The AI Paradox

We Have Seen This Before. But This Time It Is Harder.

In 1998, John Thorpe introduced what he called the Information Paradox. Companies were investing heavily in IT yet struggling to demonstrate measurable improvements in business performance. I met John Thorpe in Australia in 2005 where we discussed this at length, a conversation that I often reflect on.

When I look at what is happening with AI today, it feels very familiar. Many organisations are investing in AI, running experiments, pilots, and building capability. But when you speak to boards, executives and managers, there is often a consistent question in the background. Where is the actual value? Hence, the AI Paradox: AI investment is accelerating, but value does not meet expectations.

We have seen this pattern before

Over the past 30 years there has been a recurring pattern in how organisations adopt new technology:

• In the 1990s, IT investments did not translate into productivity gains (hence the popularity of the Information Paradox book, frameworks and techniques).
• In the 2000s, ERP and CRM implementations were large, complex, and often underdelivered against expectations.
• From 2010s, SaaS adoption accelerated, but value outcomes were inconsistent.

Today, AI is following a similar trajectory. Technology moves quickly. Value capture takes longer.

A structural issue created over the last 20 years

There is, however, an additional challenge today that makes the AI Paradox more difficult to solve. During the ERP and SaaS waves, organisations were encouraged to adopt standard processes embedded in systems. This approach meant that companies would often skip the step of documenting their own processes to accelerate the implementation of the new system. The principle was to avoid customizing the system to fit the organisations processes, as this would eventually result in lower delivered efficiency, scalability and higher maintenance/upgrade costs. But post system implementation, this approach also had an unintended consequence. Many organisations no longer have clearly documented processes or workflows. Instead, processes are embedded inside systems, configurations, and tools. The organisation runs on workflows that are implicit rather than explicit.

Why this matters for AI

AI does not simply automate tasks. It rewires workflows. To redesign a workflow, you need clarity. You need to understand inputs, outputs, decision points, and ownership. Without this, AI gets applied at the edges rather than at the core. This is why many AI initiatives result in incremental improvements rather than material performance uplifts.

What stays the same

Some of the lessons from the Information Paradox are still highly relevant today. Organisations need:

• Value-based investment decisions, linking AI investments to performance uplifts.
• Portfolio governance, selecting and funding the right AI initatives.
• Benefits realisation discipline, measuring (continuously) that benefits are achieved.

These are still necessary foundations. Every company understands this. But they are no longer sufficient on their own.

What needs to change

To realise value from AI, organisations need to go one level deeper. They need to reconstruct how work gets done. This means mapping workflows end-to-end. It means identifying high-frequency, high-friction, high-cost activities. It means redesigning workflows with AI embedded into them, not layered on top. In practical terms, this is less about technology deployment and more about operating model change.

A simple way to think about it

Most organisations are currently focused on two areas:

•  Capability, which includes AI models, tools, and platforms, and
• Use cases, which include AI pilots and proofs of concept (POCs).

But value sits in a third area: Workflows. If AI is not transforming workflows, it is unlikely to materially uplift performance.

A practical way forward

This is not a new problem. It is a familiar pattern, now playing out in a new context. But the solution is more demanding. It require organisations to rebuild their understanding of how work gets done and then redesign it. Until that happens, AI will continue to feel promising but underwhelming in terms of measurable outcomes.

Getting value from AI is less about doing more, and more about doing a few things right with much greater clarity and intent.

Three practical actions that will make a difference:

1.       Select 3–5 high-value workflows and redesign them end-to-end with AI embedded.

• Identify where the economic value sits, such as cost, revenue, or risk, and focus there.
• Redesign the future-state workflow rather than documenting everything as-is.
• Be explicit about decision points, where AI is used, how humans interact with it, and who owns the outcome.

2.      Build an AI portfolio that compounds over time, not a list of disconnected use cases

• When prioritising new initiatives, assess how each one reuses or strengthens what is already in place, including data, models, and workflows.
• Track how capabilities are being reused and scaled, not just the standalone ROI of each use case. This requires deliberate sequencing, not just prioritisation.

3.      Stand up roles and capacity to drive adoption and workflow change

• Assign clear accountability for embedding AI into day-to-day operations. This includes people responsible for workflow redesign, change management, and ongoing performance tracking. Without this layer, even well-designed AI solutions tend to stall after initial deployment.

References

Thorpe, J 1998, The Information Paradox, McGraw-Hill.