Transformation Still Has an Open-Loop Problem

A transformation programme can be on time, within budget and widely adopted while its most important result remains unclear: did it make the business work better?

The programme may have delivered exactly what it was asked to deliver. A system went live. People completed training. A new structure was introduced. The project team closed the outstanding issues.

But the investment was probably justified by something more consequential. More capacity for customers. Lower operating cost. Faster service. Stronger control. A team that could achieve more with the resources it had.

Those outcomes depend on what happened to the work after the change. That is where the learning loop is often weakest.

The evidence arrives after the commitment

A business case asks an organisation to commit to a view of the future. It describes the problem, proposes a solution and estimates what that solution will achieve. Large changes need this discipline: resources are scarce, dependencies matter and somebody has to make a decision.

The difficulty is that the organisation learns some of the most important things only after it begins.

An AI assistant may expose how incomplete the source information is. A redesigned workflow may reveal an informal check that was protecting customer quality. A new system may reduce administration in one place while creating reconciliation work elsewhere.

Those discoveries should change the organisation’s understanding of the problem. Yet a programme organised around delivering an agreed scope can struggle to absorb them. Evidence that questions the original design becomes an implementation issue to resolve. The plan remains the reference, even when the operation is revealing a better question.

The result is an expensive separation between deciding and learning. Months can go into preparing the case. Years can go into delivering it. The opportunity to test the central assumptions arrives in the middle, when changing direction is hardest.

What changed in the operation?

Consider an imagined service function introducing AI to prepare responses to customer requests. The investment case assumes that less preparation will create more customer-facing capacity.

The launch measures look encouraging: the assistant is available, employees use it and responses are drafted faster. None of that establishes whether the intended capacity has appeared.

People may be spending the saved time checking outputs. They may have to retrieve missing context from other systems. Specialists may be handling more exceptions. Or the change may be working well, with people spending more time resolving customer problems.

The leader needs to distinguish those possibilities.

The relevant chain runs from the investment through the actual division of work to the outcome. What did the assistant take on? What remained with people? What new work appeared? Where did time go afterwards? Did the service improve?

This is also why the question extends beyond AI. A CRM migration, restructure, shared-service change or revised control can alter the same relationships. Each changes some part of how the business produces its result.

Close the loop around the outcome

Closed-loop transformation begins with the outcome the organisation is trying to achieve.

For a customer function, that might mean increasing the capacity available for substantive customer conversations while maintaining service quality and manageable workloads. That intent gives the analysis direction. It makes some patterns important and others less relevant.

Next comes a credible understanding of the work: where effort goes, what creates value, where delays or rework accumulate and what people believe is causing them.

A proposed change then becomes specific. What are we changing? Why should it help? Which work will it reach? What would we expect to see if our explanation is right?

After implementation, the organisation checks what actually happened. Availability is different from use; use is different from a changed workflow. A benefits claim needs to connect to the operation it describes.

The final step is the one that closes the circuit: the result changes the next decision.

If the intervention works under some conditions but not others, the next move should reflect that. If burden has shifted downstream, the scope needs to expand. If the original explanation was wrong, the organisation should stop investing in it.

The point is to shorten the distance between an operating decision and useful feedback about its consequences.

A useful lesson from control theory

Control theory offers a helpful way to think about that continuity. A controller uses observations and a model to decide what to do, then responds to what follows. Model-predictive control repeatedly updates its plan as new information becomes available, rather than treating an initial sequence of actions as permanently correct. [1]

Organisations are more contested and less predictable than engineered systems. People interpret goals, adapt their behaviour and have legitimate differences about what a good outcome means. There is no need to pretend otherwise to borrow the discipline of comparing expectations with consequences.

Conant and Ashby’s work on regulation and models also points towards a useful question: is our representation good enough for the decision we are trying to make? Their theorem is a formal result under stated assumptions, not proof that a business can be captured in one central model. [2]

For a leader, the practical implication is straightforward. A decision to change how a function works should be supported by an understanding of that function that can be tested and corrected.

More than a series of pilots

Many organisations already experiment, run improvement cycles and revise plans. Closed-loop transformation should build on those practices.

The ambition is to make the connection between them continuous. A pilot should leave behind more than a positive slide. Its result should improve the next operating decision. A benefits review should connect to the original problem and the mechanism of change. A new programme should be able to recover what related work has already taught the business.

That requires continuity of ownership as well as information. Someone must remain accountable for the operating outcome after the project team has delivered its part. The point at which a programme closes cannot also be the point at which the company stops learning from it.

The evidence need not be elaborate for every change. The effort should be proportionate to the decision. What matters is retaining a usable connection between the intent, the intervention, what happened and what the organisation will do next.

AI increases the importance of the whole

When a company introduces agents, it changes more than the speed of a task. It changes who gathers context, who executes, who reviews, who handles exceptions and who is accountable when something goes wrong.

Those relationships can change again when the model, instructions, data or permissions change. A formal process can remain untouched while the practical division of labour moves underneath it.

This creates enormous scope for improvement. It also makes isolated measures less informative. An agent completing more tasks is valuable when the resulting arrangement helps the business achieve its goals. If every completed task creates additional checking and coordination, that also belongs in the picture.

A company able to understand these shifts and respond to them would have a powerful operating capability. It could choose where to delegate further, where to redesign the workflow and where human judgement remains essential.

Make improvement part of operating

Closed-loop transformation does not require constant reorganisation. Sometimes the evidence supports keeping a routine, retaining a control or waiting for a meaningful result. Stability is part of a well-run system.

Nor does it remove the need for long-term architecture or major investment. It makes the route responsive to what the organisation discovers along the way.

The ambition behind Teho is to make this capability part of the company itself: deep understanding of work across humans and AI, connected to decisions and the consequences of change.

Then transformation becomes more than the delivery of a designed future. It becomes a continuing ability to move the business towards its goals—and to recognise when a different move is needed.

References

[1] Mayne, D. Q., Rawlings, J. B., Rao, C. V. & Scokaert, P. O. M. (2000). Constrained model predictive control: Stability and optimality. Automatica, 36(6), 789–814.

[2] Conant, R. C. & Ashby, W. R. (1970). Every good regulator of a system must be a model of that system. International Journal of Systems Science, 1(2), 89–97.