Getting Better at Change

Before funding another programme, it is worth asking what the last related change taught the business.

The answer may be easy to locate in principle and difficult to recover in practice. There is a business case, a delivery record, a benefits report and a retrospective. The people involved remember different parts.

But the next decision needs something quite specific. Which change affected the work? What conditions mattered? What else changed at the same time? Where did new burden appear? What would we do differently now?

A self-improving company should become better at answering those questions. Each consequential intervention should leave it better equipped for the next one.

That is the role of learning in closed-loop transformation.

Preserve the explanation, not just the result

An outcome alone is a poor instruction for future action.

Suppose an imagined team improves after introducing an agent. Another team wants to repeat the result. Knowing that “the agent worked” is not enough.

The first team may have had better source information, more experienced reviewers or a workflow that made delegation straightforward. The agent may have been useful only for routine cases. The visible benefit may have depended on additional work elsewhere.

Those conditions are part of the result. Lose them and the next team inherits a claim rather than an explanation.

The record also needs the original expectation. What did the organisation think would happen, and why? If that prediction is rewritten after the outcome, the company loses the opportunity to see which assumptions were useful and which were wrong.

Retaining the relationship between intent, intervention, context and consequence makes the experience usable.

Three parts of a useful account

Before the change, capture the outcome being pursued, the part of the operation in scope and the current understanding of its work. Include the evidence, important gaps and the explanation the intervention is intended to test.

During the change, establish what was actually implemented. The approved plan may differ from practice. Record the relevant version, affected work, actual exposure, guardrails and any concurrent changes.

Afterwards, connect the observed consequences to the expectation. What moved? What stayed the same? Did effort or risk appear elsewhere? How strong is the evidence that the intervention contributed? What should happen next?

A transition record connects those three parts so the next team can recover the reasoning, not just the result.

The business value is a better decision: whether to scale, revise, stop or investigate; whether a similar intervention is appropriate elsewhere; whether a condition needs to be fixed first.

The record should be no more elaborate than the decision requires. If maintaining it becomes another large administrative burden, its design needs to improve.

Give the right thing credit

Organisations rarely change one variable at a time.

A workflow is redesigned while a new manager joins. Data quality improves while an AI assistant is introduced. Demand changes during training. Performance moves.

Determining which actions contributed is a substantive learning problem. Christina Fang’s research examines organisational learning as credit assignment: learning which intermediate steps help produce an outcome. [1]

In practice, an attractive story can fill the gap. The vendor credits the tool. The programme credits the redesign. The team credits additional effort. Several may be partly right.

A useful account should keep those explanations available until the evidence distinguishes them. It can record that performance changed after an intervention, that the expected mechanism appears to have operated, and that another change remains a plausible contributor.

This is not an argument for withholding every decision until certainty arrives. It is a way to prevent one uncertain result becoming an unquestioned rule for future investment.

A company learns badly when the confidence of the retelling grows while the underlying evidence stays the same.

The next decision is the test

Organisational-learning research distinguishes processes including knowledge creation, retention and transfer. Experience does not become useful simply because it happened; how it is interpreted and used matters. [2]

That suggests a practical test for any learning system: show the later decision it improved.

Perhaps a team avoids repeating an intervention that failed under similar conditions. Perhaps it recognises that a successful pilot depended on unusually complete data and addresses that before scaling. Perhaps it preserves a human check whose value became clear in an earlier trial.

These are benefits of experience reaching the next choice.

A searchable retrospective can help, but retrieval alone is insufficient. The current team needs to understand which part of the earlier experience is relevant, which assumptions still hold and what must be tested again.

The company becomes better at change when the quality of that judgement improves.

Experience needs updating too

Past success can become a source of error.

A practice may have worked with a different customer mix, software stack or set of controls. An agent’s capabilities may have changed. The team may have lost the expertise that made a review process effective. A previously important bottleneck may no longer exist.

The earlier result is therefore a starting point for a new judgement, not an instruction to repeat the same action.

A living understanding of the operation should connect past experience to current conditions. Which features are similar? Which differences could alter the mechanism? What would we need to observe before committing more widely?

This is especially important for AI-enabled work. “The model improved” does not establish that the workflow improved. A better model may change the appropriate review pattern, but the company still needs to understand how the work responds.

Learning should make that investigation more informed, not make it unnecessary.

Why transfer is difficult—and valuable

A change that works in one function may fail in another without either team doing anything irrational.

The functions may face different exceptions, information quality or customer expectations. They may have different authority to act. A workflow suited to experienced staff may impose too much interpretive work on newcomers.

The useful lesson is conditional. It identifies the circumstances under which the intervention helped, and the circumstances in which the result remains uncertain.

That makes transfer more precise. Instead of copying a tool, the next team can reproduce the relevant conditions or choose a different intervention.

Across companies, the same principle applies with additional care. Shared patterns may inform thinking, but private operational evidence is not automatically a common data asset. Rights, confidentiality and the limits of comparison remain part of responsible learning.

The strongest generalisation may be a better question or a more informative experiment, rather than a universal answer.

Learning across human and AI work

As the division of work evolves, a company has more to learn about its own operation.

Where does delegation genuinely remove effort? Where does it turn execution into checking? Which information improves an agent’s usefulness? Which handoffs create uncertainty? Where does human judgement protect the result?

The answers may be more durable than a preference for any particular model or platform. They describe how the company’s work functions.

But accumulated records do not automatically compound in value. They can preserve mistakes, lose context or become obsolete. The benefit depends on people and systems using them to make better choices.

For Teho, learning belongs inside the ambition of a self-improving company: the business can understand its work, change it and use the consequences to become more capable of doing so again.

Memory is part of the infrastructure supporting that capability. The reason to care is what the company can do with it.

The next operating decision should start with more than the confidence of a new proposal. It should benefit from what the business has already learned about how it works.

References

[1] Fang, C. (2012). Organizational Learning as Credit Assignment: A Model and Two Experiments. Organization Science, 23(6), 1717–1732 (first published online 2011). https://pubsonline.informs.org/doi/10.1287/orsc.1110.0710

[2] Argote, L., Lee, S. & Park, J. (2021). Organizational Learning Processes and Outcomes: Major Findings and Future Research Directions. Management Science, 67(9), 5399–5429 (first published online 2020). https://pubsonline.informs.org/doi/10.1287/mnsc.2020.3693