ML Systems Lab Open interactive version →
Advanced 26 min read Pricingwillingness-to-paycompetitive responsesegmentation

Willingness-to-Pay & Competitive Modeling

Estimating the price a segment will bear, and pricing in a market where competitors respond to your moves

Elasticity tells you how a *market* reacts to price. Two harder questions decide real pricing: how much will *this segment* pay (their willingness-to-pay), and what happens when a competitor re-prices in response to your move? Ignore the first and you leave money on the table; ignore the second and your carefully optimized price triggers a war that erases the gain.


Willingness-to-pay is the reservation price — and you estimate it three ways. WTP is the most a buyer will pay before walking away; a segment's demand curve is the distribution of WTPs. You estimate it by: stated preference — surveys like Van Westendorp ("at what price is this too expensive / a bargain?"), cheap but biased by what people *say* vs *do*; choice modeling / conjoint — show realistic bundles and infer the price coefficient from actual choices, more robust; and revealed preference — the gold standard, reading WTP from real behavior in a price experiment (they *paid*, so it's real). Segment-level WTP curves are what enable differential pricing — student vs enterprise tiers, geographic pricing — capturing more of each segment's surplus than one flat price can.


The second-order trap: your optimum assumes competitors stand still. A naive price optimizer maximizes profit against *today's* competitor prices — a one-shot best response. But competitors have their own optimizers. Cut price to win share and a rival matches you; now both of you sell at the lower price with the *same* share split — you've moved to a worse equilibrium for both. The correct object isn't a one-shot optimum, it's a reaction function: my best price *given how you'll respond*, solved to a competitive (Nash) equilibrium where neither side wants to deviate.


Which is why price wars are a strategic failure, not a modeling win. A local optimizer that ignores reactions will happily walk both firms down to marginal cost — every step looks locally profitable, the destination is ruinous. Real competitive pricing weighs the reputational and equilibrium cost of a move: matching a rival's cut may be rational defense, *initiating* one rarely is. The senior instinct is to model the competitor as a *player*, not a fixed constant — ask "and then what do they do?" before shipping the price. The math that maximizes profit against a frozen competitor is precisely the math that starts the war.

Key points

Takeaway

Willingness-to-pay is a segment's reservation price — estimated by stated preference (surveys), choice/conjoint modeling, or revealed preference from experiments — and segment-level WTP curves are what let differential pricing capture more surplus than a flat price. But a profit optimizer that treats competitor prices as fixed computes a one-shot best response that ignores retaliation; the correct object is a reaction function solved to a competitive equilibrium, because the same math that maximizes profit against a frozen competitor is what starts a price war down to marginal cost.

Recap

Check your understanding

Q1. Which method of estimating willingness-to-pay is most credible, and why?

Q2. Select the two correct statements about a price optimizer that ignores competitor response.

Q3. When is matching a competitor's price cut a defensible move, versus initiating one?

Try it interactively

ML Systems Lab is a free interview-prep platform for ML engineers — work through the full interactive module, quizzes, and drills.

Open ML Systems Lab →