Almost every company now has an AI roadmap. Walk into any leadership offsite and you will find a slide of initiatives — summaries in the CRM, a drafting assistant for support, a copilot in the analytics tool, a document-search bot for the knowledge base. The list is real, the budget is committed, and the intent is genuine. And most of it is AI-enabled, not AI-native — a distinction that matters precisely because the two are funded as if they were the same thing.

The organising question: if you removed the AI layer from this initiative tomorrow, would the business still work roughly as before — or would something structural be missing? If the honest answer is “it would work, just a little slower,” the initiative is AI-enabled. That is not a criticism. It is a description of how much is actually at stake — and therefore how it should be funded, staffed, and measured.

— The distinction

Additive versus structural.

An AI-enabled initiative adds intelligence to an existing workflow. The product category, the business model, the data advantage, and the org chart all stay the same; AI makes a known process faster or cheaper. These are often worth doing — sometimes they are necessary just to stay competitive. But they improve the old model rather than change it.

An AI-native initiative is one where intelligence becomes a core operating layer. The workflow is redesigned around what reasoning, retrieval, and bounded action make possible; a proprietary data loop compounds with use; manual coordination is replaced by orchestration; and the economics or the defensibility of the activity actually shift. AI is not additive here — it is structural.

The failure mode is not investing in AI-enabled work. It is investing in AI-enabled work while believing — and budgeting as if — it were AI-native. That is how an organisation ends up with a roadmap full of genuinely useful features and a board that quietly wonders why none of them moved the strategic needle.

— The test

Eight questions, asked of each initiative.

The same diligence an investor runs on an AI startup can be turned inward, on the initiatives you are funding. Run each one through these:

01
Central or peripheral?
Would the product or process still be valuable if the AI layer were removed? If yes, it is AI-enabled — fund it as an efficiency play, not a strategic bet.
02
Does it change the workflow?
Is the team doing the same thing faster, or a genuinely new thing? Workflow redesign is the signal that separates structural from cosmetic.
03
Is there a proprietary data loop?
Does usage generate corrections, outcomes, and context that improve the system over time — or does it depend entirely on a third-party model getting better?
04
Is governance designed in?
Evidence, confidence, permissions, approvals, and an audit trail — especially anywhere the initiative touches money, customers, or compliance. Their absence caps how far it can scale.
05
Are the economics honest?
How do inference, human review, and integration costs behave at real volume? AI-native does not automatically mean high-margin; the unit cost has to close.
06
Is it model-resilient?
What happens when the frontier providers ship this capability natively? A strong initiative benefits from model progress; a weak one is one release from obsolescence.
07
Is there a named owner and a path to adoption?
A technically strong initiative still fails commercially without a single owner, a budget, and a real adoption path inside the organisation.
08
Could it define a new operating layer?
Is this a feature inside someone else’s platform, or could it become the system of record, engagement, or intelligence for a part of the business? Only the latter compounds.
— The trap

AI theater, and how to spot it.

The most expensive pattern in enterprise AI is what might be called AI theater: the language of transformation applied to work that changes nothing structural. It clusters, and the cluster is the signal:

  • ▲Generic AI claims with no workflow named. “We’re using AI across the business” — but no single decision or process is materially different.
  • ▲No evidence of ROI. Activity is measured (pilots launched, tools adopted) but outcomes are not (cost removed, cycle time cut, revenue moved).
  • ▲No data advantage. The initiative processes generic inputs and produces generic outputs; nothing is learned that a competitor could not also learn tomorrow.
  • ▲No governance model. Nobody can answer what happens when the system is wrong, who approved it, or how to prove what it did — so it never leaves the sandbox.
  • ▲No answer to the incumbents. The differentiation is “we added AI,” which every competitor and every platform vendor is also doing.

AI-enabled improves the old model. AI-native creates a new one. Most roadmaps are quietly the first while being budgeted as the second.

— What to do with the answer

Fund each initiative as what it is.

The point of the test is not to kill AI-enabled work. Efficiency features are often the right thing to ship, and a portfolio of them can add up to real margin. The point is to stop mispricing them — to fund the AI-enabled initiatives as efficiency plays with quick paybacks and modest expectations, and to reserve strategic patience, executive air cover, and governance investment for the genuinely AI-native ones that can change the structure of advantage.

That is a portfolio decision, and it is exactly the decision most roadmaps skip. A roadmap is a list; a portfolio is a set of bets sized to their nature. Sorting your initiatives into AI-enabled and AI-native — honestly, initiative by initiative — is the first step from one to the other. Where to set the threshold, how to score value against feasibility and risk, and which three to fund this quarter is the work of the ROI Portfolio Map; how each consequential call gets made, on what evidence and through what gates, is AI Decision Architecture.

The enterprises that compound on AI are not the ones with the longest roadmaps. They are the ones honest enough to know which items on the list are features and which are foundations — and disciplined enough to fund each accordingly.

Sort your roadmap, honestly

Thirty minutes with Liron to run your current AI initiatives through the AI-native test — an outside read on which are efficiency features, which could be structural, and where the governance and data gaps are that decide the difference.

Book an AI Decision Call