Every AI strategy meeting eventually hits the same wall: “Our systems can’t do that.”

Technical debt – the accumulated cost of shortcuts, outdated technology, and deferred maintenance – is the silent killer of AI ambitions. Understanding this connection is essential for realistic AI planning.

The Problem in Plain Terms

Technical debt shows up in many forms:

• Legacy systems that can’t expose data via modern APIs
• Data silos that prevent unified views
• Outdated databases with performance limitations
• Brittle integrations that break when touched
• Undocumented systems that only one person understands
• Security vulnerabilities that prevent cloud migration

As MIT Technology Review has documented, AI doesn’t make these problems disappear. It makes them worse.

Why AI Amplifies Technical Debt

AI projects need things that technically indebted organisations struggle to provide:

Data access: AI needs to read from and often write to multiple systems. Legacy systems with poor APIs make this painful or impossible.

Data quality: AI exposes data quality issues that were previously hidden. That “good enough” data suddenly isn’t.

Integration speed: AI projects move fast. Legacy integration that takes months kills momentum.

Scale: AI inference at scale requires infrastructure that legacy environments often can’t provide.

Security: Modern AI platforms require modern security controls that legacy systems lack.

The result: AI projects spend 60-70% of their budget on data and integration work rather than AI itself.

Assessing Your Debt

Before AI planning, honestly assess technical debt:

Data Access Debt

• Can you query data from core systems via API?
• How long does it take to get a new data extract?
• Is real-time data access possible?
• Who has to approve data access, and how long does that take?

Data Quality Debt

• Do you have master data management?
• Are key entities (customers, products, etc.) consistently identified across systems?
• What percentage of critical fields are complete and accurate?
• Do you have data quality monitoring?

Integration Debt

• How many point-to-point integrations exist?
• What’s the average time to build a new integration?
• How often do integrations fail?
• Do you have an integration platform or is everything bespoke?

Infrastructure Debt

• Can your infrastructure support modern AI workloads?
• Are you cloud-capable for AI services?
• Do you have adequate compute for model training/inference?
• Is your network adequate for data movement?

Score yourself honestly on each dimension. If you’re “red” on most, AI will be a struggle.

The Strategic Choice

Organisations face a choice:

Option A: Fix debt first, then AI

• Invest in modernisation before AI initiatives
• Longer timeline to AI value
• More solid foundation when you get there

Option B: AI despite debt

• Accept higher costs and constraints
• Build workarounds for technical limitations
• Get to value faster but with more fragility

Option C: Parallel investment

• Fund both AI and modernisation
• AI pilots inform modernisation priorities
• Modernisation enables AI scaling

Most organisations end up with Option C, but that requires more budget and coordination.

The Debt Payment Plan

If you’re carrying significant technical debt, a realistic plan:

Year 1: Stabilise and Assess

• Comprehensive assessment of technical debt
• Prioritisation based on AI and other strategic needs
• Quick wins that unblock near-term initiatives
• Architecture vision for target state

Year 2: Foundation Work

• Core data infrastructure improvements
• API layer for key systems
• Data quality improvements for priority domains
• Cloud migration for AI-relevant workloads

Year 3: Accelerated Capability

• Broader AI deployment on improved foundations
• Continued modernisation aligned with AI expansion
• Platform maturity enabling faster delivery

This timeline is realistic for significant debt. Trying to compress it creates failure risk.

The Budget Conversation

Technical debt has real costs:

Hidden costs of debt:

• Higher cost per integration
• More time for every data project
• More incidents and outages
• Higher security risk

Investment to address:

• Legacy modernisation
• Data platform development
• Integration platform
• Cloud migration

The conversation with finance: “We can continue paying the debt interest (higher costs for everything) or we can pay down the principal (modernisation investment).”

Quantify current debt costs where possible. The numbers often make the case.

What AI Can’t Wait For

Some AI initiatives can proceed despite debt:

Standalone productivity tools (Copilot, etc.) that don’t require deep integration can deploy now.

External-facing AI that uses cloud data can sometimes bypass legacy systems.

Pilots in modern islands where you have clean data and modern systems can prove value.

Use these to build AI experience and demonstrate value while debt gets addressed.

What Must Wait

Some AI initiatives should wait for debt reduction:

Cross-system AI that needs data from multiple legacy sources will struggle.

Real-time AI that requires fast data access won’t work on slow systems.

AI at scale that needs production-grade infrastructure can’t run on inadequate foundations.

Being honest about what’s not ready prevents expensive failures.

The Conversation to Have

Technical debt is often an uncomfortable topic. Leadership may not want to acknowledge it. Here’s the productive framing:

“To achieve our AI ambitions, we need certain capabilities. Here’s our current state versus required state. Here’s the investment needed to close the gap. Here’s what we can do now versus what needs to wait.”

Focus on enabling AI rather than criticising the past. It’s easier to build support for “enabling the future” than “fixing old mistakes.”

Final Thought

Technical debt isn’t a moral failing. It’s the accumulated consequence of years of decisions, many of which were right at the time.

But ignoring debt while pursuing AI leads to expensive failures. The organisations that succeed with AI are usually those that face their technical reality honestly and invest accordingly.

You can’t AI your way out of technical debt. But you can address both together, with realistic expectations and appropriate investment.

That’s the path that actually works.