How to run a pricing experiment without losing customers

Changing a price is not symmetrical in its effects. A reduction is received as good news by everyone who hears it and requires no explanation. An increase is received as a loss by people who had already decided what your product costs, and it reopens a decision they had closed. That means the cost of a pricing experiment falls almost entirely on the customers you already have, while the information you want is mostly about customers you do not have yet.

The design principle follows directly: run the experiment where the information is and the cost is not. New products have no established price, so any price is simply the price. New customer segments have no history with you. New geographies, new channels and new tiers are all places where a number can be tested without contradicting something a customer was previously told. Existing customers on an existing product are the one population where a test is expensive, and they are the population most experiments reach for first because they are the easiest to reach.

A new product is the cleanest instrument available. Launch it at the price you want to learn about, and you are measuring willingness to pay with no incumbent expectation to contradict. A premium tier alongside your standard offering works similarly, and it has a second useful property: it tests the ceiling without touching the floor, because customers who do not want it simply keep buying what they bought before. If nobody takes the premium tier, you have learned something about your ceiling at no cost to your existing revenue.

A new geography or a new channel lets you run a genuinely different price for genuinely different buyers, which is defensible on its own terms since the costs of serving them usually differ. And if you have enough web traffic to split it, showing two prices to two comparable groups of visitors measures the trade-off between price and conversion directly, which is the only method here that produces something close to a clean comparison. Two things to hold to: pick which measure decides the outcome before you look, and remember that the higher price wins if it produces more total margin, not if it produces more sales.

If you do raise a price on an existing product, the standard mechanism for protecting relationships is to hold existing customers at their current price — either permanently or for a stated period. This is not merely a kindness. It converts an increase from something done to your customers into something that applies to new ones, and it removes the incentive for a long-standing customer to re-evaluate you at exactly the moment a competitor's price looks appealing.

The trade-off is worth being clear-eyed about. A permanent grandfather clause means you are running two prices indefinitely, which complicates your systems, your quotes and eventually your conversations, since customers do talk to each other. A time-limited period — the old price for a stated number of months, then the new one — captures most of the goodwill while ending the complexity, provided the end date is stated at the start rather than announced later. What does the most damage is a grandfather arrangement that is quietly withdrawn, because that is experienced as two increases and a broken promise.

The instinct when raising a price is to explain at length and apologise, and both make it worse. Length invites scrutiny of the reasoning, and apology invites negotiation, since apologising signals you are not sure the new price is right. What works is short, specific and settled: what the new price is, when it takes effect, what happens to the customer specifically, and where they can ask a question. If a grandfather period applies, that is the most important sentence and belongs near the top.

Giving notice is the part that matters most, because the objection is usually less about the amount than about being surprised. Notice lets a customer plan, place a last order at the old price, or decide calmly — and a customer who decides calmly mostly stays. Reasons are fine in one line if they are true and concrete, such as input costs having risen. Avoid reasons that invite argument, particularly anything implying the old price was a mistake, and avoid framing the rise as beneficial to the customer. It is not, and saying so costs credibility that the honest version keeps.

Decide before you start what you are measuring and for how long. Total margin over a defined period is usually the right measure, since a higher price with fewer sales can be the better outcome and a unit-volume measure would call it a failure. Set the period long enough to cover a full cycle of your normal variation, and long enough for slower-deciding buyers to appear, because a price rise often reduces immediate purchases while leaving eventual purchases roughly intact.

The honest difficulty is that a small business rarely has enough volume to separate a price effect from everything else that moved. A season, a competitor's promotion, a festival, weather, one large customer's unrelated decision — any of these can be larger than the effect you are testing. Two habits help: run the test long enough that a single event does not dominate, and write down beforehand what else is happening so you are not reinterpreting the period afterwards from memory. Where volume is genuinely low, accept that the result is a weak signal and treat it as one input rather than a finding.

It cannot tell you what would have happened at the other price, which is the whole difficulty. Unless you ran a genuine split with comparable groups at the same time, you are comparing two different periods and attributing the difference to the price. Sometimes that is reasonable; often the periods differ in ways you cannot enumerate. A test also measures the response of the customers who found you during the test, not of the market — and a price rise changes who arrives, which means the customers you are measuring are partly a consequence of the change.

It is also silent on the slow effects. A price that holds volume this quarter may change how buyers describe you, which competitors you are compared against, and who recommends you, and none of that shows up inside a test window. That is not a reason to avoid testing; it is a reason to treat a result as evidence about direction rather than a precise measurement, and to keep the change reversible where you can. Where a test says nothing conclusive, say so. A pricing decision made on an honest "we could not tell" is better than one made on a number the data did not support.