Experimentation & Learning

What this dimension measures

This dimension looks at how the organization handles the test itself: how a trial starts, what it is expected to prove, how it ends, and whether anything carries forward. When a new tool is tried, is the trial small enough to be safe and real enough to be informative? Is there a point at which someone decides, based on what was learned, whether to continue, change course, or stop? Does the next trial look different because of the last one?

The assessment asks respondents to walk through a recent trial from first idea to real feedback: what people believed beforehand, what changed, and how it ended. The shape of that story is the evidence. A trial that was small, ended in a decision, and shaped a later effort indicates a working learning loop. A trial that was large, or that never resolved, or that left no trace, indicates that the organization may not yet have one.

An organization with high Experimentation & Learning tends to prefer modest trials on real tasks, expects to revise its view as results arrive, and closes trials rather than letting them drift. An organization sitting lower on this dimension may still be curious and willing; the difference is that trials either do not start without a full specification, or start and never conclude.

How the organization treats the person who brings back bad news from a trial is a separate matter, covered by Openness & Psychological Safety.

Why this matters for AI

AI is unusually hard to specify in advance. The same tool can be excellent at one version of a task and unreliable at a slightly different version, and the only way to find out is to try it on real work. Its behavior shifts as models change. Its value depends on details of the local process that nobody fully documents. An organization that requires certainty before acting will find that AI rarely provides it, and will either stall or commit to something large on the strength of a demonstration.

Learning by trial is the alternative. Small efforts on real tasks, with a clear point at which results are examined, turn uncertainty into evidence. The organization finds out where the tool helps, where it fails, and what the surrounding process needs to look like, at a cost it can afford to write off. Each round narrows the question for the next one.

This matters for mistakes too. AI systems make errors, some of them confident and plausible. An organization that is used to examining trial results has a habit of checking that catches these errors early. One that does not may discover them only when they reach a customer or a decision.

It also matters for pace. AI capabilities move quickly, and the organizations that keep up are not the ones that predicted correctly but the ones that could revise. A learning loop that runs in weeks rather than quarters lets an organization keep adjusting as the ground shifts. That loop is what this dimension is trying to see.

How this can work in your favor

When your responses describe small trials on real work that end in a decision and shape what comes next, AI adoption can proceed without demanding a certainty the technology cannot offer.

Uncertainty becomes affordable. Nobody has to be right about an AI tool in advance. The organization can try it on a bounded task, look at what happened, and decide. A wrong guess costs a few weeks rather than a year.

Evidence accumulates. Each trial leaves something behind: what worked, what did not, what the process needed. Over several rounds, the organization builds a real understanding of where AI fits its work, rather than relying on claims from vendors or from early enthusiasts.

Errors surface early. The habit of examining results means that an AI tool's mistakes get noticed while they are still cheap, before anyone has built a dependency on output nobody checked.

Judgment improves. Teams that have run several trials get better at predicting which ideas are worth testing and at designing tests that actually answer the question. The learning loop sharpens itself.

Adoption keeps pace with the technology. When a new capability appears, there is a known way to find out whether it matters here, so the organization can respond in weeks rather than waiting for a plan.

How this can make AI adoption harder

When your responses suggest that trials are rare, large, or left unresolved, AI adoption tends to meet friction at the start.

Certainty expected too early. Efforts cannot begin without a full specification, and AI does not supply one. Proposals wait for a confidence that never arrives, or the organization commits to a large project on the strength of a demonstration rather than its own evidence.

Trials that never end. A pilot starts, generates interest, and then drifts. Nobody decides whether it succeeded, so nothing is learned and nothing changes. Months later it is still "being evaluated," and the next proposal has no result to build on.

First versions judged as final. Without a habit of revision, an AI tool's early rough edges are read as a verdict on the whole idea. The organization concludes that AI does not work for it, when what did not work was one configuration on one task.

Mistakes found late. If results are not examined routinely, an AI tool's confident errors can travel a long way before anyone notices. The first serious problem is discovered by its consequences.

Learning that does not travel. Even when a trial goes well, the lesson stays with the people who ran it. The next team starts from scratch.

This is friction, not a verdict. It describes where AI efforts in this environment are likely to stall: at the point where someone is asked to act without knowing the answer.

What this dimension is not

Experimentation & Learning is not recklessness. Running trials without bounds, checks, or a point of decision is not learning; it is activity. The dimension is about trials that are small enough to be safe and structured enough to teach something.

It is not the same as being able to change the production process afterward. An organization can learn a great deal from pilots and still be unable to act on it; that is Adaptability's question.

It is also not about whether people feel safe reporting a disappointing result. The treatment of the messenger belongs to Openness & Psychological Safety. This dimension looks at the treatment of the test: whether it started small, ended in a decision, and shaped what came next.

What we look for

The assessment looks for the shape of real trials rather than statements about a culture of innovation. Signals that tend to be informative include:

• Whether the most recent trial of a new tool was on a real task, with a bounded scope, rather than a demonstration built to impress.
• Whether that trial ended in a decision to continue, change, or stop, and who made it.
• Whether people can say what they believed before the trial and how the result changed their view.
• Whether a later effort was designed differently because of an earlier one.
• Whether there are trials that have been "in evaluation" for many months with no resolution.
• Whether results were examined closely enough that an error in the tool's output would have been caught.

Claims such as "we encourage experimentation" are not evidence. The story of one trial, from first idea to what happened next, is.