Pricing Research Analysis

Is the intersection of a price sensitivity meter the optimal price?

No. The point where the too-cheap and too-expensive curves cross is where equal proportions reject the price for opposite reasons. It contains no information about volume, revenue, margin or competition, and nothing in the instrument's construction makes it an optimum. The instrument's honest output is a range of acceptable prices and the shape of resistance across that range. Report the four curves rather than only their crossings, because a steep resistance curve and a shallow one imply very different pricing strategies at the same crossing point.

How different is stated willingness to pay from what people actually pay?

Systematically different, in a direction that depends on the category, and by an amount no correction factor reliably fixes. Respondents overstate willingness to pay for products they like the idea of, understate where naming a high price feels foolish, and are pulled toward whatever price the question implies. The practical response is to lean on relative readings (where resistance accelerates, how segments differ, how the shape breaks) rather than absolute levels, and to calibrate against any observed price point available. If a comparable product has ever sold at a known price and volume, the size of the adjustment needed is the most honest measure of the instrument's bias you will get.

What is wrong with sequential price laddering?

Two things, both systematic. Acquiescence: respondents tend to keep agreeing through a series of similar questions, so accepted prices drift upward. Anchoring: the starting price sets a reference, and thresholds found from a high start are higher than those found from a low start for the same population. If the design randomised the starting price or direction, compare the resulting curves and you can size the effect directly. If it used one start, the level of the curve is conditional on that start and only the shape is robust. Look for spikes at the starting price and at round numbers, which indicate the ladder measured the anchor rather than the threshold.

Should I use monadic price testing?

It is the cleanest of the stated-preference approaches, because each respondent sees one price, so there is no within-respondent anchoring and no acquiescence chain. Its cost is sample: every price point needs its own cell. Analyse it as a comparison of independent groups, check the cells are balanced on the variables that matter (an imbalance on category usage will masquerade as price sensitivity), and resist fitting a smooth curve through a small number of noisy points. Plot the points with their intervals first.

How do I build a demand curve from survey data?

By plotting stated intent or acceptance at each price against price, and then writing down every assumption that sentence contains: that intent equals behaviour, that the sample represents the buying population, that competitor prices hold constant, and that the curve exists only where it was measured. Points between tested prices are interpolation and points outside are extrapolation with no evidence behind them. If an observed price and volume are available, calibrate the curve to pass through that point and disclose the size of the shift.

Is the revenue-optimal price the same as the volume-optimal price?

Almost never. Volume is maximised at the lowest price tested, which is why volume alone is not a pricing answer. Revenue peaks higher. Margin, where cost data is available, peaks higher still, because every unit sold cheaply carries the same cost. Presenting one optimum without the others is a common route to a wrong pricing decision. Report the flatness too: where revenue barely moves across a band, the research has told you price within that band is not the lever.

Can I get a price elasticity from a survey?

You can compute one, and it needs three caveats to be usable. It is local to the prices where it was estimated, not a constant of the product. Its interval is wide, because it inherits the uncertainty of two proportions and then divides them. And it is elasticity of stated intent, not of sales, which for most categories will be too elastic, since a survey has no switching costs and no habit. Where the interval spans both elastic and inelastic values, the honest conclusion is that the study cannot distinguish them.

Does it matter that we did not show competitors?

Considerably. A price tested with no competitor present is tested in a market that does not exist. Respondents anchor on their own reference prices, and acceptable ranges come out systematically wider and higher than the same product priced against alternatives. If the decision is competitive and the study had no competitive frame, the study cannot answer it. At minimum, add known competitor prices to every chart as reference lines, which is the cheapest way to make an abstract price curve commercially readable.

Why will you not give me a single recommended price?

Because a single price implies a precision the method does not have, and because the choice within a defensible range usually turns on things the study did not measure: positioning, competitor response, channel relationships, the cost of changing the price back. The useful output names the range, identifies the price within it the evidence best supports, states the assumptions that would have to hold, and says what would change the answer. A decision-maker who knows the answer depends on a competitor holding steady can watch for the thing that would invalidate it.

Can AI analyse pricing research reliably?

It can run the curve arithmetic consistently, which is genuinely useful. The characteristic failures are specific: producing a precise optimal price with decimal places from an instrument that only supports a range, fitting a smooth demand function through five noisy cells, extrapolating beyond the tested prices, converting stated intent into revenue without naming a conversion assumption, and dropping the stated-preference framing because the prose reads better without it. Require the framing statement in the deliverable, a hard boundary at the tested range, and every assumption written where a decision-maker will actually read it.