There are four types of pricing experiments available to a SaaS company: price level tests, plan structure tests, feature gate tests, and trial length tests. Each has a different minimum sample size, contamination risk, and signal quality. Most teams run the wrong type for the question they are trying to answer — or run the right type with insufficient power to detect a real effect.
Before changing any price, product usage data tells you what you need to know. Customers who use many features and return to the product frequently are insensitive to moderate price increases. Customers who use one or two features and visit infrequently will churn at nearly any increase. Segmenting by these two dimensions — feature breadth and session depth — before running an experiment lets you predict outcomes rather than discover them the hard way.
- The four types of pricing experiments and what each one is actually testing
- Why most pricing experiments are underpowered and return false signals
- The cohort contamination problem and why concurrent A/B testing is harder in B2B than it looks
- How to use product usage data to predict price sensitivity before changing anything
- The ethical and legal constraints on differential pricing in SaaS
The Four Types of SaaS Pricing Experiments
Every pricing experiment changes one variable. The problem is that teams frequently change multiple variables simultaneously — adjusting price levels, restructuring tiers, and re-gating features in a single "pricing refresh" — and then wonder why the results are uninterpretable. Disciplined experimentation requires isolating the variable under test.
Type 1: Price level tests
A price level test changes the dollar amount at a fixed packaging structure. Everything else stays the same — the plan names, the feature set at each tier, the trial terms — and you measure whether the new price level affects conversion rate, win rate, or time-to-close.
Price level tests sound simple. They are the hardest to run cleanly in B2B SaaS. The reason is sample size. Detecting a meaningful difference in conversion rate at typical B2B conversion volumes requires several hundred qualified opportunities per variant — a number most growth-stage SaaS companies cannot generate in a reasonable time window without running the test for six months or longer. Running it for six weeks on forty prospects per variant and calling the result significant is the most common source of false pricing signals in the industry.
The insight: Price level tests require more data than almost any growth-stage SaaS company has available in a short window. If your conversion volume is low, use cohort analysis on a historical price change instead of a concurrent A/B test.
Type 2: Plan structure tests
A plan structure test changes the architecture of tiers — adding a plan, removing a plan, splitting one plan into two, or reordering the hierarchy — without necessarily changing the price at any individual tier. The outcome variable is plan mix: which tier do new customers choose, and how does that affect average contract value and first-year expansion?
Plan structure tests are more tractable than price level tests because the signal is visible faster. A change in the distribution of new customers across tiers shows up in the first few weeks of conversions, long before a meaningful retention signal is available. The risk is that you optimize for initial plan selection without knowing whether the new structure improves or degrades expansion behavior downstream.
The insight: Plan structure tests optimize for the top of the monetization funnel. Always pair the initial conversion signal with a 90-day expansion check before drawing conclusions about whether the new structure is better.
Type 3: Feature gate tests
A feature gate test moves a specific capability between tiers — adding a feature to a lower tier, removing it from a higher tier, or creating a new tier gate around a feature that was previously ungated. The outcome variable is both tier conversion and usage depth: does gating the feature drive upgrades, or does it simply reduce usage without creating upgrade pressure?
Feature gate tests are the most tractable experiment in SaaS pricing because they operate on a defined feature boundary rather than on the price number itself. But they require a clear hypothesis about which customers are most likely to upgrade in response to the gate. Gating a feature that your highest-engagement users do not care about produces no upgrade pressure and frustrates users who valued it without ever feeling the need to pay more for it.
Gating a feature creates upgrade pressure only if the blocked customers value the feature more than the cost of upgrading. If they do not, the gate just reduces product value without generating revenue.
The insight: Before designing a feature gate test, check the usage data. A feature used heavily by customers who do not upgrade is a candidate for gating. A feature used heavily by customers who already upgraded tells you nothing about upgrade motivation.
Type 4: Trial length tests
A trial length test changes the duration of a free trial — shortening it, extending it, or converting a time-based trial to an action-based trial (activated when a user completes a specific workflow rather than after a fixed number of days). The outcome variable is trial-to-paid conversion rate and the quality of customers who convert.
Trial length tests are among the most tractable experiments available because they operate on a large population (everyone who starts a trial) and produce a clear binary outcome (converted or not) in a short time window. They also directly influence the activation audit: whether customers reach the moment of value within the trial window is a function of both the trial length and the activation sequence, not the trial length alone. Extending a trial without improving the activation flow typically produces a modest conversion lift that disappears once you account for the lower quality of late converters.
The insight: Trial length tests measure the combined effect of time-in-trial and activation quality. Isolating the trial length effect requires holding the activation sequence constant — something most teams do not control for.
| Experiment Type | What It Changes | Min Sample Size | Test Duration | Risk of Contamination | Signal Quality |
|---|---|---|---|---|---|
| Price level test | Dollar amount per tier, billing period, or both — packaging unchanged | 200–500 qualified opportunities per variant | 2+ full sales cycles; typically 60–120 days | High — B2B buyers compare notes; concurrent splits expose differential pricing | Low at typical B2B volumes; cleaner as historical cohort analysis than concurrent A/B |
| Plan structure test | Tier count, tier names, plan hierarchy — not price levels | 100–300 new customers per variant | 30–60 days for initial mix signal; 90 days for expansion read | Medium — structural changes are visible on pricing pages but less personally offensive than price differential | Medium — initial plan selection signal arrives fast; expansion signal requires patience |
| Feature gate test | Which features are available at each tier — not price or tier count | 50–150 users who regularly use the gated feature | 30–45 days to observe upgrade conversion; usage changes visible faster | Low — gates are product decisions, not differential pricing; visible to users but not framed as a price test | High relative to effort — produces a clear upgrade signal if the feature was genuinely valued |
| Trial length test | Duration or trigger condition of the free trial period | 200–400 trial starts per variant | trial length + 30 days to measure conversion and quality | Low — trial terms apply to new users only; current customers are not affected | Medium — conversion rate is observable; downstream quality of converters requires longer tracking |
Why Most Pricing Experiments Are Underpowered
An underpowered experiment is one designed to detect an effect that would be commercially meaningful, but with a sample size too small to tell the difference between a real effect and random variation. Most SaaS pricing experiments are underpowered by a factor of three to five — not because teams are careless, but because the required sample sizes are larger than intuition suggests.
The sample size problem in B2B SaaS
Consumer pricing experiments can accumulate thousands of conversions per week. B2B SaaS rarely can. A company with 200 qualified pipeline opportunities per month has perhaps 60 conversions. To detect a 10% difference in conversion rate at 80% statistical power requires approximately 400 conversions per variant — more than six months of data, split across two conditions, holding all other variables constant for the entire duration.
Most pricing tests are abandoned long before that threshold is reached. The team sees a directional trend at week four and makes a decision. That directional trend is sampling noise more often than it is a signal. Publishing a "pricing experiment" result based on 40 conversions per variant is equivalent to concluding that a coin is biased after flipping it eight times.
The typical B2B pricing experiment is underpowered by a factor of 5–10×. Teams run tests on 30–80 conversions and interpret directional results as signals. At those sample sizes, a 15% conversion rate difference would need to be observed consistently to pass a basic significance threshold — and most tests are not run long enough to accumulate the necessary observations. The correct response to a small conversion volume is not a shorter test. It is a different experimental design.
The false precision of short tests
Short pricing tests do not just lack power — they actively mislead. Early-converting customers are not a representative sample of the customers who will convert after seeing the new price for sixty days. Early converters are typically the most motivated, the most value-aware, and the most price-insensitive. A three-week test on motivated early converters will systematically overestimate the conversion rate of a price increase, because the customers who would have churned in response to the new price have not yet had time to decide.
This is especially pronounced in B2B SaaS with long sales cycles. If your average sales cycle is 45 days and you run a pricing test for 30 days, you are measuring behavior in the first two-thirds of the consideration window. The deals most likely to convert at the old price but reject the new price are the ones that close in weeks five through eight — precisely the ones your short test did not capture.
The insight: The minimum test duration for a B2B pricing experiment is two full sales cycles after reaching minimum sample size — not two sales cycles total. Most teams count the ramp period as part of the test duration, which systematically underestimates the required duration.
When cohort analysis beats A/B testing
If your conversion volume is below the threshold needed for a concurrent A/B test, the right tool is cohort analysis on a historical price change. Every time a SaaS company has changed its pricing — even informally — there is a natural experiment embedded in the data. Customers who evaluated and converted before the change form a control cohort. Customers who evaluated and converted after form a treatment cohort.
Cohort analysis on natural price changes has its own confounds — seasonality, sales team changes, product updates — but it operates on real conversion events at the actual scale of the business, without the contamination risks of concurrent testing. For most growth-stage SaaS companies, it is the more honest tool.
Know which cohorts can absorb a price increase before you change anything
ProductQuant's Foundation diagnostic maps feature breadth and session depth across your customer base — the two signals that predict price sensitivity. You get a pricing sensitivity read as part of the activation audit, so you know which segments to grandfather and which to test before committing to a change.
Start with The FoundationThe Cohort Contamination Problem in B2B Pricing Tests
Cohort contamination in pricing experiments happens when control and treatment groups are not truly isolated — when information about the price difference leaks between groups and distorts the measured effect. In consumer markets, contamination is rare because strangers rarely discuss what they paid for the same software. In B2B SaaS, contamination is nearly inevitable.
Why B2B buyers are uniquely contamination-prone
B2B buyers compare notes. Procurement teams discuss vendor costs with peers in industry communities, Slack groups, conference hallways, and direct LinkedIn messages. Finance teams reviewing contracts compare line items with other portfolio companies. A buyer who discovers they are being shown a 30% higher price than a competitor who mentioned the tool in a conversation last week does not quietly accept the difference — they escalate it, complain about it, or walk away.
The contamination risk scales with industry concentration. In a market where the target buyers all participate in three or four professional communities, running concurrent A/B tests on price levels means you are effectively running those tests in the same social network. Contamination is a question of when, not whether.
"In B2B SaaS, differential pricing exposed mid-funnel does not just lose a deal — it poisons the relationship with the prospect and everyone they talk to. The risk is not statistical noise. It is reputational damage at a scale that a consumer A/B test would never produce."
Kyle Poyar, Operating Partner at OpenView — OpenView Partners Blog
Reducing contamination risk: geographic and temporal segmentation
Two structural approaches reduce but do not eliminate contamination risk in B2B pricing experiments.
Geographic segmentation assigns price variants to non-overlapping regions. All prospects in one market see price A; all prospects in another market see price B. This approach is clean if the two markets have genuinely different buyer communities with minimal cross-pollination. It breaks down when buyers in the two regions frequently interact — which, in verticals where buyers attend the same global conferences, is often the case.
Temporal segmentation eliminates concurrent testing entirely. You run price A for a defined window, then switch to price B, and compare cohorts across the two windows. This avoids any contamination from simultaneous exposure but introduces time confounds — seasonal patterns, market changes, product updates, and sales team evolution all become noise in the comparison.
Neither approach is fully satisfying. The honest conclusion is that in concentrated B2B markets, controlled pricing experiments have meaningful limits. The alternative is not to avoid testing pricing — it is to reduce reliance on concurrent A/B tests in favor of observational methods and pre-change analysis.
The insight: In B2B SaaS with a concentrated buyer community, concurrent price-level A/B tests carry contamination risks that most teams underestimate. Sequential testing with cohort comparison is safer, even if the confounds are different.
How Product Usage Data Predicts Price Sensitivity Before You Experiment
The most underused input in SaaS pricing decisions is the usage data sitting in the product database. Before running any pricing experiment, the data already tells you which customers are price-sensitive and which are not — if you know where to look.
Feature breadth as a price sensitivity predictor
Feature breadth measures how many distinct capabilities a customer uses regularly. A customer who logs in to use one core workflow is behaviorally exposed: if your price increases, their cost-to-continue grows while their perceived value stays constant. There is no expanding surface area of the product holding them in place.
A customer who uses six or eight distinct feature areas has a much higher switching cost. Replicating that breadth of functionality with a different tool requires either a single expensive replacement or assembling multiple tools. Feature breadth is the single strongest predictor of price tolerance in SaaS products with modular feature sets. Customers with high feature breadth have demonstrated that they extract value across the product, making them far less sensitive to moderate price increases than their single-feature-using peers.
Session depth as a retention signal
Session depth measures both the frequency and duration of product engagement. Customers who return to the product daily or multiple times per week are embedding it in their workflow — the product is not a peripheral tool but an operational dependency. Customers with low session frequency are using the product episodically, which means there are long windows where the cost of cancellation is invisible to them.
Combining feature breadth and session depth creates a two-dimensional view of your customer base that is directly actionable in pricing decisions. High breadth, high depth: price-insensitive — these customers can absorb a meaningful price increase without significant churn risk. Low breadth, low depth: price-sensitive and already at churn risk — increasing their price accelerates an outcome that may already be probable.
The pricing sensitivity read in the activation audit
At ProductQuant, the Foundation engagement includes a pricing sensitivity read as part of the activation audit. The audit maps which customers have reached activation depth — using multiple features, returning frequently, completing the core value workflows — and which are still in shallow adoption. That segmentation is not only an activation problem. It is a direct input to pricing decisions.
Shallow adopters who are paying current rates are not good candidates for a price increase. Deep adopters who are paying entry-tier rates are the highest-value candidates for expansion conversations or plan structure changes. Knowing this before a pricing experiment begins lets you design the test around the right population and avoid contaminating the pricing signal with churn driven by activation failure rather than price objection.
The insight: Usage data segments your customer base by price sensitivity before you run a single experiment. Skipping this segmentation and running a price level test on your full customer base means your results reflect the average of a highly heterogeneous population — and averages in heterogeneous populations rarely tell you what to do next.
Run pricing experiments that are actually designed to detect real effects
Growth LAB includes one experiment per month, designed by the ProductQuant team with proper power analysis, contamination controls, and pre-experiment segmentation using your product usage data. You get a result you can act on — not a directional trend from an underpowered test.
The Ethical and Legal Constraints on SaaS Pricing Experiments
Pricing experiments in SaaS sit at the intersection of two distinct concerns: the legal constraints on differential pricing, and the ethical obligations around transparency with buyers. These are not identical, and conflating them produces the wrong framework for decision-making.
What the law actually governs
Price discrimination law in the United States primarily governs business-to-business sales of physical goods under the Robinson-Patman Act. Software-as-a-service is generally not covered by Robinson-Patman because software is not a tangible commodity. Most jurisdictions do not have direct legal prohibitions on offering different SaaS prices to different buyers at the same time, provided the differentiation is not based on a protected class.
Where legal constraints do apply: consumer protection laws in many markets prohibit deceptive pricing practices — showing one price prominently and charging another, or using dynamic pricing that is not disclosed. In the European Union, price personalization based on inferred personal characteristics carries obligations under the General Data Protection Regulation (GDPR) and the Digital Services Act. Any pricing experiment that uses behavioral data to personalize prices shown to individual users in the EU requires careful legal review.
The insight: For most B2B SaaS companies operating in North America, differential pricing is not a legal problem — it is a trust and ethics problem. Do not confuse legal permissibility with ethical acceptability or business prudence.
The ethics of concurrent differential pricing
The ethical concern with concurrent differential pricing in B2B is straightforward: you are showing two buyers who are in the same market and potentially know each other a different price for an identical product. If discovered, this is experienced as deception, not as a pricing experiment. The buyer who received the higher price does not feel like they were randomly assigned to a test condition — they feel like they were taken advantage of.
This is distinct from legitimate forms of price variation that are widely accepted in B2B SaaS: annual versus monthly billing discounts, volume discounts for larger contracts, geographic pricing tiers that reflect different market rates, and promotional pricing for a defined campaign window. These are structural price differences that buyers understand and expect. Concurrent undisclosed testing of different price levels on equivalent buyers is not a structural difference — it is a differential that has no principled justification from the buyer's perspective.
Grandfathering as an ethical obligation
One ethical constraint that operates regardless of legal requirements is the grandfathering of existing customers when prices increase. Existing customers made a purchase decision under a specific price expectation. Changing that price retroactively — even with notice — violates the implicit contract of their original decision.
Grandfathering existing customers at their current rate for a defined period (typically 12 months minimum) is both an ethical obligation and a business calculation. The churn risk from retroactive price increases applied to existing customers consistently outweighs the revenue gain from the increase — particularly when the increase affects customers who are not yet deeply embedded in the product and have low switching costs. A price increase experiment that is designed to test new pricing on new customers only is both cleaner as an experiment and more defensible as a business practice.
The insight: The most defensible pricing experiment is one that applies a new price to new customers only, grandfathers existing customers explicitly, and has disclosed pricing pages that reflect the actual price being charged. Anything that requires concealment from buyers is a signal that the experiment design has an ethical problem.
Frequently Asked Questions
How long does a SaaS pricing experiment need to run?
Duration depends on conversion volume, not calendar time. A price level test needs at least two full sales cycles of data after reaching minimum sample size — for most B2B SaaS products, that is 60 to 90 days minimum. Running a test for three weeks and making a decision based on 40 conversions is the most common cause of false pricing signals in the industry. If your conversion volume is too low to reach statistical power in a reasonable timeframe, skip A/B testing and use cohort analysis on natural price changes instead.
What is cohort contamination in pricing experiments?
Cohort contamination happens when customers in the control and treatment groups communicate about price. In B2B SaaS, this is nearly unavoidable when both groups contain companies in the same industry or geography. If a buyer in your treatment group discovers they are being shown a higher price than a peer in the control group, you lose not just that deal — you lose trust with both. Geographic or temporal segmentation reduces but does not eliminate contamination risk. The cleanest approach is sequential testing — full rollout in defined windows — rather than concurrent A/B splits.
Can you run differential pricing on the same product in SaaS?
Differential pricing — showing different prices to different buyers for the same product at the same time — is legally permitted in most jurisdictions but creates significant ethical and business risk in B2B SaaS. Unlike consumer markets where price variation is accepted, B2B buyers frequently compare notes. Discovery of differential pricing in a professional context can permanently damage trust. The safer path is pricing differentiation through legitimate tiers, add-ons, or geography-specific pricing pages — not concurrent differential pricing in the same market.
How does product usage data predict price sensitivity?
Price sensitivity correlates with two usage signals: feature breadth and session depth. Customers who use a wide range of features and return to the product frequently have a higher cost to switch and a lower sensitivity to price increases. Customers who use only one or two core features and visit infrequently are far more likely to churn in response to a price change. Segmenting your customer base by these two dimensions before a pricing change tells you which cohorts can absorb a price increase, which need grandfathering, and which are already at churn risk regardless of price.