Data Lake / Data Warehouse Modernization
Telia wanted to become a more data-led telecom operator, but one of its major data warehouses had metadata for only a tiny fraction of fields, making it hard for teams to find, understand, govern, and use information such as call records and customer interactions efficiently.
Reported impact
Telia reduced the cost and time of metadata creation, made internal data easier to discover and use, improved governance and privacy management, and created a stronger data foundation for future sales, service, and marketing AI use cases.
Telia and Accenture built a generative AI metadata-enrichment workflow on Amazon Bedrock to make a large telecom data repository easier to use at scale. The implementation starts with prompts, examples, and warehouse context that Bedrock uses to generate tags and descriptions for missing metadata fields. Accenture applied the approach across 87,000 data fields, then stored the outputs in a centralized metadata repository so teams could search, interpret, and trust datasets more easily. The repository also strengthens governance by making assets easier to manage under privacy and regulatory controls. To keep quality high, the workflow includes a human-in-the-loop review model that flags AI-generated descriptions for Telia's data experts to verify and maintain. Accenture also provided training and change-management support so Telia could operate the solution independently and extend the same data foundation into later generative AI use cases for sales, service, and marketing.
Data-field definitions from Telia's warehouse are passed to Amazon Bedrock with prompts and examples so the model can generate tags and descriptions for missing metadata. Those outputs are stored in a centralized metadata repository and flagged through a human-in-the-loop review system so Telia's experts can verify accuracy before teams across the company use the enriched repository for analysis, governance, and downstream AI use cases.
"Accenture used Amazon Bedrock to analyze Telia’s data and automatically populate corresponding metadata descriptions."
"It relies on a 'human-in-the-loop' review system that flags AI-generated column descriptions"
"leverage Amazon Bedrock, building an intelligent solution that updated tags and descriptions for 87,000 data fields"
"only 1% of its data had corresponding metadata"
"saving Telia 10,000 data engineer hours and €700,000 in data entry costs."
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
Accenture Sample Aug 14, 2026 Novartis Novartis needed to unify fragmented data across R&D, manufacturing, commercial, and corporate functions so teams could analyze growing biomedical and operational data sets faster and apply AI to drug development and… 10 days to three days Amazon Web Services (AWS) · Microsoft Azure · analytics tools supported by AI and ML 
Google Cloud Sample Archived · Jul 23, 2026 TELUS TELUS needed to unify siloed customer and network data so teams could trust the same foundation for generative AI, launch new services faster, and scale agentic platforms without each business unit building its own… 90% of AI model traffic runs through Gemini Enterprise Agent Platform BigQuery · BigQuery Analytics Hub · Cloud SQL 
Snowflake Sample Archived · Jul 17, 2025 CityFibre create a single, event-based data ecosystem that could bridge lines of business, integrate core SaaS systems, and support operational reporting plus ML at CityFibre’s pace of growth. eight lines of business with access in 18 months SageMaker · Snowflake · Snowpark 