64% of all features built in software projects are rarely or never used. Not because they were poorly built, but because they failed to address what people actually needed. Mind you: this figure dates from the pre-AI era. With AI, the risk is even greater.
Without the right context, AI lacks the judgement to challenge your assumptions or question whether you're solving the right problem in the first place.
AI can accelerate processes and amplify outcomes: the good ones and the bad ones. If you start with the wrong question, AI will simply help you build the wrong solution faster.
The good news? Most of this is avoidable. The key is knowing what to do before a single line of code is written. Here's how to set AI up for success and help people work with it, not around it.
Invest in the problem, not the solution
In quality management, there's a simple rule of thumb: the 1-10-100 principle. A problem that costs £1 to prevent will cost £10 to fix during development and £100 to fix once it reaches the customer.
The same applies to AI initiatives. If you spend weeks building a solution before fully understanding the problem, you risk creating something nobody wants to use. There's a quote often attributed to Einstein: whenever he had one hour to solve a problem, he'd spend 55 minutes defining it and five minutes solving it.
A practical place to start is with the people who are doing the work every day. Follow them. Ask questions. Run a workshop. Where do they lose time? Which tasks are endlessly repeated? Where do errors and frustrations occur? Most importantly, which problem would genuinely improve their working day if it were solved?
“Oh no, not another new tool…”
One of the biggest mistakes organisations make, is delivering a ready-made solution and expecting people to adopt it.
But the secret to real adoption is engaging users from day one. The people who will use the system need to help shape it. When employees see their feedback reflected in what is being built, they begin to feel ownership of the solution. This turns a pilot group into something much more valuable: a group of advocates who not only use the solution but will actively help improve it.
But the benefits go far beyond adoption. The people using a system every day notice things developers never will.
Finding the right problem: a real-world example
A national organisation that connects vocational education with industry works with more than 250,000 accredited training companies and 9 sector councils. Its advisors conduct around 144,000 accreditation visits every year.
Each visit follows the same process: preparation, the conversation itself, observations across eight quality dimensions, reporting and registration within an internal system. For several weeks, advisors were observed in their day-to-day work. The team followed them, conducted interviews, ran workshops using sticky notes and whiteboards, and even facilitated role-playing sessions.
The outcome was a detailed journey map showing everything an advisor does before, during and after a visit. What worked well? What caused friction? What motivated people, and what slowed them down?
You don't need a large-scale research programme to achieve the same result. Simply take one process and map it with the people involved. Identify three things:
Where time is being wasted
Where quality varies unnecessarily
Which information is difficult to find
The result is a far more useful starting point for AI than asking: "Where can we apply AI?"
What the research uncovered
The biggest opportunity turned out to be reporting. Writing up visit reports took advisors an average of 30 minutes per conversation. Across 144,000 visits, that amounts to roughly 72,000 hours of administration every year.
Important work, yes. But most advisors described it as a chore.
The research also revealed that:
Reporting quality was inconsistent because everyone worked slightly differently.
Advisors were sometimes assigned cases outside their primary area of expertise without sufficient support.
Preparing for visits was time-consuming because information was scattered and difficult to access.
Only once these root causes were understood did the team begin discussing technology. Problem first. Solution second. AI was not the goal. It was simply a means to an end.
Start small, learn fast and scale with purpose
One of the first solutions developed was an AI-powered conversation report. Immediately after a visit, advisors record a short spoken reflection, often while travelling to their next appointment.
The process works as follows:
The advisor records a spoken reflection.
Speech-to-text technology converts the recording into text.
An AI agent reviews the report against the eight quality dimensions.
The advisor reviews the draft, approves it or makes adjustments.
The final version is stored automatically in the internal system and CRM.
This approach keeps the advisor in control while using AI to support and strengthen their expertise. As a result, reporting time fell from 30 minutes to 15 minutes per visit. Rolled out across the entire organisation, that reduction represents a potential saving of 36,000 hours per year.
The same approach is now being applied elsewhere in the process. AI also assists advisors with preparation by bringing together relevant information from multiple sources. Advisors remain responsible for validating the information and determining whether it is useful. In this way, a single use case gradually evolves into broader support across the entire workflow.
Three lessons for your organisation
1. Start with the problem, not the technology
What problem is genuinely worth solving? Invest time in answering that question before looking at tools or platforms. The earlier you identify the right problem, the less time and money you'll spend fixing mistakes later.
2. Build a platform, not a collection of pilots
Pilots often deliver isolated value but rarely scale. A modular architecture allows each new application to build on existing foundations, reducing costs and accelerating deployment.
3. Involve employees from the outset
Don't make people users of a system. Make them co-owners of the solution. That's the difference between a tool that sits unused and a new way of working that benefits the entire organisation.
Empower your people
AI doesn't replace people. It makes them more effective, provided the foundations are in place: the right problem, the right context, and people who feel ownership of the solution.
When those elements come together, AI becomes more than a tool. It becomes a force multiplier for the expertise that already exists within your organisation.
