Skip to main content

Three Questions to Ask Before AI Transformation

Technology will bring benefit if and only if it diminishes an existing limitation — and you change the rules that were built to live with that limitation.

Dr. Eli Goldratt's book Necessary But Not Sufficient makes one claim that explains most AI project failures before they happen:

Technology by itself is necessary but not sufficient. For a technology to deliver value, it must diminish an existing limitation. And diminishing the limitation is still not enough. You also have to change the rules your organization built to work around that limitation before the technology existed.

The MRP lesson​

In the late 1970s and early 1980s, MRP software made it possible to run net requirement calculations overnight. Before MRP, this took 20 people almost a full month of manual work. So the rule everywhere was: run the calculation once a month.

MRP eliminated that constraint. Companies could recalculate daily.

Most companies kept running the calculation once a month.

The limitation was gone. The rule stayed. Only 3-4% of MRP adopters captured any meaningful bottom-line benefit. The rest paid for expensive software, expensive data cleanup, and 20 people checking data instead of running calculations — with the same output as before.

The technology worked. The implementation failed because the old rules were never touched.

The three questions​

Goldratt's framework produces three gates you must pass before any technology implementation is ready to succeed. Applied to AI:

1. What specific limitation does AI diminish in this process?

Not a vague answer. Not "AI makes us faster." A specific step in a specific process that is constrained by something AI can now address: a routing decision that previously required human judgment, a classification task that took hours of manual review, a drafting step that bottlenecked the team before anything else could proceed. If you cannot name the limitation precisely, you do not yet know what you are building.

The agent opportunity analysis is the tool for finding and ranking these across your functions.

2. What rules exist today specifically because of that limitation?

This is the question almost nobody asks — and it is the one that determines whether the project produces results or a good demo.

When a step required human judgment, the organization built habits around it: scheduling approval meetings, batching work so a person could review in bulk, adding headcount at the constraint, creating escalation paths for the decision. Those are not incidental habits. They are rules put in place to function despite the constraint. If you deploy AI and leave those rules in place, you get the MRP outcome. The force that keeps those rules alive past their reason is inertia, and it is the default state of every organization, not the exception.

3. What should the new rules be?

With the limitation diminished, the old rules are no longer load-bearing. New rules need to replace them, and there is more than one option — some of which are not obviously better. The answer is not written in the technology. It requires deliberately designing the new workflow once the old constraint is gone.

Why this matters for AI specifically​

The more powerful the technology, the larger the limitation it removes, and the bigger the cost of leaving the old rules intact.

AI's specific breakthrough is making reasoning steps automatable: judgment calls that previously halted a process and waited for a human can now be handled inline. That removes a category of limitation that has been in every organization since organizations existed. The old rules are everywhere. Approval workflows. Escalation paths. Daily standups to surface decisions. Review queues. Org structures built around who makes what calls. Many of those rules exist because a step required a person. When that requirement goes away, the rules need to change with it.

This is also why proven process first matters so much: if the process has not been documented, you cannot accurately answer Question 1 and you cannot identify the rules in Question 2. The SOP is how you read the rules that have accumulated around a limitation over time. Without it, you are guessing at both the constraint and the fix.

The gate test​

Before approving any AI transformation project, treat these three questions as a gate. The project sponsor needs to answer all three in specific, concrete terms:

  1. Which step in which process does this diminish, and what is the current constraint?
  2. What rules exist today because of that constraint? Name them.
  3. What replaces them? Name those too.

If any answer is vague, the project is not ready. Not because the AI is wrong. Because the implementation is not yet aimed at anything specific enough to change.

Goldratt's observation from three decades of watching technology implementations: the bigger the limitation a technology removes, the worse the outcome when you keep the old rules. AI is the most powerful process technology in a generation. The cost of getting the rules wrong is proportionally large.

Further reading​

Source: Dr. Alan Barnard, "Why is technology like AI often Necessary but not Sufficient to create value? Dr Eli Goldratt."