Part 1: In the AI Race, Trust is King

Note: This is Part 1 of a planned 3-part blog series on artificial intelligence (AI). 

AI doesn’t create intelligence. It amplifies what already exists. That single observation should be the starting point for every AI investment decision your organization makes, but it rarely is.

Instead, most organizations start somewhere else entirely and with urgency. Boards are asking why other companies seem to be doing more with less. Executives have experienced AI firsthand whether it is synthesizing a competitive analysis in seconds, turning an hour of meeting notes into a clean action plan before they got back to their desk or watching a developer vibe code a working prototype over a weekend, perhaps even with tools like ChatGPT. The technology is genuinely impressive. The pressure for AI adoption is entirely legitimate.

But urgency without deliberateness is how organizations end up doing what I’d call expensive guessing by deploying AI onto fragile foundations, generating confident-sounding outputs that turn out to be wrong, and discovering the cost of that only after trust has already been lost.

The Gap Nobody Talks About Honestly

There is a version of AI that most executives have experienced: personal, immediate, often astonishing. And there is a version of AI that actually has to perform inside a complex enterprise: reliable, auditable, accurate across hundreds of users with different roles, different workflows, and very little patience for tools that occasionally get it wrong.

Those are not the same thing. And the distance between them, between a prototype that works brilliantly in a demo and a system that holds up under the weight of real enterprise use is where most AI investments quietly fail.

The what and the why always dominate these conversations. The how and what it actually takes to do this right, across real workflows, with real users, in environments that do not forgive errors almost never gets the same honest attention. That is where AI investments go wrong.

The result is predictable. A lot of money is spent, frustration sets in quickly, and the roadmap that was never built properly at the start becomes very expensive to fix later.

Three Failure Patterns and Why Moving Faster Won’t Fix Them

A 2025 Gartner survey of over 700 CIOs found that 72% reported their organizations were breaking even or losing money on AI investments.¹ This is not a model problem. The models are remarkable. It is a decision-making problem specifically, the decision to move before the harder questions have been asked.

I see three failure patterns repeat across industries, and none of them are solved by accelerating the timeline.

The first is picking use cases based on excitement rather than impact. Organizations launch multiple AI initiatives in parallel, frame them as innovation projects rather than business outcomes, and end up with a collection of successful pilots that fail to scale. A pilot that cannot survive contact with the full complexity of your operations is not a success. It is a deferred problem.

The second is what I call the illusion of readiness. The assumption: we have an ERP, we have a data warehouse, AI can pull it all together. The reality: ERPs are transactional systems. They capture what happened, not the full context of why, how, or what the relationships between data objects actually mean. The data is present but it is siloed, inconsistent, and missing the connections that AI needs to reason reliably. Buying an AI tool and expecting it to magically bridge organizational data silos is one of the most common and costly mistakes I see.

The third is underestimating the human side. How difficult it is to implement a solution. How difficult it is for users across different roles whether a food scientist, a packaging manager or a compliance officer to actually change how they work, not just adopt a new tool. Organizations run these as technology programs and wonder why users tune out. The functions need to own it. Every persona experiences the change differently, and each one needs a different reason to believe it makes their work better.

Trust Is the Asset You Cannot Afford to Lose

There is a deeper risk that does not show up in any budget conversation but shapes everything: trust.

When an AI solution produces a wrong answer,  it does not just create a compliance problem or a cost. It creates an asymmetric trust equation that is very hard to recover from. It takes months to build the kind of trust that makes an organization actually change how it works. It takes one bad experience to destroy it. Users who lose confidence in an AI tool do not give it another chance. They work around it, quietly, and the investment calculates to zero.

Your competitive advantage in the AI era will not come from the model you choose. Every one of your competitors has access to the exact same base AI capabilities, whether from OpenAI, Anthropic, or others. It will come from how deliberately you build the foundation those models run on.

AI can be very confident even when it is not accurate. That combination of being confident and wrong is uniquely damaging in an enterprise context. It is far more corrosive than a system that is obviously limited, because people stop checking.

That is the real question most organizations are not asking. Not which AI tool to deploy. Not how fast to move. But: what does it actually cost to build an AI solution that works reliably at scale and do we understand what we are truly committing to?

That is the question Part 2 of this series takes head on. Because building an AI solution that actually works at scale is harder than almost anyone tells you and the gap between what your team can prototype and what your organization can sustain is where most of the real cost lives. 

In the meantime, download my latest AI Executive Brief to learn more about my perspective on why most AI projects fail or contact our team to learn more about our spec-first approach.