It seems like every day, a major brand is making news around their latest AI implementation, pilot or innovation.
What can be harder to come by, however, is what happens after the announcement. What these projects are turning into, what lessons have been learned, and what problems have been solved?
This brings to light one under-discussed issue which businesses are increasingly facing: how to ensure these AI projects don’t get stuck in the pilot phase. Because the stark reality is that many projects are abandoned before they provide ROI or glimpses into their future potential.
In fact, as of this year, 90% of enterprise AI pilots fail before they generate a single pound. Conversely, however, there wouldn’t be this level of investment if no one were seeing success.[Text Wrapping Break][Text Wrapping Break]The question then becomes: why is this technology working for some, but not all? And how can I ensure it works for me?
General AI is no longer enough
When we look back on the first wave of Generative and Analytic AI projects, we can see that their primary focus was promoting accessibility through the installation of more general-purpose use cases. In 2024, Gartner forecasted that 30% of generative AI projects would be abandoned after the proof-of-concept phase by 2025 due to poor data quality and unclear business value.
This change Gartner is ultimately forecasting comes down to the increasing desire for contextual relevance within agents. Over the past year, businesses have started to realise the immense value which can stem from agents which are truly immersed in both a business and industry. Without this specific, unique and contextual knowledge, agentic insights can provide only limited utility. That’s because the insights still need to be corroborated against industry trends and regulations. In short, businesses have realised that general AI is no longer enough.
Simultaneously, the rate of innovation – along with corporate FOMO – remains rapid. So there’s pressure on organisations to invest quickly, but to do so strategically. [Text Wrapping Break]
Bridging the context gap
As the race toward operational AI use cases continues to grow at a rate that can be hard to conceptualise, so does the “context gap.”
The right decisions are driven by context derived from specific industry domain understanding and the unique business data. When companies use incomplete data, they risk what’s known as the “accuracy paradox”, which is the phenomenon where models may demonstrate high technical accuracy, but fail to translate those into real-world scenarios. This can result in significant miscalculations when making major business decisions, inefficiency, and missed opportunities.
The gap exists between general AI tools that lack this capability, and those that are tailored to be industry-specific.
To bridge this gap, technical architectures are shifting. By prioritising contextualised semantic models, businesses can transform raw, fragmented data into trusted, relevant insights. This is how you turn a pilot project into something that has real business value and a clear ROI.
It’s not about the tools, it’s about the right tools for you
Let’s consider health care providers as an example. They struggle with information fragmented across electronic medical records, data warehouses, and clinician notes. They need an intelligence layer that can find new insights by connecting unstructured data with clinical, claims, and financial data sources.
Compare this to retailers who are facing a widening intelligence gap as customer, brand sales, product inventory, and supply chain intelligence remain trapped in disconnected silos.
Businesses in both industries need to maximise their Agentic Analytics investments to get ahead of these burgeoning issues. They need a platform which will deliver consistent, deterministic answers governed by a leading semantic layer. However, the insights they need require different information to ensure their relevance. The competitive intel, internal connectors, regulations and industry data which need to be considered to ensure contextual accuracy are widely different.
Trying to make something retro fit all of these into a governing semantic layer can be time consuming and create the bottlenecks that an analytic program was meant to remove. Ensuring the right contextual information and language informs these insights from the very start is key to ensuring immediate ROI for these projects.
And hence, a key driver in moving pilot projects into the production phase.
Moving beyond pilots, and into what’s next
As organisations move from AI experimentation to full-scale production, ensuring that a comprehensive semantic and contextual framework that reflects the realities of a business is vital. The organisations which stop treating AI as a “plugin” and start treating it as a deeply integrated, context-aware layer of their enterprise architecture will be the winners.
And when this happens, I suspect we will start to see headlines around ROI, rather than just investment.
