“Businesses have spent $30-40bn on generative AI – 95% of initiatives show no measurable P&L impact” (MIT)

It’s a truism, the devil is always in the detail.  AI is not the problem, it’s the ‘pipework and plumbing’ of AI into existing business structures, 3rd party systems and daily operational ways of working, where it can all fall apart. 

An AI demonstration can produce a convincing result in minutes. But, making that result useful to a business means connecting it to the information, systems and people responsible for getting the work done.  It’s not easy and that is where the metaphorical pipes start to leak. 

Much of our current client work is here - connecting AI with existing systems and data, and defining outputs people can actually use. The quality of those connections determines whether a promising idea becomes a dependable and valuable part of a business. These practical applications befuddle both the tech and business teams in equal measures and leave leaders of the business between a rock and a hard place.

In July 2026 analysis by the Office for National Statistics reported on the cost of not getting wider adoption right.  Only 10% of UK businesses with ten or more employees, using AI, said they used it extensively. The ONS cautioned that these new, self-reported measures were early indications that better implementation/wider adoption needs to be addressed.[1]   

Start with the outcome and follow the work

Our client work, is a tried and tested formula to ensure businesses unlock the transformational value of AI. 

Consider a team preparing a customer proposal. AI may generate a persuasive draft, but the team still needs accurate prices, current product information, delivery capacity and approval from the right people. If those inputs sit in separate systems, a fluent answer can leave most of the underlying work untouched. Or worst case, getting it wrong.

Our start point is to define the output and its destination. Who will use it? What decision will it support? Which information must be current? What happens when the AI cannot resolve an exception? An answer displayed in a chat window is only useful if someone can act on it.

This brings the less glamorous ‘plumbing’ work into view - agreeing which system holds the authoritative record, giving people and agents appropriate access, connecting applications, and deciding how errors will be handled. These choices need, technical understanding, the involvement of the people who run the process and a very good ‘plumber’.

The same applies to collaboration. Sales, finance and operations bring different knowledge to a proposal. A shared conversation with AI allows them to contribute, challenge assumptions and develop the work together. Our AI native tech stack, with 6 layers of architecture, Karrik, supports that real-time collaboration between groups of people and AI.

Connect the workflow to a larger ambition

A useful individual application project, can be the first step in a broader strategy. The leadership task is to be guided to where that step should lead and then make that decision.

For example, reducing the time spent producing a report may be worthwhile. A more ambitious opportunity might be a service that gives customers timely insight and helps them respond as circumstances change. That would involve changes to the offer, the workflow and the responsibilities of the people delivering it. The AI component alone would not create the new service.

That is how our discussions about upgrading a task can develop into a strategy for changing how the business competes. The opportunity must still be tested with customers and supported by workable economics.

Scaling AI means making the complete process repeatable across teams. Measure the time taken from request to useful outcome, the amount of correction required and the quality of the result. A faster intermediate step has limited value if it creates more work somewhere else.

At Informed.AI, we help leadership teams connect their AI ambitions with the operating changes needed to deliver them. Bring us a workflow that matters to your customers or teams, and we can make AI deliver a lasting difference, with no leaky ‘pipes’.

Discuss your AI strategy at informed.ai, or explore collaborative AI at karrik.com.

Sources

[1] Office for National Statistics. Artificial intelligence in UK businesses 2023 to 2026. 20 July 2026.

About the author

Professor Andy Pardoe is a Professor of AI and founder of Informed.AI. He brings more than 30 years of technology, business and management experience to his work with CEOs and boards on AI strategy, governance and adoption.