Price Discrimination:
Set SaaS Prices by Customer Segment

Charging two B2B SaaS buyers different prices is not automatically price discrimination; the label fits when the same or an economically similar service carries different net prices because willingness to pay differs rather than marginal cost. That can be a defensible strategy only when the segment predicts buying response, eligibility can be enforced, contribution improves, and the rationale survives buyer-legitimacy and legal review.

First rule out a genuine cost difference#

Begin with classification, before discussing segments or fences. A visible price difference is not automatically price discrimination. The economic definition used by the OECD is narrower: similar offers with the same marginal cost are sold at different prices. A difference caused by extra implementation work, higher service cost, currency or tax treatment, payment risk, or a genuinely different product can instead be cost-based price differentiation.

In SaaS, compare realized net prices for comparable rights, not price-page labels. Credits, free months, implementation waivers, support, usage commitments, renewal caps, payment timing, and termination rights all change the economic offer. Two accounts can have the same list price and different net prices; two plans can have different list prices without being economically comparable.

Once comparable rights and cost differences are isolated, the three-degree taxonomy identifies how the buyer is sorted. It does not yet say whether a particular mechanism is profitable, fair, enforceable, or lawful.

Perfect first-degree discrimination—charging every buyer exactly their willingness to pay—is a theoretical endpoint. Individual negotiation or algorithmic personalization can approximate it, but neither reveals willingness to pay perfectly. A SaaS tier can implement second-degree discrimination because buyers reveal something by choosing a version; if the versions also carry materially different entitlements or cost to serve, not all of the price difference is discrimination.

The OECD distinguishes price discrimination from differences caused by marginal cost, describes first degree as individual willingness-to-pay pricing, second degree as buyer choice from a menu of versions, and third degree as different prices for observable customer groups.

Three conditions make discrimination possible. The seller needs a downward-sloping residual demand curve and therefore some ability to set price; it needs a way to observe or elicit differences in valuation; and arbitrage between low- and high-price buyers must be difficult. This does not require a monopoly or prove substantial market power. The OECD explicitly warns that the size of a price difference is not a measure of the extent of market power.

For SaaS, arbitrage is broader than literal resale. It includes a buyer splitting usage to reach a cheaper band, an ineligible account accessing a discount through an affiliate, a high-service account buying a low-service package while expecting exceptions, or procurement using another customer’s noncomparable quote as a reference. A fence must control the economic path, not merely rename the segment.

The formula explains the logic, not the price#

The taxonomy names the mechanism; the formula tests its economic logic under a deliberately narrow model. For separable segments in the standard constant-marginal-cost model, profit is maximized where marginal revenue in every served segment equals marginal cost:

MR_i = MC

The inverse-elasticity pricing rule writes the same logic as:

(P_i - MC) / P_i = 1 / |epsilon_i|

or:

P_i = MC / (1 - 1 / |epsilon_i|)

Here, P_i is the segment price, MC is marginal cost, and |epsilon_i| is the absolute own-price elasticity of demand facing the seller in that segment. A less elastic segment has fewer or weaker substitutes and receives the higher modeled markup. A more elastic segment receives the lower one.

Illustrative worked example—not company data and not currency. Normalize marginal cost to one price unit. Segment A has an absolute elasticity of 2, so its modeled price is 1 / (1 - 1/2) = 2 price units. Segment B has an absolute elasticity of 4, so its modeled price is 1 / (1 - 1/4) = 1.33 price units. The corresponding price-cost margins are 50% and 25%.

The arithmetic does not authorize those prices. It assumes known demand, constant marginal cost, separable segments, profit maximization, and elastic demand at the chosen points. Actual SaaS marginal cost can differ through support, infrastructure, implementation, payment risk, and sales effort. Historical negotiated prices are also selected rather than random: salespeople change discounts when they expect a deal to be difficult, and only won deals appear in invoice data. Estimate price elasticity of demand from a defensible exposure and counterfactual before using the rule.

Research from California Institute of Technology course materials notes that under the stated inverse-elasticity model, the price-cost margin equals the reciprocal of the segment’s absolute demand elasticity, so a less elastic segment receives a higher modeled price than a more elastic segment.

There is no widely accepted benchmark for a “fair” segment gap, a correct number of SaaS segments, or a guaranteed profit lift. The formula is a conditional relationship, not an industry target.

Dynamic pricing, personalized pricing, and price fences are different mechanisms#

The formula assumes separated demand but does not tell an operating team what created the price difference. These mechanism labels do, and each creates a different measurement and risk surface.

Dynamic pricing changes with market conditions such as time, supply, or aggregate demand. If every comparable buyer shopping at the same moment sees the same condition, the price can be dynamic without being personalized. Personalized pricing uses an individual’s characteristics or conduct to set or influence that individual’s price. Price steering changes which offers a buyer sees or their order; it may affect outcomes without changing the price of an identical item. A/B price testing creates experimental price variation to measure response; the test is not, by itself, a durable segmentation policy.

The OECD’s digital-pricing analysis separates these practices because “the algorithm changed the price” does not identify the input. Market state, account attributes, an individual’s behavior, and randomized assignment are different causes.

A segment is an analytical claim that a defined set of buyers has a different demand response. A price fence is the operational condition that determines who can receive a price or what trade-off accompanies it. Kimes and Wirtz’s revenue-management monograph describes fences as the conditions attached to rate categories; well-designed fences let customers self-select while restricting access to the lower-price path.

A segment is a claim about buyer behavior. A fence is the rule that implements the claim. The quote is the audit trail that shows whether either one was real.

A SaaS segment must predict a different buying decision#

With the mechanism named, the next question is whether the proposed segment predicts behavior at all. “Enterprise,” “mid-market,” and “small business” are convenient reporting labels. They become pricing segments only when they predict a different response to a comparable economic offer. Company size can correlate with value, procurement burden, support cost, bargaining power, or substitute availability; those are different mechanisms and should not be collapsed into one willingness-to-pay story.

The segment hypothesis has to exist before anyone searches for a favorable price. A usable statement has this form:

For buyers in [predefined eligible group], the comparable offer solves [distinct job or constraint], the relevant alternatives are [named categories, not invented vendors], and demand is expected to be [more or less] price-sensitive over [defined interval and horizon]. We will distinguish willingness to pay from cost to serve using [evidence].

The hypothesis states the claim; the minimum segment record below makes that claim testable over a defined population and horizon:

A B2B nonlinear-pricing working paper illustrates why both won and lost opportunities matter. Its method uses intended deal size from successful and unsuccessful sales efforts to estimate price sensitivity under stated assumptions. The application is one educational-services firm, not a SaaS benchmark, but the measurement lesson travels: invoice data contains no observation of the quantities and prices rejected buyers might have accepted.

The same paper also models incentive compatibility. If adjacent quantity bands have sharply different marginal prices, buyers can change purchase size to reach the cheaper schedule. Optimizing each band separately can therefore produce a globally weak tariff. In SaaS terms, model downgrade, account splitting, delayed expansion, and negotiated exceptions before celebrating one segment’s modeled margin.

Firmographic labels are not sufficient segment evidence. California Institute of Technology course materials and the arXiv paper indicate that a pricing segment needs a distinct demand response, and a multi-band schedule must account for buyers changing quantity or package to reach a more favorable price.

Use fences buyers can understand and systems can enforce#

A segment record supplies evidence about demand; a fence turns the chosen distinction into buyer access and system behavior. It earns its place when it is relevant to the transaction, applies consistently to every eligible buyer, and can be enforced without a growing exception queue. Common SaaS mechanisms perform different jobs:

Not every row is pure price discrimination. A higher service level can cost more to deliver. A longer commitment can reduce risk. Taxes can change the invoice without changing the seller’s economic price. Decompose the gap into cost, risk, product, and willingness-to-pay components instead of giving every difference one label.

The matrix describes intended behavior. The four scenarios below challenge it from the paths most likely to erase the distinction before launch:

1. A buyer who values the high-price offer tries to qualify for the lower price without changing its economic behavior.
2. A current high-price account learns the lower condition and requests parity at renewal.
3. Sales overrides the rule to close a quarter-end deal.
4. A buyer crosses the eligibility boundary after purchase through growth, contraction, merger, location, or channel change.

The answer is not always a harder fence. A restrictive contract can reduce conversion or trust more than leakage costs. Sometimes the right response is a smoother quantity curve, a genuinely different package, a standardized discount authority, or one transparent price.

Fairness is part of demand, not an ethics appendix#

Leakage asks whether buyers can bypass a fence; fairness asks what happens when they see it working as designed. Economic legality and buyer legitimacy are different questions. A practice can be lawful yet commercially brittle if the buyer cannot understand why comparable accounts receive different treatment. In B2B, that reaction can appear as a longer security-and-procurement cycle, a most-favored-price request, a renewal dispute, lower reference willingness, or a loss of champion trust.

The strongest direct fairness evidence in the researched set comes from consumer settings, so it needs a clear boundary. In two experiments, fence-context fit affected fairness judgments: a pricing variable that fit the purchase context was evaluated differently from one that did not, and suspicion played a role. In three e-commerce experiments, personalized pricing produced negative reactions among both disadvantaged and advantaged participants, mediated by fairness perceptions.

Those results do not quantify enterprise churn or prove that every negotiated quote is disliked. They establish a useful design warning: receiving the discount does not automatically make an opaque mechanism feel legitimate, and familiarity with price variation does not excuse a fence that has no credible connection to the offer.

Consumer experiments have linked price-fairness judgments to the fit between a fence and its purchase context, and have found negative responses to personalized pricing even among some participants who received the advantageous price. Research from Journal of Business reports that the studies do not establish an enterprise-SaaS effect size.

The buyer-legitimacy test is not another segment definition. It evaluates whether the implemented difference remains understandable, symmetric, predictable, and correctable after prices become visible:

• Comparable basis: Normalize entitlements, term, quantity, support, risk, and concessions before calling two prices unequal.
• Relevant reason: Tie the rule to value, cost, commitment, service, or access—not to a hidden estimate of how trapped or uninformed one buyer may be.
• Symmetric eligibility: Give the same result to every account meeting the same documented condition, subject to a governed exception path.
• Explainability: Make the account-facing rationale understandable without exposing another customer’s confidential contract.
• Predictability: State what happens when eligibility changes and how acquisition, renewal, migration, and grandfathering interact.
• Reviewability: Keep a human appeal and correction path when data or classification can be wrong.

Monitor realized net-price dispersion among truly comparable accounts, exception and override rates, segment migration, complaints, sales-cycle delay, win-loss reasons, contraction, renewal, and retention. None has a universal pass threshold. The useful signal is a change against the predeclared counterfactual and guardrail.

Test the scheme against a uniform-price counterfactual#

Even a well-evidenced segment, enforceable fence, and legible rationale leave the main economic question unanswered: what happened relative to one governed price? Price discrimination does not always harm buyers, and uniform pricing does not always help them. The OECD’s competition analysis explains that segment pricing can lower the price for buyers who otherwise would not purchase, expand output, intensify competition, or help recover fixed investment. It can also transfer surplus to the seller, raise prices for some buyers, distort a downstream market, or fund costly efforts to track valuation and prevent arbitrage. The welfare result depends on the market and counterfactual.

Economic theory provides no universal welfare verdict for price discrimination. It can expand access or output and can also transfer surplus or harm particular buyers; market-specific outcomes and the uniform-price counterfactual determine the result.

For an operating decision, compare contribution rather than list-price lift:

incremental contribution = sum across segments of [(realized net price - variable cost) x realized quantity] - operating cost of the scheme - contribution under the uniform-price counterfactual

This decision worksheet consolidates the economic consequences; it is not an industry formula. Calculate it for the same eligible population and horizon, then add the effects the compact expression hides:

• acquisition, renewal, expansion, downgrade, contraction, and cancellation;
• migration between packages, bands, entities, channels, and geographies;
• service and implementation exceptions;
• collection timing, credits, refunds, and payment risk;
• buyers newly served at the lower price and buyers excluded at the higher price;
• measurement uncertainty and the cost of maintaining the pricing system.

A segment scheme passes the economic gate only when its improvement survives those adjustments. If the result depends on ignoring later renewal loss, treating a package change as a pure price response, or comparing selected negotiated deals with self-serve buyers, the model has not isolated the effect.

Put law, personal data, and competition outside sales discretion#

The economic worksheet cannot approve the legal or data boundary. “Price discrimination” is an economic label, not a legal conclusion. The law varies by jurisdiction, buyer type, product classification, data use, market position, and competitive effect. Use the following as review triggers, not legal advice.

In the United States, the FTC says price differences are generally lawful, particularly when justified by different costs or a good-faith response to a competitor. Its Robinson-Patman overview also lists specific elements: the statute applies to commodities rather than services, covers goods of like grade and quality, requires at least two purchasers and possible injury to competition, and has additional commerce requirements and defenses.

That does not mean adding “SaaS” to a contract settles the analysis. A mixed software, equipment, implementation, resale, channel, promotional-allowance, or services arrangement can require fact-specific classification. Ask qualified counsel which rules apply to the actual offer and customer relationship.

The FTC describes price differences as generally lawful under U.S. federal antitrust law, as Federal Trade Commission guidance explains, while listing specific Robinson-Patman elements, including commodities rather than services, like grade and quality, multiple purchasers, and possible injury to competition.

Keep competitor information outside the segment model. The DOJ’s 2025 proposed RealPage settlement arose from alleged conduct in rental housing, not ordinary SaaS packaging. Its clean operating boundary is still relevant: competitors must make independent pricing decisions. Do not use a shared system, meeting, survey, or nonpublic data exchange to align prices with rivals. Public competitor research and independent pricing judgment are different from pooling current, sensitive commercial information; counsel should set the boundary.

In the RealPage matter, the DOJ stated that competing companies must make independent pricing decisions and proposed restrictions on using competitors’ nonpublic, competitively sensitive information and features that aligned pricing. The matter concerned alleged rental-housing conduct.

Personalization creates a separate data surface. FTC staff’s preliminary surveillance-pricing findings describe consumer prices and promotions influenced by location, browsing, shopping, and behavioral data. The findings were preliminary, consumer-focused, and not proof that B2B SaaS uses the same practices. They show why a model built from an individual champion’s behavior is not merely “account segmentation”: it can involve personal data, inference, opacity, and a different set of review obligations.

For EU consumer distance contracts, European Commission guidance says a consumer must be told when the price is personalized on the basis of automated decision-making and notes that data-protection obligations may also apply. The guidance distinguishes that from non-personalized dynamic pricing. A pure enterprise contract may fall outside the consumer rule, while a self-serve offer, an individual purchaser, or personal-data profiling can change the facts. Do not infer coverage from a “B2B” CRM field.

Any use of protected characteristics, close proxies, sensitive personal data, opaque vulnerability scores, or automated individual decisions should trigger legal, privacy, security, and fairness review before testing. The purpose is not to declare every such input illegal; it is to stop a revenue team from making that jurisdiction-specific decision by itself.

The approval record should expose the whole argument#

The preceding artifacts do the analysis; the approval record routes the remaining authority and preserves the stop decision. Before approving a segmented price, put one page in front of pricing, finance, sales operations, product, data, and counsel where needed. It should link the segment record, fence tests, buyer rationale, counterfactual, and scoped legal review, then answer four gates:

After the gates, classify the approved mechanism as cost-based differentiation, second-degree self-selection, third-degree segment pricing, individual negotiation, dynamic pricing, or experiment. More than one label can apply, but each mechanism needs its own owner and evidence. Also record the launch cohort, review date or trigger, rollback path, and the person authorized to approve exceptions.

If all four gates pass, use the smallest segmentation scheme the evidence supports and measure it through the next meaningful contract event. If the economic gate is weak, collect better demand evidence. If the fence is brittle, redesign the offer or schedule. If buyers cannot understand the rule, change the rule rather than its explanation. If the legal or data boundary is unresolved, do not ship the test.

Approve price discrimination only when it expands access or contribution through a measured segment difference, an enforceable fence buyers can navigate, and a rationale the company can defend after prices become visible. When those conditions are absent, one governed price is not unsophisticated. It is the more accurate model of what you know.

Frequently asked questions#