Pricing Changes Are Experiments, Not Forecasts

A SaaS company we worked with last year raised list prices on its mid-tier plan by twenty-two percent. The CFO, who had championed the change, presented it to the board as a $14M lift to ARR over the following twelve months. The number was sourced to a model that combined the company's stated price elasticity (an internal estimate, not a measured one), the renewal cohort distribution, and an assumed pass-through rate on new bookings. The board approved the change. Two quarters in, ARR was tracking roughly flat against the prior trajectory. The discussion in the room was not about pricing. The discussion was about whether the sales team was hitting quota and whether the macro environment was softer than expected.

The change might have worked. The change might have failed. Nobody in the room could tell, because the question had never been framed in a way that admitted of an answer. The forecast — $14M to ARR — was, by the time results came in, indistinguishable from the noise around it. There was no counterfactual. There was no observation window. There was no pre-committed rule for what to do if the number came in below expectations. The change had been launched as a promise, and the company was now living inside the consequences of the promise without having ever defined what would constitute keeping it.

This is the typical pattern. Pricing changes are framed as forecasts when they are in fact experiments, and the framing destroys the company's ability to learn from the change or to reverse it cleanly.

Why the framing matters

The forecast framing — "this change will lift ARR by X percent" — has two failure modes, and both are structural rather than cosmetic.

The first failure mode is interpretive. Once the change has been made, the data the team collects is unable to distinguish "the change worked" from "we had a good quarter for other reasons." Revenue moves for many reasons in any given period. Without a counterfactual — without an articulation, in advance, of what we would have expected to see in the absence of the change — every post-change number is ambiguous. The team that wants to claim success can do so by pointing at the gross number. The team that wants to claim failure can do so by pointing at the underlying mix. Both arguments are unfalsifiable, and the company learns nothing from the experience. The next pricing change is debated on the same priors as the last one, because the last one did not actually generate evidence.

The second failure mode is political. A forecast is a commitment. The CFO who said the change would lift ARR by $14M is now on the hook for a specific number. If, six months in, the team realizes the change is not delivering, reversing the change reads as admitting the forecast was wrong. The reversal is therefore deferred — past the point at which it should have been made, sometimes indefinitely. The company holds a pricing structure that the data, if anyone were looking honestly, would say should be revised. Nobody is looking honestly, because the political cost of looking is too high.

A forecast asks "what will happen?" An experiment asks "what would tell us we were wrong, and what would we do about it?" The second question is the one that compounds.

The experiment framing fixes both failure modes — not by changing what the team does (the change still gets made), but by changing how the team thinks about it before the change is made.

What an experiment-framed pricing change looks like

An experiment, properly framed, has four components. They are not exotic. They are the components every introductory statistics course teaches, applied to a business decision. The reason they are rarely applied is not technical. It is that the components require the team to commit, in writing, to interpretations they would otherwise have the flexibility to reshape after the fact.

The first component is a hypothesis, stated as a belief with a confidence range. Not "we forecast $14M of ARR lift," but "we believe this change will produce between $6M and $20M of incremental ARR over the next twelve months, with eighty-percent confidence, with our point estimate at $12M." The range is the honest version of the forecast — it reflects what the team actually believes, rather than the number the team is willing to put on a slide. The act of widening the range usually reveals that the team's actual conviction is much weaker than the deck implied. That weakness is information. It tells the board what kind of bet this actually is.

The second component is a counterfactual. What would we have expected to see in the absence of the change? In some pricing changes, the counterfactual can be measured directly — by holding out a cohort, a region, or a segment from the change. In others, a direct holdout is operationally or politically impossible, and the counterfactual has to be modeled from prior cohort behavior. Either is acceptable. The unacceptable case is the one where the counterfactual is never specified, because that is the case in which any outcome can be reconciled with any narrative.

The third component is an observation window and a set of metrics. Not "we'll keep an eye on it" — a specific date, a specific list of metrics (ARR is rarely sufficient; the team should also track churn, downgrade rate, sales cycle length, win rate against alternatives, and discount depth), and a specific threshold on each. The window has to be long enough for the signal to emerge through the noise. For most B2B pricing changes, six to nine months is the floor; for changes that affect renewals more than new bookings, twelve months is more honest. The window also has to be short enough that the company can actually act on the result. Eighteen months is, in our experience, the practical ceiling — beyond that, too much else has changed for the experiment to remain interpretable.

The fourth component, and the one most teams skip, is a pre-committed action on each outcome. If the result lands in the upper half of the predicted range, what do we do? Almost always: ratify the change, consider extending it. If the result lands in the lower half but inside the range, what do we do? Hold the change, but do not extend it; revisit in another cycle. If the result lands below the range, what do we do? Revert, or restructure. The pre-commitment is the entire point. The pre-commitment is what converts the change from a one-way bet into a reversible experiment, and what allows the reversal — if reversal is what the data demands — to happen without political cost.

The reversibility problem

The reversibility problem is the one most companies do not think about until it is too late. A pricing change that is forecast and approved as a promise becomes, on the day it is made, much harder to reverse than it was to put in place. The CFO who advocated for the change owns the outcome. Reverting reads as the CFO being wrong. The CFO, accordingly, does not propose to revert.

A pricing change framed as an experiment, with pre-committed actions on each outcome, has the reversal already built into the original commitment. Reverting is not the CFO being wrong. Reverting is the pre-committed response to a particular range of outcomes, agreed in advance by the same room that approved the change in the first place. The political cost is borne up front, in the form of the harder conversation at the moment of commitment, rather than amortized into the slow refusal to revisit a change that is no longer working.

A retail business we worked with had run a pricing change a year prior to our engagement that, on any honest read of the post-change data, had produced a material lift in margin but a much larger drop in conversion than projected. The net was negative. Nobody had reverted it. The reason, as the head of pricing told us privately, was that reverting would have required a board conversation in which the original $30M lift forecast came up. Easier to leave the change in place and let the numbers absorb back to baseline organically. The cost of that decision — a year of suppressed top-line — was several multiples of what an honest revert in month four would have cost. The original forecast framing had made the revert politically unaffordable.

This is closely related to the broader point we have made about the question to ask before committing significant capital: the question is not "what do we expect to happen?" but "what would tell us we were wrong, and what would we do about it?" Pricing changes are simply the version of this question that companies face most often, and run worst. A pricing change happens, in most subscription businesses, every twelve to twenty-four months. The compounding effect of running each one as an experiment rather than a forecast — better data, faster learning, cleaner reversibility — is, over a decade, the difference between a pricing function that gets sharper over time and one that produces the same conversations on the same priors year after year.

The cultural cost is up front

The cultural cost of moving to an experiment framing is paid once, up front, in the first meeting where the CFO is asked to widen their range and pre-commit to actions on the downside. The conversation is uncomfortable. The CFO experiences it as having their forecast questioned. The CEO may experience it as the finance team going soft.

The cultural benefit, however, accrues from the first cycle onward. The first time a change is run as an experiment and reverted cleanly because the pre-committed criteria were not met, the company learns that reversion is not failure. The second time, the team proposes the experiment framing themselves. By the third or fourth cycle, the pricing function is operating on a different basis from the rest of the company's planning processes, and the difference shows up in the numbers.

If your next pricing change is being framed as a forecast — a single number, with no counterfactual, no observation window, no pre-committed action on the downside — that is the change most worth reframing before it is launched. We are happy to discuss how that reframing would look in your specific business. The change still happens. It just happens in a way the company can actually learn from.