The AI Value Gap

Business Case First, Data Foundations Next

Over the past 6 weeks I’ve been travelling extensively and meeting clients, vendors, and attending various tech and industry conferences and events. I’ve spoken with tech start up founders, CTOs, operations leaders, and data teams across industries.  There is a lot of excitement in the space, and at times it feels like the AI sector is moving at the speed of light, and it is very easy to get caught up in the fear of being left behind.

The Generative AI (GenAI) hype seems to be reaching a crescendo. Leaders want copilots, agents, and predictive systems that drive efficiency, productivity and lower costs. But scratch beneath the surface and many organisations are stuck in a pattern of limited use case proofs-of-concepts (PoCs) – not scalable, value-producing technology.

Where is the value?

If appears to me the core issue is: in chasing ‘shiny’ GenAI outcomes, companies forget to link the technology to the value levers that drive business goals.  The result is impressive outputs with nice visualisations which aren’t anchored to the KPIs that matter – e.g. revenue and growth, cost reduction, asset productivity. Without a clear business case priorities drift and adoption stalls. Start with the lever and KPI, then shape the use case.

Once the business case is defined, another major pitfall that we’ve seen time and time again is a lack of focus and investment in data architecture and quality.  It’s unglamorous work, but the organisations that get it right are the ones who have the solid foundations to build AI solutions that have measurable impact.

Where the bottlenecks show up:

1. Data architecture. Data sits in silos and/or in proprietary formats rather than a single, searchable, scalable and easily-accessible repository. Data labelling is inconsistent/incomplete, limiting usability. To give an example in the context of what we’ve seen in predictive maintenance, sensor tags are often not mapped to the actual equipment, making meaningful analysis challenging and time-consuming.
2. Data quality. Quality assurance is done as a one off clean-up for running proof-of-concepts on limited data sets rather than an a broader continuous top down strategic process.   This stop–start approach forces teams to rebuild foundations for every initiative, consumes scarce SME time, and creates a persistent delivery bottleneck.  An example of this is misalignment of timestamps across data sets,  alignment is required every time a PoC is done.
3. Governance & access:  Unclear ownership and ad-hoc role-based access.  For example, we’ve seen organisations use 3rd parties to manage their data making it commercially prohibitive to do analysis at scale.

Across conversations, there seems to be a consistent trend: lots of narrow POCs with minimal impact. Scaling AI for business impact requires the “unglamorous” business case and structural work to be done. The companies realising value aren’t the loudest on stage; they’re the ones treating these foundations as non-negotiable.

A simple test for leaders

If AI disappeared from your business tomorrow, which decisions would slow down or get worse? If you can’t list them, you haven’t crossed from experimentation to impact.

Key questions to consider

• Business case: What action will change, and when, because of this AI use case? What is today’s baseline?
• Impact: Which specific value levers has AI improved, by how much, and who owns the result?
• Data readiness: Have we got the building blocks in place to allow us to use our data effectively?  What specific issues are we encountering?

AI value isn’t constrained by algorithms or models, it’s constrained by foundations. Build the basics first – a use case linked to the business value levers and a coherent data architecture and quality strategy. Do that, and AI has the best chance of impacting business performance; skip it, and you’ll likely have nice dashboards but limited value creation.

If you’d value a conversation about putting the building blocks in place to maximise returns from your data and AI, I’d welcome the opportunity to connect.