The easiest way to waste an AI programme is to let every team call its own shots. One group buys a tool, another runs a pilot, a third builds a chatbot nobody asked for, and by quarter-end the organisation has a slide deck full of motion and almost nothing it can point to on a balance sheet.
That is the pattern Colin Nelson was circling in a recent session with Gavin McClafferty of Subsea7 and John Toon of HLB. The problem is not a shortage of ideas. It is the familiar corporate talent for producing activity without coordination. AI just makes that habit louder, faster, and more expensive.
The pilot graveyard is getting crowded
Most companies have already crossed the first hurdle. They have pilots running, teams experimenting, and a few brave people testing tools across departments. Then the programme hits the same wall. The experiments are disconnected. The use cases are interesting but not tied to the business plan. Nobody can quite say which projects should scale, which should die, and which are just eating time.
That is where AI differs from the usual innovation theatre. It spreads quickly. It is easy to access. It gets adopted in pockets before leadership has properly decided what problem it is meant to solve. Nelson’s point was blunt enough: the technology is rarely the issue. The organisation is.
In regulated or risk-heavy environments, governance adds another wrinkle. Data rules, compliance checks, and security reviews can become a traffic jam if they are bolted on after the fact. The result is a familiar corporate compromise. Everyone agrees AI matters. Few can explain how it will produce measurable value without turning into a permanent side project.
Subsea7 started with the problem, not the tool
Subsea7 did not begin with a grand AI strategy document and then go looking for a business reason. Gavin McClafferty described a more practical route. Leadership first identified the issues that were genuinely bothering teams, the ones that were already costing time or creating friction. Those pain points were then translated into specific AI opportunities.
The company also pushed the search outward. An internal campaign pulled ideas from engineering, supply chain, HR, and other parts of the business. That matters because a central innovation team on its own will always miss half the useful problems. The volume of engagement, McClafferty said, was stronger than anything they had seen before, which is usually what happens when staff stop being asked for abstract ambition and start being asked for fixes to real work.
From there, the process became more disciplined. Ideas were not treated as equal. Early concepts were explored quickly. Strong ones moved into proof of concept. Better ones became minimum viable products. Only after that did they earn a route to production. That sequence sounds almost boring. It is also the part most companies skip, then wonder why nothing lands.
HLB chose focus over sprawl
John Toon described a similar shift at HLB. The temptation, early on, was to explore everything. That is a common mistake. It feels energetic. It is usually just undirected curiosity with a budget.
HLB moved instead toward experimentation with intent. The point was to stop treating AI as a broad promise and start treating it as a series of testable choices. In a more structured or distributed business, that is a useful correction. It forces teams to decide what matters before they spend six weeks fiddling with tools that never get used.
The effect is less glamorous than the AI hype cycle suggests, but more useful. Teams make cleaner decisions. Promising ideas can be validated. The business gets a narrower set of opportunities, but better ones. That is generally preferable to a long list of mediocre experiments dressed up as innovation.
AI is changing the work, not just the speed
Most corporate AI conversations still begin and end with efficiency. Faster workflows. Lower costs. Less manual effort. Fine. Those gains are real, but they are not the whole story.
At Subsea7, McClafferty pointed to the scale of data that sits inside a large organisation and never gets used properly. AI changes that. Engineering data that once took a long time to sift through can now be processed immediately. Teams can surface similar past projects, identify relevant specialists, and pull out useful context without digging through systems by hand. That is a different operating model, not just a quicker one.
HLB is seeing a similar shift in research-heavy work. Tasks such as assessing a new client or getting a handle on an unfamiliar sector used to take days. Now they can be done in minutes, and with more depth than the old manual process allowed. The next layer is more interesting still. Multiple AI agents can now split parts of a workflow between them, and rough ideas can become working prototypes in hours rather than weeks.
That compresses time in a way most organisations are not prepared for. The old standard for value was whether a tool saved a few hours. The new standard is whether it changes how quickly the business can see, decide, and act.
Culture will decide who scales and who stalls
The hardest part is still human. Across any sizeable organisation, reactions to AI sit everywhere from enthusiastic to openly hostile. Some people lean in. Some hover around the edges. Some are waiting for the whole thing to blow over.
McClafferty’s view was straightforward. Leadership has to bring people along, not just push adoption from the top. If employees see AI as something being done to them, the rollout slows. If they understand where it helps and where it does not, momentum is easier to keep.
Toon addressed the job anxiety directly, which is more useful than pretending it is not there. If AI can already handle parts of process-driven work, ignoring that fact does not protect the role. It only delays the adjustment. People do adapt, but only if the organisation gives them enough room to experiment, learn, and build confidence with the tools.
That is where culture stops being a poster and becomes operating reality. AI adoption is not a software purchase. It is a shift in how people judge work, ownership, and capability.
The organisations that win will run AI like a portfolio
The most effective teams are not the ones with the biggest number of pilots. They are the ones that can connect use cases, governance, experimentation, and people into one operating system.
Subsea7’s model shows the shape of that. A small core team drives work across the business. A wider stakeholder group helps assess and prioritise. The company keeps just enough structure to move fast without losing control. HLB takes a similar line, using clear frameworks for tool selection, data handling, risk, compliance, and training rather than trying to control every move from the centre.
That balance matters. Too much experimentation creates noise. Too much governance creates inertia. Put the two together properly and AI starts to look less like a shiny add-on and more like a portfolio of working bets.
The companies that get there first will not be the ones with the flashiest demos. They will be the ones that know which problems matter, can test ideas quickly, and have the discipline to turn the useful ones into production. That is the unglamorous part. It is also the only part that pays.