I once dated a woman who seemed to love me.
Over time, I realized she loved a version of me she had already written. He had my name and some of my history. His contradictions had been edited out.
The relationship worked whenever I played the part. Reality kept interrupting.
Many executive teams are doing the same thing with AI.
They love the version from the presentation: fast, scalable, intelligent and ready to transform the company. Then the real system arrives with probabilistic output, integration work, operating cost, weak data, awkward exceptions and people whose jobs must change.
The romance cools. The pilot remains.
Attraction creates activity
AI is easy to admire from a distance. A strong demonstration creates energy. A competitor's announcement creates urgency. A board request creates a budget.
The organization starts several pilots because starting keeps the promise alive. Each team can point to movement. Nobody has to decide which workflow will change, who will lose discretion or which old system will be removed.
That gap now appears in current survey evidence. In McKinsey's 2026 global survey, nearly nine in ten respondents said their organization used AI regularly in at least one business function, and 44 percent reported enterprise scaling. Only 37 percent attributed any EBIT impact to AI. Six percent met the survey's definition of an AI high performer.
These are self-reported results from an advisory firm's survey. They still expose a useful contrast. Use is widespread. Material enterprise value remains concentrated.
Love for the idea can fund adoption. It cannot carry implementation.
Commitment starts when the demo ends
The real relationship begins with a specific workflow.
Take customer complaints. The imagined AI reads every message, understands emotion, resolves the issue and improves loyalty. The actual work crosses a customer database, product rules, billing records, legal obligations and escalation teams. A confident answer can still be wrong. Someone must own the exception and the customer consequence.
This is where avoidant strategy appears.
The sponsor keeps the scope vague because a clear outcome can fail. The team adds features before it has changed the workflow. Leaders wait for competitors to prove the model, then copy the visible tool without the invisible operating choices. When costs rise, everyone blames the technology.
The organization was committed to the promise. It never committed to the change.
Run the commitment test
Choose one AI initiative that leadership calls strategic. Ask five questions:
- What result are we promising? Name the customer or business outcome and the baseline. “Use AI” is an activity.
- Which workflow will change? Show the current steps, the future steps and the work that disappears. Integration belongs inside the design.
- Who gives up or gains a decision right? A new system changes who may recommend, approve, override and stop. Name those changes before production.
- Which costs are we accepting? Include data, integration, model use, review, change, controls and the cost of handling failures.
- Who can end the relationship? Give one person authority to stop the initiative when dated evidence misses the agreed threshold.
An initiative with clear answers may deserve commitment. One without them is still an attractive idea.
The test also exposes compatibility. Your company may want speed while its approval structure rewards caution. It may want personalized service while its customer data remains fragmented. It may want autonomous agents while every exception still needs three committees.
Those tensions are operating facts. Funding another demonstration will not resolve them.
Pay the price of reality
Commitment becomes visible in the budget and the calendar.
If the workflow depends on clean customer history, someone must own the data repair. If employees need to review exceptions, the operating plan must include their time. If a model answer can trigger financial or legal harm, the control cannot be an instruction to “use judgment.” It needs a defined review point, an audit trail and a person with authority to intervene.
These choices can make the business case look worse in the short term. They also make it honest.
An initiative that works only when integration, change and control are treated as free is a presentation artifact. Once the full system is visible, leaders can compare the expected value with the real cost and choose with open eyes.
Decide what you are willing to change
Secure AI strategy is specific. It chooses a material outcome, changes the surrounding system and accepts the consequences of that choice.
This may mean redesigning a role, retiring a process or concentrating investment on one workflow while popular ideas wait. It may also mean stopping an initiative that looked exciting and lacks a credible path to value.
The decision belongs to the executive team. Vendors can show capability. Technology leaders can describe architecture. Advisers can help structure the choice. Only the people who run the company can decide what the company will become and what it is prepared to change.
Look at your AI portfolio as it exists today.
Which initiative has a real relationship with the operating model? Which ones survive because everyone still loves the person in the photograph?
Commit to one. Release the rest.
Your move.