Meta Tagging

Company Profile

Fortune 500 Telecommunications

Industry

Telecommunications

Function

Executive

Region

Global

Executive Summary

A Fortune 500 Telco faced challenges with data fields that were difficult to understand for data and business teams alike because there was no easy way to tag complex technical terms with business friendly terms. This resulted in large amounts of valuable business data being unusable until the fields were manually renamed. To address these issues, the Telco giant implemented fine-tuned models, powered by Aible, to tag complex industry-specific variable names to improve business consumability.

Challenge

A large telecommunications provider had data fields that were tagged with unintuitive and complex naming conventions. For example, the team used the first three letters of a word as an abbreviation which resulted in same cases where the same abbreviation applied to multiple terms. The confusing naming conventions led to:

Overwhelming Abbreviations & Acronyms: The data was unconsumable for their teams. The data engineers, architects, and business analysts could not easily understand what each field represented.

Far From Automated: The organization had tens of thousands of fields that would need manual renaming to provide more plain and interpretable terms so that there could be universal understanding.

Solution

The Telco was able to automatically create and assign plain language descriptions of the once complex data fields by leveraging fine-tuned models, powered by Aible within 10 days. The fine-tuned AI was able to learn context and understand the meaning for identical abbreviations in different contexts which helped solve associated challenges.

Painless Analysis: ~1K different variable names were analyzed and then auto-assigned plain language descriptions of the different variables.

Fine Tuning For Industry Specificity: The Telco then provided feedback on which looked good and which needed additional curation to then create a fine-tuned model.

A Bespoke LLM for Less Than $100: It cost less than $90 in cloud-compute spend to fine-tune the model.

Results

The fine-tuned model was able to successfully tag the variable names to a standard that exceeded the expectations of the Fortune 500 Telco. Within 10 days, this unlocked the potential for millions of dollars of potential business value by making previously inaccessible data usable across a diverse range of business units.