The pattern
A growth company makes a decision with the information it has. A quarter later, the evidence has moved — a channel underperformed, a cohort behaved differently than modelled, a larger customer reshaped the roadmap. So the company re-cuts the decision. Then it does it again.
From the inside, this can feel like instability. Leadership starts to wonder whether the business lacks discipline, whether the team is indecisive, whether the plan is worth the document it is written on.
It is none of those things. Deciding before the evidence is complete, and revising as it arrives, is not a defect in a scaling company. It is the core activity. The question is not whether the business will change its mind — it will, repeatedly — but whether it can do so without losing coherence in the numbers, the team, and the plan.
That is the difference between learning and drift.
Why incomplete information is the normal condition
Mature businesses operate on settled models. They know their unit economics, their channels are understood, and their forecasts are calibrated against years of actuals. Decisions can wait for data because the data is reliable and the cost of waiting is low.
Scaling businesses have none of that. They are simultaneously discovering market fit, model fit, and operating fit — often while the market itself is moving. The data is thin because the company has not existed long enough to generate thick data. Priorities evolve because the company is learning what actually drives the business, not because leadership is fickle.
Waiting for certainty is not an option, because certainty never arrives in time to be useful. The business has to act on partial evidence, observe what happens, and adjust. The companies that scale well are not the ones that guess right the first time. They are the ones that can revise quickly and cleanly.
What makes that possible is not better information. It is better operating discipline.
The two failure modes
When a company tries to operate on incomplete information without the right structure underneath it, it fails in one of two directions.
Drift. The business reprioritises constantly, but the changes are never made as explicit decisions. Priorities shift in practice while the plan, the metrics, and the team’s understanding stay anchored to where things were. Numbers stop reconciling because nobody updated the definitions when the model changed. The leadership team is busy, but the busyness no longer maps to a coherent direction. This is iteration without discipline — motion mistaken for progress.
Rigidity. The opposite, and just as damaging. The business treats the plan as fixed because changing it feels like failure. Evidence accumulates that an assumption is wrong, but the operating model has no mechanism for absorbing that evidence, so the company keeps executing against a plan it has quietly stopped believing in. This is discipline without learning — coherence preserved at the cost of being wrong on purpose.
Both failure modes come from the same root: the absence of an operating system designed to let the business change its mind in a controlled way.
What learning without drift actually requires
The goal is not to eliminate ambiguity. It is to build enough structure that the business can move through ambiguity without fragmenting. In practice, that rests on a small number of operating disciplines.
Definitions that hold when the model changes. The fastest way for learning to turn into drift is for the numbers to stop agreeing — the metric disagreement problem that surfaces the moment two teams cut the same number differently. When the business re-cuts a decision, the metrics underneath it have to be defined, owned, and reconciled to a single source of truth — so that a change in strategy does not become a change in what the numbers mean. Shared definitions are what let a company argue about the decision rather than about the data.
Driver-based forecasting instead of target-based forecasting. A forecast built backwards from a target tells you nothing when reality diverges. A forecast built forwards from operational drivers — conversion, sales cycle, churn, collection timing — quantifies the cash impact of every new piece of evidence the moment it arrives. That is what makes a course correction a calculated decision rather than a leap. When the data is partial and the outlook is moving, the driver model — the heart of forecasting discipline — is what keeps the choices credible.
Decision rights that make revision cheap. Companies drift when reprioritisation happens informally, and they go rigid when every change has to escalate. Explicit decision rights — who owns which call, at what threshold — let the business adjust at the level where the evidence actually lands, without either silent drift or a bottleneck at the top.
A cadence that turns evidence into decisions. Learning only compounds if there is a rhythm that forces it to. A working weekly and quarterly cadence is where new evidence is reviewed, where priorities are formally re-cut rather than allowed to slide, and where a revised plan is recorded and communicated. Without it, course corrections happen in hallways and never make it into the plan everyone is executing.
None of these remove uncertainty. Together, they give the business a way to be uncertain without being incoherent.
The COO’s actual role
This is the work a strong operating partner or COO is really doing in a scaling business — and it is frequently misunderstood.
The job is not to impose certainty on a business that does not have it. It is not to slow the company down until the data is complete, and it is not to defend a plan against the evidence. It is to build the operating architecture that lets a business decide with imperfect information, learn from the outcome, and adjust without losing trust, pace, or control.
Done well, this is nearly invisible. The numbers reconcile. Priorities change as explicit decisions rather than as drift. The leadership team spends its time on the decision in front of it instead of relitigating what the scoreboard says. The board sees a business that is adapting deliberately, not thrashing.
Good COO work does not eliminate ambiguity. It gives the business a way to move through it credibly.
The diagnostic
Four questions reveal whether a business is learning or drifting:
- When you change a priority, is that change recorded as an explicit decision — or does the old plan quietly stay in place while the work moves on?
- When the strategy shifts, do your core metrics still reconcile, or does each change leave the numbers meaning something slightly different?
- When new evidence contradicts a forecast assumption, how quickly is the cash impact visible — and to whom?
- Can your leadership team revise a decision without it escalating to you, and without the rest of the business losing the thread?
If the honest answers are uncomfortable, the issue is rarely the quality of the decisions. It is the absence of an operating system strong enough to let the business keep learning without coming apart.