise to begin resolving company identities without first introducing a separate parsing layer.
Golden Record Creation
Matched enriched records were connected to their corresponding authoritative company identities.
This created a more consistent company master that could be used across the organization’s expert data while preserving the trusted company source already in place.
Ongoing Match Tuning
Matching was not treated as a one-time configuration.
The implementation continued through additional tuning cycles designed to improve match quality and efficiency as the company-mastering process matured.
Business Value Delivered
The central business outcome was straightforward: experts could be organized around a consolidated employer identity instead of multiple versions of the same company.
More Consistent Expert Organization
Different company-name variations could be connected back to one trusted employer record.
This gave the enterprise a more consistent way to categorize experts based on their employment history.
Better Expert-to-Client Matching
A clearer employer identity improved the context available when matching experts to relevant client engagements.
Instead of working from fragmented company-name variations, the organization could rely on a more consistent employer foundation.
Existing Data Authority Preserved
LakeFusion improved company matching without changing the role of the enterprise’s authoritative company-data source.
Enriched records were connected back to trusted identities rather than replacing or independently remastering the authoritative dataset.
Complex Data Became Matchable
AI-based entity resolution gave the enterprise a way to resolve company identities across highly unstructured information where traditional rules-based approaches were not practical.
This expanded the types of company data that could participate in a consistent mastering process.
Foundation for Continued Optimization
Working across approximately 100 million records also provided practical insight into how preprocessing strategy, matching architecture, production scale, and AI cost interact.
Those learnings created a foundation for continued optimization as the implementation evolved.
What Made the Initiative Successful
Several factors helped keep the initiative moving despite the complexity of the underlying data.
Strong Stakeholder Collaboration
Close coordination between enterprise stakeholders and the LakeFusion product team helped keep technical decisions aligned with the business requirement.
Technical Fit for Unstructured Data
LakeFusion’s ability to apply AI-based entity resolution directly against unstructured company information made the project viable where conventional field-based matching was not sufficient.
Iterative Match Refinement
The implementation treated entity matching as an evolving process rather than a one-time configuration exercise.
Continuous tuning created a path to improve matching quality and efficiency over time.
Support for Custom Requirements
The initiative involved a high degree of non-standard requirements, making sustained coordination between technical and business teams particularly important.
Learning at Production Scale
Operating at very large record volumes reinforced the importance of validating preprocessing decisions, matching architecture, and AI-cost assumptions against production data.
From Company Name Variants to a Trusted Employer Foundation
The enterprise began with a clear operational challenge: the same company could appear under multiple names across its expert data, making it difficult to consistently understand employment relationships.
LakeFusion addressed that challenge by establishing a company master in Databricks and applying AI-based entity resolution to connect enriched company information back to trusted employer identities.
The organization now has a more consistent foundation for organizing experts, supporting client matching, and continuing to improve company-data quality as the implementation matures.
.webp)
.webp)
