• Jun 7

Loops and Cycles in R&D

Thirty years in R&D leaves you with a small number of observations you keep returning to. This is one of them.

An R&D organization runs on loops. Not one loop, many. A test rig comparing a measurement against a specification. A stage-gate process being refined year on year. A prototype built to settle a technical question. A portfolio review reacting to a shifted market. A program closing and, in principle, leaving the organization wiser than it started.

These all look alike. They are iterative, they cycle, they have a before and an after. So they get treated as one thing, usually filed under a single word: learning, or improvement, or feedback. That conflation is a mistake, and an expensive one. The loops differ in purpose, in mechanics, in the activities they demand, and in what they actually produce. Treating them as interchangeable means an organization manages some of them well, manages others not at all, and rarely notices the difference.

The five loops at a glance

Feedback / deviation loop

What it changes: restores the system to a fixed target.

How it closes: an error signal, automatically.

Cadence: continuous.

Continuous improvement

What it changes: raises the standard incrementally.

How it closes: a cadence, semi-automatically.

Cadence: regular.

Experimentation

What it changes: produces knowledge, a verdict on a question.

How it closes: a result, when it is read out.

Cadence: bounded by the experiment.

Learning cycle

What it changes: builds retained, transferable capability.

How it closes: only a deliberate act.

Cadence: slow, retrospective.

Decision loop (OODA)

What it changes: produces an action and an updated model.

How it closes: the situation forcing a response.

Cadence: fast.

This summary is deliberately compact. The prose below does the work it only points at.

The loops, elaborated

Three of these loops carry their own closing mechanism.

The feedback or deviation loop is the most familiar. Something is measured, compared against a target, and the gap drives a correction. A test result against a spec. A project tracked against its plan. The reference is fixed and known, and the loop's only job is to return the actual to it. It closes by itself. The error signal is what triggers the correction, and nothing about it needs to be remembered to make it run.

Continuous improvement raises the reference rather than restoring it. The stage-gate process, the design-review checklist, the way estimates are made: each is a standard, and CI is the discipline of making that standard incrementally better. It does not wait for a failure. It runs on cadence, and the cadence is what closes it. As long as the rhythm holds, the loop turns.

Experimentation is different again. Its purpose is not to correct or to improve but to reduce uncertainty. A prototype that settles whether an approach will work. A pilot of a new way of running reviews. At the start the answer is genuinely unknown, which is the point. The loop closes when the result is read and the verdict recorded. It will not close on its own the way a feedback loop does, but an experiment has an end built in, and reaching that end forces the closure.

The remaining two do not behave so conveniently.

The learning cycle converts experience into capability the organization keeps. A completed program contains a great deal the organization did not know before it started. The learning cycle is the deliberate conversion of that experience into a changed way of operating, retained, and available to people who were not there. The difficulty is structural. The learning cycle has no built-in signal. Nothing measures a gap, no cadence forces it, no experiment ends and demands a readout. A program finishes and the organization moves to the next one. The learning cycle closes only if someone decides to close it. Left alone, it does not run at all.

The decision loop, what Boyd called the OODA loop, produces an action under a moving, ambiguous situation. A market shifts, a competitor moves, a technical assumption collapses, and a decision has to be made before the situation is fully understood. It is the fastest loop and the one most about tempo. It closes because the situation will not wait. The pressure of events is its closing mechanism.

The underlying pattern

Set the five side by side and the pattern is not hard to see.

Every loop is a comparison followed by a change. Something intended is held against something actual, and the difference drives an act. What separates the loops is what plays the part of the intended reference, what the loop is allowed to change, and one factor that matters more than the others: whether the loop closes on its own or has to be closed by hand.

That last distinction is the heart of it. A feedback loop, continuous improvement, and experimentation all carry their own closing force: an error signal, a cadence, a result. They are, in different degrees, self-closing. The learning cycle is not. It is the one loop with no internal mechanism that forces it to complete. The decision loop sits closer to self-closing, because events supply the pressure, though the quality of its orienting step still has to be worked at.

So the pattern is uncomfortable. The loop that builds the most durable value, the learning cycle, the one that turns a thirty-year organization into something more than a thirty-year-old organization, is the only one that will not run unless it is deliberately made to run.

There is a second half to the pattern, and it follows from the first. Because most loops close themselves, an organization is full of visible loop activity: corrections logged, processes improved, decisions taken, experiments concluded. That visible activity gets read as learning. It is not. It is the other four loops doing their work. The learning cycle can be entirely absent and the organization will still look busy with loops. The mistake always runs in the same direction: self-closing activity is taken for the loop that does not close itself.

One more element belongs here. The loops run at different altitudes and cadences. The decision loop is tactical and fast. Feedback and improvement are operational and steady. The learning cycle is slow and strategic. They are not competitors, and the goal is not to pick the right one. A healthy R&D organization needs all of them turning. Feedback and CI also operate inside fixed assumptions, correcting and improving without questioning the frame, while the learning cycle and the orienting work of the decision loop are where the assumptions themselves get examined. An organization that runs only the first kind gets steadily better at what it already does and never asks whether it is doing the right thing.

What the Head of R&D must understand and manage

This is where the observation stops being interesting and becomes a job.

The loops described here are mostly unmanaged. They run, or fail to run, by default. No one is assigned to them as a set. And because four of the five close themselves, the absence of management is invisible most of the time. The system appears to be working.

The Head of R&D is the only person with the vantage point to see the whole set. Four things follow.

Distinguish the loops, and refuse to let one stand in for another. A retrospective is not a learning cycle unless it closes into changed practice. A status gate is not a learning checkpoint. Activity in the self-closing loops is not evidence the learning cycle is turning. The discipline is to look at the organization and name, honestly, which loops are completing and which are not.

Direct management attention to the loops that do not close themselves. The self-closing loops will largely take care of themselves; they are durable by construction. The learning cycle will not. It needs an owner, a defined trigger, and protection from the constant pressure of the next program. Management attention is scarce, and it should be spent on the loop that disappears without it, not on the loops that run regardless.

Match the loop to the problem. A measured deviation, a genuine technical uncertainty, a recurring inefficiency, and a fast-moving external change each call for a different loop. Treating an uncertainty as a deviation to be corrected, or running experimentation on the cadence of continuous improvement, wastes effort and produces the wrong outcome. Misclassification is quiet and common.

Connect the loops. An experiment's result should feed the learning cycle. The learning cycle should revise the standards that continuous improvement then maintains. The decision loop should draw on what the learning cycle has produced. When the loops run in isolation, the organization can be busy in all of them and still move in place.

Is this a core competence?

Yes, and it is worth being precise about what the competence is.

It is not the ability to run a loop. Engineers run loops instinctively. They design experiments, track deviations, and improve processes without being told to. That skill is abundant in any capable R&D organization.

The competence is at the level above: seeing the organization's loops as a system, knowing which are turning and which have stalled, deciding which need to be engineered and owned, and keeping them connected. Call it cycle management, or loop literacy. It is a competence of the Head of R&D specifically, because the role is the only one with sight of all the loops at once. An individual contributor sees their own. The Head of R&D is responsible for the set.

It is also, in my experience, an underdeveloped competence. R&D organizations are full of people expert at running individual loops and short on anyone managing the loop system. That is the practical reason the learning cycle so often has no owner: managing the system of loops is rarely written down as anyone's job, so the one loop that needs an owner does not get one.

It belongs alongside portfolio management as a system-level R&D leadership capability. Both are about managing something no single project or person is positioned to manage, and both have to be built deliberately rather than assumed.

One honest qualification. My experience is concentrated in hardware-intensive R&D with long development cycles. In fast-iteration software environments the cadences compress, and the learning cycle is somewhat more self-closing because cycles are short and feedback is quick. The competence still matters there, but the management challenge differs in degree. The longer the cycle, the more the learning cycle depends on deliberate ownership, and the more this is a competence the Head of R&D cannot delegate.

Closing

Separating the loops is not taxonomy for its own sake. It is that the separation makes one specific failure visible: an organization can run four loops well, look busy and competent, and never turn the fifth. Naming the loops is what lets the Head of R&D see whether the learning cycle is actually running, or whether its absence has been hidden by everything else that is.