I’ve noticed a shift in my own practice recently. With one organisation at the team level, I’ve tended to prefer micro-nudges more often than formal experiments. In other words, using small probes accompanied by signals to watch for, rather than formal hypotheses with success criteria. However, with the same organisation on more strategic business challenges, the formal experiment has felt more useful.
That’s worth examining. Generally, I have defaulted to “experiment” almost reflexively, though Backbriefing and Experiment A3s. State a hypothesis. Define what I would expect to see. Run it. Inspect the result. A clean PDSA loop that has served many people well. However, not everything needs that loop.
What a probe actually is
In Cynefin terms, a probe is the first move in the Complex domain. You can’t analyse your way to a useful next action because cause and effect aren’t visible until after the fact. So you do something safe-to-fail, watch what happens, and let the system’s response tell you what to do next. That is Probe – Sense – Respond.
A probe doesn’t have to be naive, though. It can be grounded in experience, informed by what’s worked in similar systems and what’s recognised as good practice elsewhere. But it doesn’t try to install that practice wholesale. Good practice from elsewhere is at most a hypothesis about a different system. The probe asks which element might land here, or what shift in context might move things closer to where a recognised good practice could take hold.
A probe doesn’t carry a hypothesis. It carries an intent and a set of signals, which are the things you’ll watch for that show whether the system is moving in a direction you find beneficial. Not “did X cause Y?” but “what shifts when we do this?” This sits naturally with the Vector Theory of Change. Rather than aiming at a defined end state, you describe where you are, and the direction you’d like to travel, and the probes nudge the system along. Signals tell you whether you’re moving along the right vector. And, more interestingly, what new possibilities have become possible from where you’ve ended up.
When hypotheses get in the way
At the team level, the moment I commit to a hypothesis, I risk narrowing what I’m willing to notice. The hypothesis becomes a flashlight pointed at one corner of the room, and the rest goes dark. Most of what’s interesting at this level is dispositional. Small shifts in how people work together, where attention falls, and what conversations become possible. Those things rarely show up where a hypothesis is pointing. They show up in your peripheral vision, and micro-nudges with signals leave that peripheral vision intact.
This connects to something I touched on in the recent SOAP notes post. Confirmation bias is a constant danger, and a formal hypothesis can quietly enforce the very bias the discipline of experimentation is meant to guard against if it excludes what counts as a relevant signal too early.
Where the experiment earns its keep
But the experiment isn’t wrong. It’s just not always right. At the organisational or business-unit level, the hypothesis may work better. It coordinates and rationalises by enabling a conversation about what we believe, what we’ll do to check assumptions, what we’ll see if we’re right, and what we’ll do next either way. In doing so, it forces clarity on what success looks like and makes the cost of the intervention transparent.
That formality also earns its keep when solving wicked problems and collaboratively building a portfolio of possible things to spend time on. When comparing potential bets and deciding what’s worth pursuing, the discipline of stating hypotheses and outcomes makes those conversations productive, because everyone is considering the same form of information. More broadly, strategic-level interventions usually involve too many stakeholders and too many entangled commitments to be run as undirected probes.
For very small, quick, cheap changes, though, that overhead is rarely worth paying. A probe with signals lets you act with much less ceremony, and you often learn just as much. In addition, the probes can then inform subsequent experiments.
Probes across teams (such as adjustments to work definition and workflow, WIP limits, entry and exit policies, cadences, and meetings) can be a form of distributed sensemaking. Each team explores locally, picking up signals (such as flow metrics, team health surveys, retrospective narratives, and conversations). Aggregate across enough teams and patterns surface that no single team could have seen alone. Those patterns inform the larger hypotheses and experiments at the strategic level (such as governance, team structures, and technology adoption). And it’s rarely a single experiment at that level: a strategic hypothesis often has several safe-to-fail experiments under it, each testing a different facet of the same bet.
Matching the approach to the context
So it’s not a case of probes versus experiments. It’s a question of matching the approach to the context. Probes work well for small, local, dispositional changes, where the cost of a single move is low, signals come back quickly, and peripheral vision matters more than a defined target. Experiments work well for the larger, more committed, more positional changes, where you need to coordinate, justify, and focus, and where the hypothesis is doing useful work for the organisation as well as for you.
Those size markers are useful shorthand, but they’re not really what’s doing the work. The experiment is still meant to be safe-to-fail. That is, not so large or expensive that being wrong creates real risk. What it requires more of than a probe is coordination and alignment: several people, possibly across functions, sharing an understanding of what’s being tested and what would count as a result. A small intervention that spans several teams or requires sponsor agreement might still call for an experiment; a large intervention in genuinely complex territory, contained within one team’s autonomy, might be better treated as a probe.
Through a Cynefin Lens
There’s another Cynefin framing too. Probes operate in the Complex domain, discovering what’s possible through action. Experiments can be more liminal, stabilising what the probes have surfaced and moving the system toward Complicated, where what was emergent can settle as recognised good practice. That direction isn’t always what you want, though: some parts of a system need to stay in Complex, and over-stabilising what should remain emergent is its own failure mode. Equally, something that has stabilised may no longer be appropriate and may need to shift back into being treated as Complex.
The mistake I was probably making before was defaulting to the experiment too readily. Some of those experiments might have been better as probes. I might have learned more, sooner, and with more variety, by treating them that way.