Few technologies have created as much urgency in boardrooms as AI. For business leaders, there is a genuine sense that the decisions being made now could determine who pulls ahead over the next decade.
AI is already changing customer interactions, data analysis, and everyday workflows. Agentic AI goes a step further, shifting systems from assisting people to completing tasks on their own.
McKinsey estimates that agentic AI could generate $450 billion to $650 billion in additional annual revenue across advanced industries by 2030, representing a potential 5 to 10% revenue uplift.
There is another number worth paying attention to. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls.
Having worked with AI across the automotive industry in more than 25 markets, I’ve found that the technology itself is rarely the hardest part. The more difficult questions tend to sit underneath it, such as: what data can AI actually access, how well your systems work together, how it fits into the way your people operate, and who is accountable when AI acts independently?
This challenge isn’t unique to automotive. It surfaces anywhere customer journeys cross multiple systems and teams. Almost every business must now ask themselves if they are building a cohesive AI capability, or simply accumulating standalone tools? This distinction will define long-term success over the next decade.
AI can’t see what your business doesn’t connect
In automotive retail, a dealership’s information is spread across CRM, inventory, websites, finance, test drives, service history, marketing, and other systems. Customers move between these touchpoints without thinking about the technology behind them, while the systems supporting that journey often remain disconnected. We are asking AI to deliver a more intelligent, personalised customer experience while giving it access to only part of the information it needs to understand the customer.
An AI assistant might know what vehicles are in stock but have no idea that the customer has already spoken to a salesperson. It might know the customer’s enquiry history but not their service history or recommend a vehicle without being able to book a test drive. None of this is really a problem with AI. Its usefulness is limited by the information and systems it can access.
The same challenge exists across almost every industry. A bank, retailer, or manufacturer can have important information spread across systems that were never designed to work together. If the information needed to make a decision isn’t connected, an AI agent will inherit that limitation.
The phrase “AI strategy” can be misleading. Before building one, you need to understand your customer data, your integrations, and the processes you are actually trying to improve. The better those foundations are, the more useful AI becomes, and the more likely it is to still be relevant a decade from now.
The risk of buying AI one use case at a time
The AI market is moving so quickly that it is easy to end up buying technology reactively. A vendor comes along with an interesting use case, while another team wants to solve a different problem, and someone else sees a new model and wants to try it. Before long, AI can be sitting in marketing, sales, customer service, and operations, with very little connecting the systems.
I’ve seen this happen with other technologies before. A team solves one problem today, only to create more complexity tomorrow. Businesses should experiment with AI. Some of the best applications come from teams trying things that weren’t obvious a year ago. However, there is a big difference between experimenting with AI and building a strategy around a collection of unrelated tools. The latter quickly becomes expensive.
When looking at an AI investment, I would start with the immediate problem it solves and ask what it enables the business to do next. Does it connect another useful source of data? Does it improve a process that another AI system could build on later? Does it make it easier to introduce another workflow? If the answer to these questions is no, the investment is not worth making, since a tool that solves only today’s problem is unlikely to be relevant a decade from now.
Agentic AI changes the stakes
Moving AI from passive helper to active participant changes how businesses operate. Assisting someone with an email is straightforward. However, when AI begins choosing recipients, pulling account history, updating the CRM, and booking follow-ups, it takes charge of a broader workflow. The economic value shows up right here.
Businesses have thousands of processes that are important but don’t necessarily require a person to make every individual decision. AI can handle more of those processes while people focus on the decisions that genuinely require judgement. For this to work, businesses need to be clear about what AI can see, what it can do, and where a person needs to step in. Those decisions need to be made as part of the design, rather than once the technology is already deployed. Human approval shouldn’t be required for every minor action. If it is, we defeat the purpose of having an agent in the first place.
Governance also needs to reflect the level of autonomy an AI system has. Gartner warns that applying the same governance model to every AI agent, regardless of its autonomy or access, can in itself lead to failure.
Ownership matters just as much. AI cannot be an IT project if it is changing how the sales team sells cars, how the service department manages appointments, or how customers interact with the business. The people running those operations need to be involved in deciding what AI should do, while being accountable for the outcomes.
The model isn’t the advantage
There is a lot of attention on which AI model businesses should use to future-proof themselves, yet fundamentally this will keep evolving.
What is much harder to alter is everything around the model: customer data, processes, integrations, an understanding of how the business operates, and the feedback collected from every interaction. Getting that infrastructure right is what creates staying power.
If an AI system handles ten thousand customer interactions and those interactions improve the data and processes behind the next ten thousand, you have something that can compound. An AI system that simply completes a task in isolation may still deliver a return, but the value is much more limited.
A better question than “What can AI do for us today?” is “What will this help us do tomorrow that we can’t do today?” Using that simple benchmark will tell you if an AI investment is actually strategic.
