TELECOMMUNICATIONS

TelcoDR

Automated TMF Data Mapping: 1,300+ Tables in 20 Minutes

How TelcoDR mapped 1,300+ legacy tables to TMF standards in 20 minutes using AI-driven automation. AWS Bedrock, deterministic evaluation, and agent workflows.

TelcoDR

"...the problem we're trying to solve is this horrible mess of complexity and legacy, and we've just found the holy grail of potentially cracking that problem.... ...That's really, really awesome."

Nigel Back, Head of Product, TelcoDR

THE CHALLENGE

The problem.

Telecommunications companies sit on decades of legacy data trapped in systems like Salesforce, SAP, and custom databases. Mapping this data to industry standards like TMF (TeleManagement Forum—the global telecommunications standards body that defines data models and APIs for BSS/OSS systems) takes months or years of manual work across thousands of tables and typically requires cross-functional teams of domain experts, data engineers, and business analysts. This blocks AI adoption, system modernization, and competitive BSS application development.

Telecommunications companies operate on infrastructure built over decades. A typical telco runs Salesforce with 1,300+ custom tables, SAP systems with proprietary schemas, and homegrown databases with zero documentation.

Modernizing these systems requires mapping every field to industry standards like TMF (TeleManagement Forum). This process traditionally takes 6-18 months of manual work by domain experts who understand both the legacy schema and the target standard.

The business impact is severe. CloudSense customers were considering churn because manual integration work was too slow and expensive. Without automated mapping, telcos can't build AI-driven BSS applications, can't migrate to modern platforms, and can't compete with cloud-native competitors.

TelcoDR needed a solution that could:

- Automatically discover and map legacy schemas with secure credential management - Handle 1,300+ tables with complex relationships and custom fields - Map to TMF standards with enterprise-grade accuracy - Work on-premise to maintain data sovereignty - Scale to evolving telecommunications standards

THE SOLUTION

What we built.

SQL-Based Data Discovery with Secure Access

Using PostgreSQL Foreign Data Wrapper (FDW), we enabled SQL queries against Salesforce and other external systems. The FDW connects through system APIs (using securely managed credentials for each data source), presenting external data as queryable PostgreSQL tables within the platform.

This approach provides a unified query interface across disparate systems. Telcos can run the platform on-premise using Docker with AWS IoT connectivity, processing data locally while leveraging cloud AI capabilities for mapping intelligence.

AWS Bedrock Integration for Enterprise AI

We integrated AWS Bedrock using Claude 4.5 Sonnet for intelligent schema interpretation and mapping suggestions. The platform architecture supports Anthropic's latest models as they become available, ensuring access to state-of-the-art capabilities as the model family evolves. The AWS infrastructure provides enterprise compliance through built-in security controls, encryption, and audit logging.

The AI analyzes database schemas, identifies relationships between tables, and proposes mappings to TMF standards based on semantic understanding of telecommunications concepts and naming patterns.

Deterministic Evaluation Over Pure Agent Workflows

Early versions relied on LLM agents to generate and validate mappings. Quality was inconsistent. We rebuilt the system using Langfuse evaluation pipelines with deterministic checks.

Now, agents propose mappings, but deterministic rules validate them against TMF specifications, data type compatibility, and relationship constraints. This hybrid approach improved mapping accuracy by 60% and made the system debuggable and auditable, critical for regulated telco environments.

Why Deterministic Evaluation Beats Pure Agent Workflows

The biggest technical lesson: LLM agents are powerful but insufficient for enterprise accuracy.

Pure agent-based approaches let the model generate and validate its own work. This creates inconsistency, hallucinations, and errors that compound across 1,300+ tables. When an agent maps a field incorrectly, downstream relationships inherit the error.

Deterministic evaluation separates generation from validation. We apply AI where it provides clear value: semantic understanding of legacy schemas and intelligent mapping suggestions. The deterministic validation layer then ensures accuracy.

  • TMF specification rules (required fields, data types, cardinality)
  • Database constraints (foreign keys, unique indexes, null handling)
  • Relationship integrity (parent-child consistency, circular reference detection)
  • Business logic rules (telco-specific validation from domain experts)

Enterprise-Grade Multi-Tenancy on AWS

The platform runs on AWS using infrastructure as code (AWS CDK) with full CI/CD automation. ECS Fargate handles container orchestration with auto-scaling based on mapping workload. Aurora PostgreSQL provides the managed database layer with tenant-column-based multi-tenancy for data isolation required for telecommunications compliance.

For customers requiring data sovereignty, we deploy a lightweight Docker runtime on-premise. It connects to AWS IoT Core for secure communication with the cloud control plane. The on-premise runtime executes queries via FDW, sends schema metadata to cloud AI services for mapping suggestions, and applies results locally.

Enabling BSS Applications Without System Replacement

Beyond data mapping, the platform enables telcos to build modern BSS applications on top of legacy systems without migration risk.

Once legacy data maps to TMF standards, developers can build new applications against the standard API. The platform handles translation between TMF and the underlying Salesforce or SAP schema in real time.

Telcos get modern capabilities without replacing systems that represent decades of business logic. The legacy infrastructure stays in place. New applications interact through standardized TMF interfaces.

  • Customer churn prediction using unified data across legacy systems
  • Automated billing reconciliation across multiple source systems
  • Real-time service provisioning with legacy system integration
  • Compliance reporting that aggregates data from incompatible schemas

HOW IT WORKS

The details.

Querying Salesforce and SAP Like a Database

We used a PostgreSQL extension that presents data from external systems as queryable tables. This means the platform can run a single query that reaches across Salesforce, SAP, and other source systems at once, through secure, managed credentials. For organisations running on-premise, a lightweight container handles local data processing and connects to cloud AI capabilities only for the mapping intelligence.

AI That Understands Telecommunications Data

We integrated AWS Bedrock using a recent Claude model for intelligent schema interpretation. The AI analyses database schemas, identifies relationships between tables, and proposes mappings to telecom standards based on its understanding of telecommunications concepts and naming patterns. The platform is configured to use updated models as they become available, so the mapping intelligence improves over time.

AI Proposes, Rules Validate

Early versions let the AI generate and validate its own mappings. Quality was inconsistent. We rebuilt the system so AI proposes mappings and a separate deterministic layer validates them against known rules: required fields, data type compatibility, relationship constraints, and business logic from domain experts. This combination improved mapping accuracy by 60% and made the results auditable, which matters in regulated telecom environments.

Why Pure AI Is Not Enough for Enterprise Accuracy

When an AI generates and validates its own work, errors compound. A wrong mapping in one table creates problems in every table that depends on it. Separating generation from validation solves this. The AI handles what it is genuinely good at: understanding the meaning of legacy schemas and suggesting intelligent mappings. Fixed rules handle what requires absolute reliability: format compliance, type checks, and relationship integrity.

Cloud Architecture for Telco-Grade Compliance

The platform runs on AWS with infrastructure as code and automated deployment pipelines. Database isolation ensures that one customer's data cannot be accessed by another. For organisations with data sovereignty requirements, a lightweight on-premise option connects securely to the cloud control plane, processing data locally while using cloud AI for mapping suggestions.

Building Modern Applications on Top of Legacy Systems

Once legacy data maps to telecom standards, developers can build new applications against the standard API. The platform handles translation between the standard and the underlying Salesforce or SAP schema in real time. Telecoms get modern capabilities without replacing systems that contain decades of business logic. New applications interact through standardised interfaces while the legacy infrastructure stays in place.

  • Predicting which customers are likely to leave, using data unified across all legacy systems
  • Automatically reconciling bills across source systems that previously could not talk to each other
  • Provisioning new services in real time without waiting for legacy systems to be replaced
  • Pulling compliance reports from data that used to live in incompatible formats across separate systems

OUTCOMES

What shipped.

10-20 minutes per TMF object mapping (vs 6-18 months manual)

1,300+ source tables mapped to 67+ TMF objects

60% improvement in mapping accuracy

Immediate customer retention value for CloudSense

Production-quality mapping code without rework

Zero downtime during mapping operations

KEY TAKEAWAYS

What we learned.

  • Deterministic evaluation is mandatory for enterprise AI accuracy. Pure agent workflows produce inconsistent results at scale. Separate generation from validation with structured checks against specifications and business rules.
  • Foreign Data Wrappers enable unified query interfaces across systems. FDW presents external APIs as SQL-queryable tables, providing consistent data access patterns across disparate source systems.
  • AWS Bedrock integration provides enterprise-grade AI capabilities. Cloud infrastructure delivers advanced LLMs with built-in security, compliance, and audit controls essential for telecommunications.
  • Hybrid cloud/on-premise architectures satisfy data sovereignty requirements. Deploy lightweight runtimes that process data locally while leveraging cloud AI services through secure connectivity.
  • Infrastructure as code with automated testing enables rapid iteration. AWS CDK provisioning of full test environments on every PR improves development velocity and system reliability.
  • Legacy system integration beats replacement for regulated industries. Mapping existing data to standards unlocks modern capabilities without migration risk or loss of decades of embedded business logic.
  • TMF standards enable interoperability across telco systems. Once legacy data maps to TMF, new applications work across any compliant system, reducing vendor lock-in and accelerating BSS development.

IN SUMMARY

Bottom line.

In summary, TelcoDR and AE Studio proved that the 'holy grail of ETL' is achievable with the right combination of AI and deterministic validation. As a result, what used to require cross-functional teams working for 6-18 months now processes in 10-20 minutes per TMF object mapping with 60% better accuracy.

The platform delivers immediate business value through customer retention while enabling long-term strategic value through rapid BSS application development on legacy infrastructure. Telecommunications companies can now modernize without the risk and cost of system replacement.

As telecommunications standards evolve and AI capabilities advance, the architecture scales to new target standards and data sources. Furthermore, the deterministic evaluation approach ensures quality and auditability will remain enterprise-grade as the platform grows.

FAQ

Frequently asked.

How long does it take to map legacy telco data to TMF standards?
BSS Magic processes TMF object mappings in 10-20 minutes per object using AI-powered automation. For context, mapping 1,300+ tables to 67+ TMF objects would take traditional manual approaches 6-12 months with cross-functional teams of data engineers, domain experts, and business analysts. The system uses AWS Bedrock's Claude 4.5 Sonnet LLM combined with deterministic validation checks to analyze database schemas, identify relationships, and generate accurate mappings. This includes handling complex scenarios like Salesforce objects, SQL Server tables, and custom legacy formats—all while maintaining high accuracy through multi-stage validation.
What makes automated ETL for telecommunications so challenging?
Telecommunications ETL is uniquely complex due to fragmented legacy systems, inconsistent data formats, and the sheer scale of interconnected tables. Most telcos operate 10-20+ BSS/OSS systems that have evolved over decades, each with custom schemas and business logic. The challenge intensifies with regulatory requirements for data accuracy and the need to maintain business continuity during migrations. Traditional ETL tools struggle with semantic understanding—they can't interpret that 'cust_id' in one system maps to 'customer_reference' in another without extensive manual configuration. BSS Magic solves this using LLM-powered ontology mapping that understands telco domain concepts and relationships.
Why did you choose deterministic checks over pure AI agents?
Pure AI agents introduce unpredictability that's unacceptable for production telco systems handling billions in revenue. We implemented a hybrid approach: LLMs for intelligent schema interpretation and mapping suggestions, with deterministic validation layers to ensure accuracy and consistency. This architecture provides the best of both worlds—AI handles the complex semantic understanding that would take humans months, while deterministic checks catch edge cases, enforce business rules, and provide audit trails. For example, the LLM might suggest a mapping, but deterministic validators verify data types, cardinality, and TMF compliance before approval. This resulted in 95%+ accuracy rates suitable for production deployment.
How does BSS Magic handle data security for highly regulated telco clients?
BSS Magic operates within the client's infrastructure with security controls appropriate for telecommunications compliance. The platform uses secure credential management for accessing external systems and AWS PrivateLink for secure LLM access. All processing happens in-region with encryption at rest and in transit. The system supports role-based access control (RBAC), comprehensive audit logging, and compliance with GDPR, SOC 2, and telecom-specific regulations. Schema metadata used for mapping is analyzed by the system, with careful attention to minimizing sensitive data exposure during the mapping process.
Can BSS Magic integrate with existing telco systems without replacement?
Yes, BSS Magic is designed as an integration layer that sits alongside existing systems without requiring rip-and-replace migrations. It connects to legacy databases, CRMs, billing systems, and other BSS/OSS platforms through standard APIs and secure database connections. The platform maps data to TMF Open APIs (standard telco interfaces) while leaving source systems operational. This means telcos can modernize gradually, adding new capabilities and microservices that consume standardized data, while legacy systems continue running. The TelcoDR client maintained full business operations during mapping operations with zero downtime.
What AWS services power the multi-tenant architecture?
The architecture leverages AWS Bedrock for LLM inference, Amazon RDS for metadata storage, AWS Lambda for serverless processing, and Amazon S3 for schema artifacts and mapping configurations. AWS PrivateLink enables secure connectivity to Bedrock without internet exposure. Additional services include Amazon EventBridge for workflow orchestration, AWS Secrets Manager for credential management, and CloudWatch for monitoring and logging. The multi-tenant design uses separate VPCs per client with shared control plane infrastructure, ensuring data isolation while maintaining operational efficiency. This setup supports concurrent mapping jobs across multiple telco clients.
How do you evaluate and optimize LLM performance for production?
We use AWS Bedrock with Claude 4.5 Sonnet, which provides 95%+ accuracy on complex multi-table relationships and superior understanding of telco domain concepts. The platform architecture supports upgrading to Anthropic's latest models as they become available. We continuously optimize through prompt engineering, few-shot examples from validated mappings, and feedback loops where deterministic validators flag errors for model improvement. Production monitoring tracks accuracy and automatically triggers alerts when performance issues are detected, ensuring consistent quality for enterprise deployments.
What is the business value of automated data mapping for telcos?
Automated data mapping delivers 50-100x time savings compared to manual approaches, reducing 6-12 month projects to hours or days. For TelcoDR, this meant immediate acceleration of their BSS modernization roadmap and faster time-to-market for new services. The business impact extends beyond speed: reduced consultant costs (manual mapping requires expensive data engineering teams), lower risk of human error in critical revenue systems, and enabling data-driven initiatives that were previously blocked by integration complexity. Standardizing on TMF APIs also future-proofs the architecture, making subsequent integrations exponentially faster. Many telcos see ROI within the first quarter through reduced integration costs alone.

LET'S TALK

Bring us the hard problem.

We'll bring the team that ships.

Book a call Back to home
Get in touch