Why Banks Can't Deploy AI: The Data Taxonomy Crisis

The Accumulated Taxonomy Debt

Modern banks aren't single institutions - they're archaeological layers of merged entities, acquired portfolios, and legacy systems spanning 30-50 years:

• Retail banking from the 1980s with product codes designed for mainframe batch processing
• Commercial lending from a 1995 acquisition using completely different client classifications
• Wealth management from a 2005 merger with incompatible customer taxonomies
• Digital banking launched in 2015 with modern product codes that don't map to legacy
• Open banking APIs from 2020 requiring yet another taxonomy layer

Each layer uses its own classification systems. None interoperate. Nobody has authority to standardize across all of them.

Six Critical Failure Modes

1. Cloud FinOps Becomes Impossible

Banks mid-cloud migration discover that taxonomy chaos extends to infrastructure spend. FinOps teams need to aggregate costs across AWS, Azure, and GCP - but find the same classification problems that plague business operations:

• Cost centers defined differently across cloud platforms
• Application and service taxonomies inconsistent between on-premise and cloud
• Business unit hierarchies don't map to cloud resource tagging conventions
• Product ownership unclear when cloud resources span legacy organizational boundaries
• No standardized taxonomy for cloud services across multi-cloud environments

The operational reliability impact compounds the financial visibility problem:

• Incident response slowed: When ownership is ambiguous across cloud resources, critical incidents take longer to route to correct teams
• Compliance audits fail: Can't trace cloud resources to regulatory requirements when resource taxonomies are inconsistent
• Disaster recovery incomplete: Service dependencies aren't properly classified, recovery plans miss critical resources
• Security gaps: Security teams can't identify what's running where when resource naming is chaos

The competitive disadvantage: Challenger banks and fintechs have cloud-native cost allocation from day one. They know exactly what each service costs, which business lines drive spend, where optimization opportunities exist. Incumbent banks spend 20-30% of FinOps capacity just figuring out what things are called.

2. Regulatory Reporting Becomes Impossible

Basel III requires risk-weighted assets calculated one way. IFRS 9 requires expected credit loss calculated differently. MiFID II needs transaction reporting with specific classifications. BCBS 239 demands enterprise-wide risk aggregation.

Each regulatory framework assumes clean, consistent taxonomies. Your bank has:

• Product hierarchies that evolved organically over decades
• Customer classifications that differ by business line
• Transaction types inconsistent across channels and regions
• Risk ratings that mean different things in different systems
• Exposure definitions that vary by risk type

The cost isn't just reconciliation time - it's compliance risk. When taxonomies are informal and undocumented, you can't prove to regulators that your risk calculations are accurate.

3. Customer 360 Initiatives Fail Predictably

Your bank wants a unified customer view. The business case is compelling: better cross-sell, improved service, regulatory compliance (GDPR subject access requests require finding ALL customer data).

Then the project discovers that the same customer exists with different identities across systems:

• Retail banking: "Account Holder #12847593"
• Commercial lending: "Client - SME Tier 2"
• Wealth management: "HNW Investor - Private Banking"
• Credit cards: "Cardholder - Premium Segment"
• Mortgages: "Borrower - Residential Portfolio"

These aren't just naming differences - they're fundamentally incompatible classification schemes with different attributes, hierarchies, and business logic.

The business impact is concrete:

• Customer has a mortgage AND a business account, but cross-sell systems don't connect them
• GDPR subject access request requires manual searching across 20+ systems
• AML/KYC reviews miss complete customer picture because data is fragmented
• Relationship managers can't see complete client exposure across products

4. Risk Management Operates Partially Blind

Enterprise risk aggregation (BCBS 239) requires synthesizing exposure data across all risk types. But different risk functions use incompatible taxonomies:

• Credit risk: Exposures classified by internal rating, industry sector (using proprietary taxonomy), geographic region, product type
• Market risk: Positions classified by asset class, trading book, desk, risk factor sensitivity
• Operational risk: Events classified by Basel event types, business line, affected process
• Liquidity risk: Funding sources classified by maturity, counterparty type, currency

When the board asks "What's our total exposure to the technology sector?", the answer requires manually reconciling four different definitions of "technology sector" across four risk taxonomies.

5. AI/ML Projects Discover the Problem Too Late

Your bank invests £5M-20M in AI initiatives:

• Fraud detection requiring consistent transaction classifications
• Credit scoring needing standardized customer attributes across products
• Next-best-action requiring unified product taxonomy
• Churn prediction needing coherent customer segmentation

Every project follows the same pattern:

1. Month 1-3: Demo works beautifully with sample data
2. Month 4-6: Production data integration reveals taxonomy chaos
3. Month 7-12: Team attempts manual data standardization
4. Month 13-18: Project quietly shelved or dramatically descoped

6. M&A Integration Takes 18-24 Months

Your bank acquires a smaller institution. The strategic rationale is sound. The financial model shows clear synergies. Then integration begins.

Product mapping alone takes 6-9 months:

• Acquired bank has 127 product codes
• Your bank has 203 product codes
• Some products are identical but use different codes
• Some products are similar but structured differently
• Some codes map to multiple products in the other system
• Nobody documented the business logic behind the original codes

Customer data integration takes another 6-12 months. Risk data mapping takes 4-6 months. Regulatory reporting harmonization continues indefinitely.

The taxonomy reconciliation work wasn't in the M&A budget. Synergy realization gets pushed back 12-18 months. Integration costs balloon.

Why This Gets Worse

Open Banking Requires API-Ready Taxonomies

PSD2 and open banking regulations require banks to expose data through APIs. But APIs need standardized, well-documented data structures. Your internal taxonomies are neither.

Building API layers on top of inconsistent internal taxonomies creates technical debt at scale. Every API endpoint becomes a custom mapping exercise. Changes to internal systems break external integrations.

ESG Reporting Adds New Classification Requirements

Climate risk disclosure, sustainable finance taxonomies (EU Taxonomy Regulation), and ESG reporting frameworks require classifying entire loan and investment portfolios by environmental impact, carbon intensity, and sustainability criteria.

These classifications need to integrate with existing risk, product, and customer taxonomies. Adding another incompatible layer to already fragmented systems.

Digital Transformation Requires Unified Data

Cloud migration, real-time processing, event-driven architecture - all assume clean, consistent data models. Moving fragmented taxonomies to modern infrastructure just makes the fragments more visible.

The Fintech Competitive Pressure

Challenger banks and fintechs start with unified data models built for cloud-native architecture. They can deploy AI/ML in weeks instead of years. They offer real-time insights because they don't have 30 years of taxonomy debt.

Incumbent banks can't match this agility without addressing the underlying taxonomy problem.

What FireCherry Does

FireCherry specializes in taxonomy standardization for regulated industries where accuracy, governance, and audit trails are non-negotiable. Our banking-specific expertise covers:

• Regulatory taxonomy mapping (Basel III, IFRS 9, MiFID II, BCBS 239)
• Multi-jurisdiction product hierarchies
• Customer/entity data unification across business lines
• Risk classification standardization (credit, market, operational, liquidity)
• Transaction and instrument taxonomies

Our Approach for Banks

Phase 1: Regulatory Taxonomy Assessment (3-4 weeks)

We map your existing classification systems across:

• Product hierarchies (deposits, lending, investments, services)
• Customer/entity taxonomies (retail, commercial, institutional)
• Transaction classifications (payment types, channels, purposes)
• Risk taxonomies (exposures, ratings, collateral, concentrations)
• Regulatory reporting structures (Basel, IFRS, jurisdiction-specific)
• Geographic and organizational hierarchies

Deliverable: Taxonomy standardization roadmap with regulatory impact analysis, compliance risk assessment, and cost-benefit quantification.

Fixed price: £13,500

Phase 2: Core Banking Taxonomy Standardization (14-20 weeks)

We formalize your taxonomies with:

• Formal specifications with URIs and version control
• Regulatory framework mappings (Basel risk weights, IFRS 9 classifications, etc.)
• Cross-system reconciliation rules
• Migration tooling for legacy data
• Integration with core banking, risk, and reporting systems
• Governance frameworks with audit trails
• Change management processes for taxonomy evolution

Deliverable: Production-ready taxonomy infrastructure with regulatory compliance documentation.

Typical engagement: £150k-250k

Phase 3: AI/ML Data Preparation (12-16 weeks)

With standardized taxonomies in place, AI implementations actually work:

• Customer data unified for 360 initiatives and ML models
• Transaction data standardized for fraud detection
• Product taxonomies enabling cross-sell and recommendation engines
• Feature engineering based on consistent classifications
• Quality validation frameworks for model training data

Typical engagement: £100k-200k

Phase 4: Regulatory Reporting Automation (14-18 weeks)

Automate regulatory data assembly with:

• Basel III risk-weighted asset calculations from standardized taxonomies
• IFRS 9 expected credit loss staging and measurement
• MiFID II transaction reporting classifications
• BCBS 239 risk data aggregation
• Audit trails linking regulatory submissions to source classifications
• Version control for regulatory taxonomy changes

Typical engagement: £120k-250k

Why Banks Choose FireCherry

Client Infrastructure Deployment

We understand banking confidentiality and regulatory requirements. All work performed on client infrastructure:

• No data leaves your environment
• You retain complete control and ownership
• We deliver specifications, tooling, and governance frameworks
• Seamless integration with existing systems
• Audit documentation suitable for regulatory examination

"Banks don't have AI problems or technology problems. They have data taxonomy problems accumulated over 30 years. Fix the foundation, and everything else becomes possible."

Related reading: Explore our guide on why enterprise codesets need formal specifications, or see how AI projects fail when data preparation is skipped.