Most boardroom conversations about AI are not one conversation. They are three, layered on top of each other, with no one in the room agreeing on which one is being held. That is why so few of them converge on a decision anyone is willing to sign off.
Pull the three apart and the conversation becomes tractable.
Track 1 — Consumer-style AI in the enterprise
Track 1 is the rollout of general-purpose AI tools to staff: ChatGPT, Claude, Copilot, Gemini. Subscriptions, policy, training, light governance, an acceptable-use document, perhaps a data-residency decision. Wins are individual-productivity wins — document review, presentation drafting, summarisation, code assistance for technical teams, finance commentary.
This work is not difficult. The playbook exists, the supplier market is broad, and the implementation patterns are well understood. Any competent IT services firm can deliver it. The trap is to mistake the volume of internal debate — ChatGPT versus Claude versus Gemini, enterprise tier versus team tier — for the importance of the decision. The value lost through delaying implementation is greater than the value gained by agonising before deciding.
Track 2 — Embedded AI
Track 2 is the AI being built into the software you already own. SAP, Microsoft 365, Salesforce, Xero, ServiceNow, Workday — every meaningful enterprise vendor is shipping AI features as part of their normal release cycle. Most enterprises will receive the majority of their realised AI value over the next two to three years from Track 2, often without quite noticing they have done so.
This is also not difficult work, in the sense that the functionality arrives whether or not you have a strategy for it. The constraint is that Track 2 addresses the process flows your existing software already supports. It makes the work you already do faster. It does not reshape it.
Track 3 — Gap AI
Track 3 is everything in between. The messy seams. The work that today is manual because no system covers it cleanly. The analysis that is partial because pulling the data together is too painful. The decisions that get made on incomplete information because complete information has been too expensive to assemble. Some of this work is being done badly; much of it is not being done at all because the cost of getting there has been prohibitive.
Track 3 is where AI changes what an organisation can do, not just how fast it does it. And because the gaps are specific to each business — its operating model, its data, its customer relationships — Track 3 is also where competitive differentiation actually lives.
The conversation Boards are having versus the one Management is answering
When a Board asks about AI, the outcomes they are implicitly describing are Track 3. They want the organisation to do things it cannot do today, to operate in ways competitors cannot match, to extract value from positions that were not previously economical. They are not asking whether the finance team can write better commentary, faster.
When Management answers, they reach for Track 1 and Track 2 — because those are available, understood, and least organisationally disruptive. The supplier ecosystem is mature. The risks are bounded. There is an off-the-shelf answer to most questions. Management can move forward without first having to make a hard architectural call.
The result is a recurring mismatch. The Board hears about Copilot pilots and feels underwhelmed without quite knowing why. Management feels under-appreciated for delivering against what they understood the brief to be. Both are right; they are answering different questions.
What this means for how to invest
A few things follow.
Do not skip Tracks 1 and 2 — but do not mistake them for a strategy. Track 2 will happen with or without you, because your vendors are doing it. Track 1 is table stakes and should be addressed as such — policy, tooling, training, with the bar set at competent rather than distinctive. Anyone running these as a flagship initiative is signalling that they have not yet had the Track 3 conversation.
Be clear about what Tracks 1 and 2 will and will not produce. Our view, which time will test, is that Track 1 and Track 2 businesses will run at roughly the same pace as each other. The waters will rise, individual productivity will improve, headcount intensity will fall gradually, and the competitive landscape will look much as it does today — slightly more efficient. None of this is bad. None of it is differentiating.
Treat Track 3 as a different category of investment. It requires identifying the gaps in your own operations that matter, and being willing to build, configure, or commission solutions that are specific to them. It is harder, slower at the start, and demands a clearer view of what the business is actually trying to become. The return profile is also different — sustained advantage rather than incremental efficiency.
Expect the supplier conversation to look different. The Track 1 and 2 market is dominated by global platforms, and rightly so at that scale. The Track 3 market is fragmented, specialised, and judged on different criteria. The question is no longer which large vendor you trust; it is who has done the work to understand the seams in operations like yours.
Where Predictiv sits
We have made our investment in Track 3. The Predictiv platform is built around the messy, specific, integration-heavy problems that conventional ERP, BI, and standalone AI tools leave on the floor. We have a view — and the working examples — on how Track 3 work should be done: what standards apply, what the architecture looks like, where to draw the line between deterministic tooling and judgement, and how to keep it all maintainable as the underlying AI models change.
If your Board has been asking the Track 3 question and your team has been answering with Tracks 1 and 2, we would be glad to have the conversation.