Game Theory and Price Wars

What is the framework?#

1. Define the game before pricing it#

Name the players, the segment, the product boundary, the time horizon, what each side can observe, and capacity constraints. Then list the moves that are not price: packaging, contract length, service level, financing, scope, guarantees, narrower focus. Most "price wars" are lost because the price lever was the only one on the table.

2. Compute required volume lift first#

unit contribution = price - variable cost

required new volume = old volume x old unit contribution / new unit contribution

required volume lift = required new volume / old volume - 1

Use the incremental variable cost of the marginal segment, including support, onboarding, payment, and fraud. Fixed cost does not disappear from the analysis — it sets break-even and therefore how long you can fight.

3. Build a response tree and take an expectation#

For each move, enumerate plausible competitor responses — ignore, match, undercut, bundle around you, target your top accounts, change product — and attach a probability range.

expected contribution = sum(response probability x contribution under that response)

Then evaluate the move against the expectation, not against the best branch. The most common failure in practice is building the tree and then arguing from the 30% branch.

4. Diagnose whether you have a structural advantage#

A price move is defensible when it expresses something real: lower cost to serve, a self-serve motion, a different package, cheaper acquisition, perishable excess capacity, or a revenue stream the rival lacks. A cash subsidy is not a moat — see moats and economies of scale for what qualifies.

5. Preserve reversibility#

Prefer a time-boxed offer, a segment-specific plan, a usage allowance, an annual-commitment price, a migration credit, or a repackaged tier. Each tests the hypothesis without resetting the market price, and each can be withdrawn.

6. Keep the decision provably independent#

Use public market information and internal analysis; document the customer-value and cost reasoning behind the number. Do not discuss future prices, discounts, bids, capacity, or terms with competitors — directly or through an intermediary, trade association, or shared software. Where a pricing tool ingests competitors' non-public data, treat that as a legal question before a technical one: the DOJ's position is that an algorithm cannot do what a person could not.

Worked example: does the cut pay?#

A startup sells 10,000 units per month at $100; incremental variable cost is $35.

monthly contribution = 10,000 x ($100 - $35) = $650,000

It considers $75 and forecasts 35% more volume if the rival holds.

Volume and revenue both rise; contribution falls $110,000. The break-even volume is:

$650,000 / $40 = 16,250 units → required lift = 62.5%

The forecast 35% is not close. And if incremental volume carries $3 per unit of extra support and payment cost, unit contribution is $37, break-even volume is 17,568, and the required lift is 75.7%.

Now price the response#

Three branches, defined explicitly so the volume figures are auditable:

expected monthly loss = $650,000 - $464,000 = $186,000

Against $1.2M of cash earmarked for this strategy, the war consumes that reserve in about 6.5 months — before other burn. Carry the $3 incremental service cost through every branch and expected contribution falls to $429,200, the monthly loss rises to $220,800, and the reserve lasts about 5.4 months.

The narrower alternative#

Hold $100 list, and offer $80 for an annual prepaid commitment in a low-support segment. Unit contribution is $45, so break-even is 14,444 units — a 44.4% lift, still demanding, but the offer is reversible, segment-fenced, brings cash forward, and lowers acquisition cost. It also does not tell the market that list price was never real.

What this arithmetic hides#

Every number above is a point estimate standing in for a distribution. The three response probabilities are unobservable and were chosen, not measured; shifting the "holds price" branch from 30% to 50% moves expected contribution by $50,000 a month. The volume elasticities are forecasts. The model's value is not the $464,000 — it is that the two decision-sensitive assumptions are the rival's probability of matching and the incremental cost to serve. Report it as a range and name those two assumptions.

What are the common mistakes?#

• Modelling revenue instead of contribution. More units at a worse unit economic can destroy value while every growth dashboard turns green.
• Assuming no response. Price is the most observable and most cheaply copied variable you have.
• Using average variable cost. The marginal buyer usually costs more to serve than the average one; see fixed vs. variable costs.
• Discounting with no give-get. A concession granted for nothing teaches the customer that list price is an opening bid.
• Confusing a price cut with a strategy. If the cut is not the expression of a cost, packaging, or channel advantage, it is a subsidy with an end date.

When does this framework break?#

The market definition is wrong. A payoff matrix is only as good as the boundary drawn around it. Rivals with different cost structures, objectives, financing, capacity, or complementary revenue will not respond as your matrix assumes — and a firm optimising for share, strategic position, or a parent company's ecosystem is not playing the game you modelled.

Static contribution misses lifetime effects. A discount can generate learning, liquidity, referrals, or genuine switching costs — or attract low-retention bargain seekers who churn. Track discount cohorts through retention and expansion before concluding either way.

Winner-take-all conditions change the calculus. Where strong network effects or winner-take-all dynamics exist, buying share below cost can be rational because the feedback loop is real. The test is whether the loop is demonstrable, not asserted — most markets described this way are not.

The legal analysis is not reducible to the economics. Antitrust outcomes turn on facts, jurisdiction, and market position, not on payoff matrices. Get counsel before relying on competitor data, parity or most-favoured-nation clauses, resale restrictions, below-cost exclusion strategies, or automated pricing informed by non-public competitor information.