Before you buy another AI tool, map your decisions.
The Bank of Canada published a number in August that should stop every executive in the country. More than two-thirds of Canadian business leaders say they personally use AI tools during a typical work week. Yet only 8% of businesses say they use AI significantly in their core operations. Another 50% report low or moderate use.
The boss uses it. The business does not.
Why AI adoption looks like both 20% and 90%
Here is a problem you have probably run into without naming it: every AI adoption statistic seems to contradict the last one.
In the United States, nationally representative Census Bureau data put business AI use between 17% and 20% from December 2025 through May 2026. Among firms with at least 250 employees, it reached 37%. Across the European Union, 20% of enterprises used AI in 2025, up from 13.5% a year earlier, and 55% among large enterprises. Then Stanford's 2026 AI Index, drawing on McKinsey's organizational survey, reports AI use among 88% of surveyed organizations globally, including 91% in Europe and 90% in North America.
Twenty percent. Eighty-eight percent. Both can be true. They count different things. Eurostat measures enterprises across the economy above a defined employee threshold. Stanford's figure comes from an organizational survey rather than a representative census of all enterprises. Different populations, different methodologies, different definitions of use.
They point to the same distinction. Using AI somewhere inside a company and integrating it deeply enough to change how that company operates are not the same thing.
McKinsey's 2026 research makes the gap even sharper. Eighty percent of respondents say AI has improved their individual productivity. Only 37% report positive enterprise-level EBIT impact from AI, and just 6% qualify as high performers.
Using ChatGPT is not AI transformation. Giving AI a defined role inside your operating model is.
One reason so many businesses have not crossed that gap is surprisingly basic: they have never formally decided where the machine is allowed to decide, where it should prepare the work, and where a human keeps the last word.
Sort by reversibility and frequency, not importance.
When leadership teams look for places to deploy AI, they tend to start with importance or complexity. I think that is the wrong place to start.
A sales forecast rebuilt every night can involve thousands of data points, yet it can be relatively safe to automate, because a bad forecast on Monday can be replaced by a corrected one on Tuesday. Sending the wrong response to a major client is technically simple. Undoing the damage may not be. Complexity tells you how hard the work is. It tells you nothing about what happens when the machine gets it wrong.
So I start with two questions instead. How reversible is the decision, and how fast can you undo it if it is wrong: an hour, a week, a quarter, never? Then, how often does it happen: once a year, once a month, fifty times every morning?
What is the cost of being wrong, and how fast can you take it back?
That changes where you look for value. Many AI initiatives start with the technology and work backward to find a use for it. The map starts with the decision and asks how much authority the machine should actually have.
The map tells you where to look. Not how much rope to give the machine.
Reversibility and frequency tell you where to look. They do not, by themselves, tell you how much authority to hand the machine.
Picture one incorrect customer decision reversed within an hour. Now picture the same system making that decision 40,000 times before anyone notices.
Same reversibility on paper. A very different morning.
Before you grant real autonomy, four more things matter: blast radius, data quality, regulatory exposure, and human escalation. How far can one mistake travel? How much do you trust the information behind the decision? What could a wrong call cost you legally or operationally? And when something falls outside the system's confidence or its rules, how cleanly does authority return to a person?
That is the next layer, and it is Episode 2. For now, the map is enough to change where you point AI this quarter.
You need your leadership team, a whiteboard, and 90 minutes.
You do not need a consultant, a new platform, or an AI budget to start. List 20 recurring decisions your company made during the last 30 days. Not projects. Not job descriptions. Decisions. "Should we follow up with this lead?" "Which creative gets more media spend?" "Does this client request need escalation?" "Do we reorder this inventory?" Write one decision per card.
The tool question comes later. The decision question comes first. In most companies the pattern on the wall rhymes. AI is parked in Zone 2, generating documents someone reviews afterward. It sometimes wanders into Zone 4 carrying more authority than it has earned. And it is missing from Zone 1, where repetitive, reversible calls could start returning hours this week.
You should see the imbalance before the meeting ends. The map makes the invisible obvious.
Do not leave the room and launch 20 automations. Move one. Give it 30 days. Track human time recovered, error rate, and exceptions requiring a person.
If automation saves five hours but creates four hours of exception handling somewhere else, you did not come out ahead. You moved the work. That is one of the easiest ways to mistake automation for transformation. If human time falls while errors and exceptions hold, something more interesting is happening. You are beginning to build a system.
We eat our own cooking.
We ran this exercise inside THE UN KNOWN about a year ago, and it changed how we operate. Today, automations run across our operations, from email triage and prioritization to brand consistency, content quality control, and decision preparation.
Our systems work through the night across Zones 1 and 3. They sort, qualify, gather information, and prepare context while nobody is awake. By the time I start my day, I can have two or three Zone 3 decisions waiting for me. The analysis is already there. I do not spend the morning assembling what I need to decide. I spend a few minutes deciding.
At THE UN KNOWN, this is part of how we approach AI transformation: redesign the decisions first, then build the systems around them.
Zone 4 works differently. Our systems research, analyze, and challenge assumptions. They have never had the right to decide. That line does not move.
Call it AI or SI. The question is the same.
Even the language is shifting. On September 29, 2026, a US executive order directed federal agencies to use "Super Intelligence" and "SI" in place of "Artificial Intelligence" and "AI" in official non-statutory communications. For now, the order defines SI using the technologies covered by the existing statutory definition of AI, while directing the administration to propose legislative language for a federal definition. It is a change in terminology, not evidence of a separate new technology. Whether the term spreads into mainstream business remains to be seen.
The direction is still worth watching. The conversation is moving from what these systems can generate toward what increasingly capable systems should be allowed to do. The operating question does not change: where does the machine get authority, and where does the human keep it?
The exercise costs nothing. No tool, no vendor, no budget. Mapping 20 decisions takes 90 minutes. Redesigning how hundreds of them move between your people, your data, and your systems is a different problem. The map is where you start. It is not where you stop.
What leaders ask about the AI Decision Map.
Sort decisions by reversibility and frequency, not importance. Decisions that are reversible and frequent, like triaging requests or prioritizing a queue, can be automated at low risk. Decisions that are irreversible and rare, like an acquisition or a layoff, stay with a human who decides alone while AI prepares the analysis. As reversibility decreases, human authority should increase. Before granting autonomy, also evaluate blast radius, data quality, regulatory exposure, and the path for human escalation.
The AI Decision Map is a framework from THE UN KNOWN for mapping business decisions according to reversibility and frequency. It creates four zones of authority: AI decides; AI analyzes and a human approves; AI prepares context and a human decides; or AI advises while decision authority remains entirely human.
Do not start with an entire process. Start with a recurring decision inside it. Look first for high-frequency decisions where mistakes are detectable, contained, and quickly reversible. Automate one decision, measure it for 30 days, then expand only if human time decreases without an unacceptable increase in errors or exceptions.
Track three operational measures before and after implementation: human time required, error rate, and exception rate. If automation saves five hours but creates four hours of human exception handling, most of the apparent gain disappears. Financial ROI can then incorporate the value of capacity recovered, revenue created, implementation cost, and the cost of errors.
Sometimes. Autonomous decisions are most appropriate when mistakes are reversible, detectable, and contained. Frequency can make automation more economically valuable, but it can also widen the blast radius. High-impact or difficult-to-reverse decisions require stronger human authority and additional controls. The objective is calibrated authority for each decision, not maximum automation.
Artificial intelligence remains the established technical and business term. In September 2026, the US executive branch was directed to use "Super Intelligence" (SI) in place of AI in specified official non-statutory communications. For now, the order ties SI to technologies covered by the existing US statutory definition of AI while a proposed federal definition is developed. This governmental use should not automatically be treated as equivalent to every traditional technical definition of superintelligence.
No vendor pitches. No hype. One real problem a leader is facing, the method to solve it, and one thing you can run this week.
Join the leaders building AI that actually ships. About THE UN KNOWN
