Somewhere in your organisation, there is a person who knows why the month-end report gets rebuilt by hand every time. They know which customer always needs a phone call before the system's answer can be trusted. They know that one approval step exists because of an incident nine years ago that nobody else remembers.
They are also, almost certainly, too busy to tell anyone.
That is the quiet problem at the heart of most AI programmes. The technology is bought, piloted and announced. The people who understand the work carry on delivering it at full stretch, hoping the new tool will somehow fit around them.
The value was never in the go-live
The gap between AI investment and business value is now well documented. McKinsey's March 2025 state of AI survey tested 25 organisational attributes and found that redesigning workflows had the biggest effect on whether companies saw bottom-line impact from generative AI. Yet only around one in five respondents said they had fundamentally redesigned any workflows. That is a correlation, not proof of cause, but it is a striking one. BCG's own guidance points the same way: roughly 10% of the effort in successful AI programmes goes to algorithms, 20% to technology and data, and 70% to people, processes and the way the organisation works.
Most budgets are shaped as if those numbers were reversed.
The hidden work
Two kinds of unglamorous work sit underneath almost every AI result. This week's headline in The Asset puts it plainly for financial institutions: AI value begins with data quality and process redesign. We would extend that well beyond financial services.
The first is data. Gartner predicts that through 2026, organisations will abandon 60% of AI projects that are not supported by AI-ready data, and its own survey found 63% of organisations lack, or are unsure they have, the right data management practices. A model can only reason over what it can find and trust. Somebody has to decide what the data means, where it lives and who stands behind it.
The second is process. AI cannot compensate for a process nobody has examined. Automate a convoluted workflow and you simply get convoluted work, faster. A report that is drafted in minutes but still waits in a queue for three sign-offs has not been transformed. It has been accelerated into the same bottleneck.
There is a real difference between making a process faster and asking why it exists in this form at all. Only the second question creates new value.
Why it gets missed
None of this is hidden because leaders are careless. It is hidden because of how progress gets measured.
Pilots have launch dates. Procurement has milestones. Deployment can be announced. Data remediation and process redesign have none of that theatre; they are slow, specific and rarely applauded. Accountability fragments too. Technology owns the tool, data teams own the pipelines, the business owns the outcome, and nobody owns the work in between.
Underneath sits a comforting assumption: that AI can be layered onto the way we already work. It is the belief that implementation equals transformation. It doesn't. Implementation changes what an organisation owns. Transformation changes how it works.
The people who know
If one part of this work is neglected more than any other, it is this: the experienced people who understand how the organisation really runs.
Their knowledge rarely appears in any process map. They know the exceptions, the dependencies and the risks that never got written down. They know which steps are genuinely necessary and which survive purely because "that's how it's always been done". And they know where a workaround is not a flaw but a record of the organisation learning something the hard way.
That is why they cannot be treated as stakeholders to be consulted, or as a knowledge source to be mined before the automation goes live. They are co-designers of the future operating model. Their judgement tells you where AI can add value, where a human must stay in the loop, and where the work itself should change.
There is also a human truth here that no framework captures. Asking people to help redesign their own work, without time, without a say in the outcome and without clarity on what it means for them, is not engagement. It is extraction. People rarely resist change they have helped to shape. They resist change that is done to them.
The capacity problem
This brings us to the uncomfortable centre of the matter.
You cannot transform work if the people who understand that work never have the time to rethink it.
Most experienced employees are already fully committed to delivering today's results. Asking them to redesign tomorrow's work in the margins of their week is asking for something that cannot happen. It will either not be done, or be done superficially and late.
So leaders have to create the conditions deliberately. That means protected capacity, with permission to step back from operational delivery. It means putting technology, data and business expertise in the same room, with a mandate to examine and challenge how work really happens. It means clear ownership for turning what is learned into actual changes. And it means treating all of this as a core workstream of AI transformation, not an optional extra that gets squeezed when deadlines tighten.
A useful test is to look at the calendar of your most experienced operator. If the redesign work is not visibly in it, it is not really funded. This is not overhead, and it is not a delay to implementation. It is where the return is made.
Where value actually comes from
Technology is an enabler. Value arrives when roles, workflows, decision rights and management practices change so that the technology can be used well.
Without that foundation, organisations end up with isolated AI tools, modest productivity gains and extra complexity. With it, they get genuinely different ways of working: clearer accountability, better judgement applied where it matters, and human and AI capability designed to complement each other.
That is the real distinction. Implementing AI is something an organisation does. Knowing how to work effectively with AI is something an organisation becomes.
At Bridge Advisory, we call this territory the Adoption Layer: the organisational work that sits between AI implementation and AI-enabled performance. The adoption challenge is fundamentally an operating model challenge, and it is solved role by role and team by team, with the people who understand the work in the room.
If you are investing in AI and want to understand what it will take to turn that investment into real business value, we would welcome a conversation.
Katy Wright
Evidence referenced:
McKinsey, The state of AI (March 2025 survey, 25 attributes tested; 21% had fundamentally redesigned some workflows): https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
BCG, 10-20-70 principle: https://media-publications.bcg.com/219-Weekly-Brief-What-It-Takes-to-Win-with-AI.pdf
Gartner, Lack of AI-Ready Data Puts AI Projects at Risk, 26 February 2025: https://www.gartner.com/en/newsroom/press-releases/2025-02-26-lack-of-ai-ready-data-puts-ai-projects-at-risk
The Asset, "AI value begins with data quality and process redesign", 5 November 2026: https://www.theasset.com/article/57467/ai-value-begins-with-data-quality-and-process-redesign