Adoption begins when the model changes a decision

Implementation research distinguishes outcomes that are often combined under a single word. A skills architecture pilot may end with strong numbers, yet three months later a critical vacancy may once again be filled using the same old filters.

The case is a composite. It does not describe a specific company. It makes it possible to observe the common distance between participating in an implementation and using it to decide.

The accountable team might defend adoption with access metrics. Human Resources might point out that the hiring decision never changed. Both would be looking at different outcomes.

Use, adoption, and effect are not interchangeable

Logging into a platform indicates contact. Completing an action indicates use. Applying the model with its intended meaning in a relevant decision provides evidence of integration. Maintaining that practice as support declines speaks to sustainability.

Proctor and colleagues proposed a taxonomy that distinguishes implementation outcomes such as acceptability, adoption, appropriateness, feasibility, fidelity, penetration, and sustainability. The framework emerged from health services research and should not be transferred as an automatic corporate scale. Its contribution to this analysis is conceptual: different outcomes answer different questions. Proctor and colleagues, Outcomes for Implementation Research, 2011 (opens in a new tab).

A model may be accepted without being used. It may be used extensively while losing fidelity. It may be applied correctly during a pilot and disappear when support is withdrawn.

Reconstructing a decision changes the conversation

Instead of asking whether leaders use the model, the team selects three recent decisions. It reviews a project assignment, a development recommendation, and an internal move.

For each one, it records five elements:

• the criterion that would have been used before.
• the new evidence provided by the model.
• the alternative that the evidence made visible.
• the decision made and who accepted accountability.
• what happened afterward, including any adjustment or reversal.

For development, the platform suggested a gap, but the final recommendation was the same general course from the previous year. There is no evidence that the decision changed.

For mobility, two profiles were compared based on the proximity of their skills. The decision stopped because no one had defined how to handle insufficient evidence. The model did enter the process, although it did not yet support a conclusion.

Changing a decision does not prove complete causality

If the project achieves better results after bringing in the person selected, we should not attribute the entire improvement to the model. Leadership, resources, context, and other people’s performance also play a part.

The defensible evidence is narrower: the model introduced information that changed the option under consideration; the decision was made under explicit criteria; and an outcome was later observed. That chain allows learning without confusing contribution with exclusive causation.

The same caution applies when the result is negative. Poor subsequent execution does not prove on its own that the initial decision was wrong. The sequence and conditions must be preserved.

Persistence reveals whether the practice changed

One different decision may be an exception promoted by the pilot team. To speak of sustained adoption, it is useful to observe what happens under pressure: limited time, the absence of an expert, or an urgent vacancy.

If the people accountable return to the previous criterion, feasibility may be lacking. The model may require evidence that is not available in time. If different teams interpret levels incompatibly, the problem may be fidelity. If no one knows who resolves a discrepancy, governance is missing.

Each diagnosis calls for a different intervention. More communication does not resolve an ambiguous rule on its own. More training does not fix a tool that arrives after the decision.

Measure adoption as a portfolio

Coverage, frequency, and satisfaction metrics remain useful. They help detect unequal access, friction, or rejection. Their limit appears when they are presented as complete proof of change.

A stronger portfolio combines signals of contact with samples of reconstructed decisions. There is no need to audit every case. A representative set can be selected by unit, type of decision, and level of risk to determine whether the initiative creates value beyond the pilot.

The review should respect purpose, confidentiality, and proportionality. It seeks to understand whether the model enters the point where it promised to add value, without monitoring individual conversations.

The case continues How to scale beyond the pilot without losing quality. Scaling preserves meaning; adoption means incorporating it into the work. It also connects with What it means for an assessment to be valid for a specific decision. A measurement can be technically consistent yet unsuitable for its assigned use.

Adoption does not begin when the dashboard fills up. It begins when a decision can be explained differently because the model contributed relevant evidence, and when that practice survives the end of the pilot.

To extend this reading, see How to connect skills with outcomes without confusing correlation and causation and How to tell whether a skills-first initiative creates value after the pilot, which develop complementary dimensions of the problem.