Why Organisations Are Only Getting 20% Value From AI Tools
AI tools such as Microsoft Copilot and ChatGPT are now widely deployed across organisations. On paper, this should be transformative. In practice, most companies are capturing only a fraction of the value they’ve paid for.
I’ve seen this first-hand.
While at Praetura Ventures, I introduced Copilot into the business. Like many organisations, the ambition was clear: improve productivity, automate repetitive tasks, and support better decision-making. What followed was far more frustrating. The automations and agents I wanted to build didn’t work. Outputs were inconsistent. The promised efficiency gains never really materialised.
At the time, it felt like a tooling problem. Looking back, it clearly wasn’t.
Why early AI initiatives often fail
The reality is simple: I had no formal training in how to use these tools properly.
What I was doing at the time, I now describe as playing. Experimenting. Trying things. Hoping something would stick. Failure was almost inevitable.
After leaving Praetura, that experience became the catalyst for me to properly learn how modern AI tools actually work, not at a surface level, but in terms of prompt structure, persona design, and how to architect repeatable solutions. With that understanding, it became obvious why my earlier Copilot experiments struggled. The foundations simply weren’t there.
AI tools are like Excel and we keep making the same mistake
The best analogy I’ve found is Excel.
Most organisations have had Excel licences for decades. Very few employees have ever been trained beyond the basics. As a result, most people use perhaps 10–20% of Excel’s capability. The rest goes untouched.
AI tools are heading down exactly the same path.
Companies roll out Copilot or similar tools, provide little or no training, and then wonder why adoption plateaus and impact is limited. People default to basic usage such as drafting text, summarising documents, asking ad-hoc questions and never move beyond that.
From a capital allocation perspective, this is bonkers. Organisations are effectively using 20% of the capability they’ve already paid for.
The real constraint isn’t technology — it’s capability
The good news is that this is not a complex problem to solve.
It doesn’t require hundreds of hours of experimentation. It doesn’t require deep technical knowledge or specialist teams. What it does require is that people are shown what good looks like.
Specifically:
How to structure prompts properly
How to set personas so outputs are consistent and relevant
How to think about prompt architecture rather than one-off queries
How to embed AI into existing workflows rather than treating it as a novelty
Once teams understand this, output quality improves quickly. Efficiency gains follow naturally.
Why training alone isn’t enough
There is a second, less obvious issue: habit and mindset.
Good AI use often requires spending a little more time upfront around thinking clearly, structuring instructions, being explicit about context and output. That runs counter to how most people approach new tools. The instinct is speed, not structure.
Mindset shifts come with understanding. Once people see why better prompts lead to materially better outcomes, behaviour changes. Without that understanding, AI remains a superficial add-on rather than a genuine productivity lever.
This is also why many early agentic AI initiatives fail. Without strong foundations in prompt design and architecture, layering automation on top simply magnifies inconsistency.
What this means for organisations
Most organisations don’t have an AI tooling problem. They have an AI capability gap.
Until leadership teams and employees understand how to work effectively with natural-language tools, broader AI strategies will underperform. The sequence matters: capability first, then automation.
Looking back at my early Copilot experiments, failure wasn’t a surprise, it was predictable. I just didn’t have the tools or understanding at the time to see it.
A final thought
AI will increasingly shape how businesses operate, but value won’t come from access alone. It will come from how well people are trained to use the tools they already have.
Show teams what good looks like. Build the right habits. Get the foundations right. Everything else builds from there.
This belief now underpins my advisory work and corporate AI training, not focused on tools for their own sake, but on building practical capability that actually delivers
