AI Isn’t Failing - Your Data Strategy Is

Organisations are investing heavily in AI. From generative AI to intelligent agents, the pressure to innovate - and fast - is immense.

But here’s the reality:

Most AI projects don’t fail because of AI. They fail because of data.
In a recent discussion between John Dee, Director of Strategy at Somerford Associates and Peter Pugh-Jones, Field CTO at Confluent, one message came through clearly: if you don’t fix your data, your AI investment won’t scale.

The AI Illusion: Why Investment Isn’t Delivering ROI

Many businesses are doubling down on AI platforms, tools, and models. But they’re overlooking the most critical component—the data that powers them.

Even before today’s AI surge:

• Up to 70% of AI models never made it into production

Why?

Because organisations struggled with:

• Fragmented data sources
• Poor data quality
• Complex, legacy infrastructure

Now, with the rise of large language models and AI agents, that complexity is only growing.

The Real Problem: Data Fragmentation

Most enterprises operate with:

• Multiple, disconnected systems
• Siloed departments
• Inconsistent or duplicated data

The result?

• “Multiple versions of the truth”
• Heavy manual data wrangling
• Slow, ineffective AI outputs

And in many cases, leadership doesn’t see the problem - because teams are working behind the scenes to manually clean and fix the data.

AI Hallucinations Start with Bad Data

When AI produces incorrect or misleading outputs—often called “hallucinations”—it’s rarely the model at fault.

It’s the data.

If AI is fed:

• Incomplete data
• Outdated information
• Inaccurate sources

It will make flawed decisions - just like a human would.

Fix the Plumbing First

One of the strongest analogies from the discussion:

“You need to fix your plumbing before you turn on the tap.”

Too many organisations:

• Invest heavily in AI tools
• Underinvest in data infrastructure

This leads to:

• Poor ROI
• Failed scaling efforts
• Increasing technical (and now AI) debt

From Batch to Streaming: A Shift in Thinking

Traditional architectures rely on:

• Batch processing
• Static data storage
• Delayed insights

But modern AI demands:

• Continuous data flows
• Real-time processing
• Event-driven architectures

Streaming technologies (like Kafka) allow organisations to:

• Process data as it happens
• Continuously enrich and validate data
• Deliver faster, more accurate AI outcomes

The Hidden Barrier: Organisational Silos

Technology isn’t the only challenge.

Internal politics and silos often prevent progress:

• Departments optimise for their own goals
• Systems are built in isolation
• Data ownership becomes fragmented

Without a holistic, enterprise-wide data strategy, AI initiatives struggle to scale.

Where Should Organisations Start?

If you’re assessing AI readiness, focus on three key areas:

1. Data & Regulation

• Understand GDPR, AI regulations, and compliance needs
• Re-evaluate how data is managed and secured

2. Cloud Strategy

• Define your approach (cloud, hybrid, on-prem)
• Ensure consistent data access across environments

3. Budget Allocation

• Rebalance investment:
• Less on “shiny AI tools”
• More on data engineering and integration

Start Small—But Think Big

The best approach?

✅ Identify a clear business pain point
✅ Build a focused AI pilot
✅ Ensure high-quality, well-engineered data
✅ Scale strategically across the organisation

But remember: scaling AI is far harder than building it.

The Competitive Divide Is Growing

AI will amplify the gap between organisations.

Those that succeed will:

• Break down silos
• Modernise their data architecture
• Adopt real-time, streaming data models

Those that don’t?

• Will remain stuck in pilots
• Struggle to realise value
• Fall behind faster than ever before

Final Thought

The uncomfortable truth is simple:

If you invest in AI without fixing your data, you’re building on unstable foundations.

The future of AI success isn’t about better models - it’s about better data.