.png)
Frameworks, core principles and top case studies for SaaS pricing, learnt and refined over 28+ years of SaaS-monetization experience.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
Using Amplitude for SaaS Pricing Tests a Strategic Guide to Optimization
SaaS pricing teams rarely suffer from a shortage of data. They suffer from a shortage of causal evidence. A dashboard can show that Professional customers retain better than Starter customers, or that conversion fell after a price increase. Neither observation proves what would have happened had the same customers seen a different offer.
Amplitude has moved closer to solving that problem. Its product now combines behavioural analytics with feature and web experimentation, while its Growth and Enterprise packages can add unlimited active experiments, holdouts, mutual-exclusion groups, stratified sampling and account-level bucketing. As of 13 August 2026, every Amplitude plan also includes the broader platform with unlimited seats, while its own commercial model has shifted towards one event-based meter.
But access to experimentation software does not make a pricing test sound. A SaaS company can run a statistically clean A/B test and still make the wrong commercial decision because it randomised individual users inside the same account, optimised checkout conversion instead of retained revenue, or tried to test a pricing metric that its billing system could not actually support.
Monetizely's position is that Amplitude should not be used to answer, "Which price converts best?" It should be used to answer, "Which offer creates the most retained economic value per eligible account for the segment we intended to serve?" Pricing logic must come first; Amplitude's job is to provide controlled exposure, behavioural evidence and a reliable link from the offer a buyer saw to what that customer did afterwards.
Pricing strategy must be settled before Amplitude starts measuring it
Pricing experimentation becomes much clearer when the commercial decision is separated from the measurement technology. Monetizely's 5-Step Pricing Framework, developed in Monetizing Agentic AI, treats pricing as an ordered system rather than a sequence of isolated A/B tests. Goals and Segmentation establishes what the company wants pricing to achieve and which customers it is serving. Packaging decides what each segment receives. Pricing Metric chooses the unit against which price scales, such as seats, events or consumption. Rate Setting establishes the actual price points. Operationalization makes those decisions work through entitlements, metering, quoting, billing and renewal. Amplitude matters because it can supply evidence across those steps, but it cannot rescue a company that starts testing £99 versus £119 before deciding what is being sold, to whom, and against which meter.
Amplitude itself is an instructive case. Its 2025 Form 10-K said customers committed to either event volumes or monthly tracked users, or MTUs. By August 2026, management said it had reduced pricing to a single meter, while the public pricing page showed event-based Growth and Enterprise plans and unlimited seats across every package.
That progression lets us grade Amplitude on the three framework steps most relevant to a pricing-testing platform.
| Exhibit: Amplitude pricing scorecard | Grade | Monetizely's rationale |
|---|---|---|
| Packaging | A- | The full platform and unlimited seats reduce artificial adoption barriers, but unlimited advanced experimentation and B2B account capabilities remain paid expansions on Growth and Enterprise. As of 13 August 2026, expanded packages were priced as a percentage of the platform plan. |
| Pricing Metric | A | Moving from contracts based on events or MTUs in 2025 to a single event-oriented meter in 2026 makes expansion easier to understand and links spend to the data flowing through the platform. |
| Operationalization | B+ | Management reported on 5 August 2026 that 70% of ARR closed in Q2 used the new model and 28% of total ARR had migrated, while the seller quoting process became simpler. Custom Growth and Enterprise rates plus percentage-priced add-ons still weaken price visibility. |
The scorecard matters for Amplitude users because the product's strongest lesson is also a pricing-testing principle: reduce the number of commercial variables before trying to measure their effects.
The best Amplitude tests change the offer without destabilising the billing model
Not every pricing decision belongs in an A/B test. Price points and package fences can often be randomised cleanly. Replacing seats with usage pricing across an enterprise product is a different matter because the change reaches metering, contracting, billing, sales compensation and customer budgeting.
Amplitude's 2025 Form 10-K says the platform is designed to let teams "quickly test hypotheses" and measure changes, and its Feature Experimentation product combines experimentation with feature management. Web Experimentation targets web experiences without requiring the same level of engineering involvement. The 5 August 2026 earnings call added another capability through Statsig: warehouse-native experimentation, 95% confidence intervals, CUPED, sequential testing, exposure checks, progressive rollouts and automatic rollback.
Those capabilities make four pricing questions particularly useful to separate.
| Exhibit: What to test in Amplitude | Suitable design | Primary commercial measure | Monetizely's view |
|---|---|---|---|
| Rate change | Randomly expose eligible new accounts to £99 versus £109 | Retained revenue per eligible account | Excellent Amplitude use case |
| Package fence | Move one material capability between tiers | Tier mix plus retained revenue | Excellent when entitlements can enforce the difference |
| Annual commitment | Test term and discount presentation among comparable new buyers | Annualised gross profit, cash collection and retention | Useful, provided contract terms are truly equivalent apart from treatment |
| Pricing metric change | Shadow-bill or run controlled sales cohorts before migration | Win rate, expansion, predictability and margin | Do not treat a seat-to-usage migration as a simple webpage A/B test |
The distinction prevents one of the most expensive mistakes in SaaS pricing: using a fast experiment to answer a slow structural question.
Current B2B pricing models make the point concrete. As checked on 13 August 2026, Jira remains fundamentally user-count based; Datadog combines product-specific usage units such as infrastructure hosts and custom events; Snowflake centres its platform economics on consumption credits; HubSpot Marketing Hub combines package pricing with Core Seats, marketing contacts and HubSpot Credits; Amplitude has deliberately moved towards events while keeping seats unlimited.
The table's implication is straightforward: a pricing experiment should measure movement in the same economic activity the eventual contract will monetise.
Datadog CFO David Obstler captured the operational side of that idea on 6 August 2026: enterprise pricing is generally "volume-based pricing", followed by customer usage against the commitment. A test that measures only landing-page conversion would miss much of the economics.
Account-level randomisation matters more than simply finding a larger sample
Consumer experimentation often treats a user or browser as an independent unit. B2B SaaS is rarely that clean. Five employees from one company may share a procurement process, discuss prices internally and ultimately buy one contract.
Randomising those five people into different price conditions contaminates the test. One employee may see £99, another £119, and the account executive may eventually quote £109. The experiment has created three prices but no interpretable treatment.
Amplitude now has the right building blocks for this problem. As of 13 August 2026, its Accounts expansion for Growth and Enterprise includes account-level analytics, group experimentation, account-level bucketing and B2B-oriented reporting.
The wider experimental literature reinforces the logic. A peer-reviewed 2024 Management Science study of an Airbnb pricing meta-experiment found that individual randomisation suffered meaningful interference bias and that cluster randomisation removed at least 20% of the treatment-effect estimate attributed to that interference. Airbnb is a marketplace rather than B2B SaaS, but the design lesson travels: when treated units influence one another, randomise at the level where the interaction occurs.
| Exhibit: Choose the experiment unit before choosing the price | Recommended unit | Why |
|---|---|---|
| Anonymous self-serve prospect with no known company | Stable visitor identity | Keeps the price consistent across return visits |
| Known B2B users under one workspace or domain | Account | Prevents colleagues seeing conflicting commercial terms |
| Sales-led enterprise prospect | Opportunity or account | Matches the unit actually quoted and contracted |
| Existing enterprise customer | Contract/account cohort | Avoids different prices for users covered by one agreement |
For most B2B SaaS pricing tests, account is therefore the default randomisation unit once identity is known. Amplitude's account-level bucketing makes that design technically possible; commercial discipline must ensure Salesforce, CPQ and billing preserve the same assignment.
Statistical power still matters. The 2013 peer-reviewed CUPED paper by Alex Deng, Ya Xu, Ron Kohavi and Toby Walker reported roughly 50% variance reduction in Bing experiments using pre-experiment data, effectively allowing equivalent power with about half the users or half the duration in their reported setting. Amplitude's 2026 Statsig demonstration explicitly showed CUPED and sequential testing in the product.
Yet variance reduction cannot repair the wrong randomisation unit. Better statistics on contaminated treatments merely produce a more precise answer to the wrong question.
Retained revenue should choose the winner, not checkout conversion
Price affects two things at once. Raising the rate increases revenue from every customer who buys, while usually putting some pressure on the number who buy. The winning price therefore cannot be identified from conversion alone.
Suppose a SaaS company tests £99, £109 and £119 among equal groups of qualified new accounts. A growth dashboard might call £99 the winner because its paid conversion rate is highest. A pricing team should calculate revenue per eligible account and then wait long enough to see whether the customers acquired at each price remain.
The economics can reverse the initial ranking.
| Exhibit: What the winning pricing test actually looks like | £99 control | £109 treatment | £119 treatment |
|---|---|---|---|
| Eligible accounts | 20,000 | 20,000 | 20,000 |
| Initial paid conversion | 5.0% | 4.7% | 4.2% |
| Initial MRR per eligible account | £4.95 | £5.12 | £5.00 |
| Retained payer rate after 90 days | 4.6% | 4.35% | 3.8% |
| Retained MRR per eligible account | £4.55 | £4.74 | £4.52 |
| Decision | Baseline | Promote | Reject |
The £109 price wins despite converting fewer accounts than £99. The £119 price looks respectable at checkout but loses its advantage after early retention is included.
A robust Amplitude implementation therefore needs an event path that connects price exposure to commercial outcomes. Rather than creating dozens of vanity events, we would require a small set of durable records:
the account's assigned experiment and variant;
the segment and acquisition source known at assignment;
the offer, package, rate and billing term displayed;
checkout, purchase and contract identifiers;
activation, expansion, downgrade, cancellation and renewal events.
Amplitude's core analytics already supports behavioural cohorts and funnel analysis, while its platform is designed to connect behavioural actions with acquisition, monetisation and retention outcomes. Its 2025 Form 10-K described 4,797 paying customers at year-end and said 98% of 2025 revenue came from subscriptions, making those post-purchase outcomes central to its own business as well.
A price test should consequently stay open conceptually after the experiment stops taking new traffic. The exposure phase may take two weeks; the decision window may require 30, 60 or 90 days of downstream behaviour.
HubSpot's 2026 experience is a useful warning against declaring victory too soon. CEO Yamini Rangan told investors on 5 August that "predictability has become a defining theme" in customer adoption and described pricing changes designed to improve visibility and control over spend. CFO Kathryn Bueker also said the deliberate product, pricing and go-to-market changes created a near-term headwind.
Pricing changes alter buyer behaviour, sales cycles and expansion together. A seven-day conversion read cannot capture all three.
Amplitude's own reset shows why simpler meters improve commercial execution
Amplitude's pricing history is unusually relevant because the company selling experimentation software has itself been conducting a multi-year commercial redesign.
In October 2023, Amplitude launched Plus starting at $49 per month and scaling to as many as 300,000 monthly tracked users. The offer bundled analytics, audience-management capabilities, A/B testing and feature management into self-service packaging. CEO Spenser Skates said at the time that analytics pricing had become too expensive and opaque, especially under event-based approaches.
By year-end 2025, Amplitude's 10-K described a model in which customers could commit to either events or MTUs, with discretionary overage charges when contracted usage was exceeded. That gave customers flexibility, but two possible scale meters also created more commercial logic for buyers and sellers to understand.
Management changed direction in 2026.
| Exhibit: Amplitude's pricing reset | Primary structure | Strategic meaning |
|---|---|---|
| October 2023 | Plus launched from $49/month, scaling to 300,000 MTUs; analytics and experimentation broadened in self-service. | Lower the free-to-paid barrier and expand product access |
| December 2025 | Contracts could commit to events or MTUs; overages could apply above committed volume. | Flexible, but customers and sellers still navigated two possible usage concepts |
| Q2 2026 | Management reduced the model to one meter; 70% of ARR closed in Q2 was on the new model and 28% of total ARR had migrated by 5 August. | Simplify consolidation, quoting and expansion |
| 13 August 2026 | Free includes 2M events monthly; Plus starts at $0 for the first 2M and can scale to 70M events; Growth and Enterprise state event-based pricing; seats are unlimited. | Make data volume the primary scale mechanism |
The sequence is more important than any individual price point: Amplitude first broadened the package, then simplified the meter.
Skates summarised the 2026 change directly: "We reduced down to a single meter to make it simpler for enterprises to add additional products." CFO Andrew Casey then connected the change to commercial execution, saying the model gives customers greater cost predictability and "simplifies the quoting process for our sellers".
Early operating evidence is encouraging but should not be over-read as causal proof. Casey reported that customers on the new approach were showing higher average ARR, greater multi-product attach and longer contract duration, while data usage against entitlement had reached an all-time high as of the 5 August 2026 call. Those results coincide with product changes, the Statsig integration and greater AI-driven data ingestion, so no public disclosure isolates pricing as the sole cause.
What Amplitude gets right is the direction. Unlimited seats remove a collaboration tax. A single event meter gives the buyer one principal variable to forecast. Broad platform access means a customer can experience adjacent products before a complex new purchase conversation.
What Amplitude gets wrong is the remaining gap in price predictability. Growth and Enterprise are still custom-priced, and advanced packages such as unlimited Feature Experimentation, Web Experimentation and Accounts are priced as a percentage of the platform plan. A customer whose event volume rises can therefore pay more for an experimentation add-on even when the number of experiments, teams or decisions has not risen proportionally.
Fixed advanced tiers would complete Amplitude's move towards predictable pricing
Monetizely's position on Amplitude's next reset is committed: keep events as the primary meter, publish event-volume bands further into the mid-market, and replace percentage-of-platform pricing for advanced experimentation and account capabilities with fixed, clearly defined expansion tiers.
Events should remain primary because Amplitude's value expands as customers instrument more product activity, and management already monitors usage relative to data entitlement as a principal monetisation indicator. Reintroducing seats would work against the company's platform strategy; returning to parallel MTU and event choices would surrender the simplicity gained in 2026.
Advanced experimentation is different. The customer is buying stronger decision infrastructure: unlimited concurrent tests, approval workflows, holdouts, stratified sampling, sticky bucketing and account-level experimentation. A fixed Growth-level experimentation package and a richer Enterprise governance package would make the incremental bill easier to explain than "a percentage of whatever you spend on event volume."
The broader market is moving in the same direction on predictability even where the chosen meter differs. HubSpot's Yamini Rangan said on 5 August 2026 that customers wanted costs that were transparent and tied to value. Datadog, by contrast, retains commitment plus usage economics; CFO David Obstler described its large-enterprise model as annual commitments alongside volume-driven consumption. Neither model implies that every SaaS business should adopt the same meter. Both show why buyers need to understand what causes the bill to move.
For operators using Amplitude to test their own pricing, five decisions matter more than adding another dashboard:
Make pricing experimentation part of the company's capital-allocation process. A pricing test changes the quality and quantity of future ARR. Review major results with the same seriousness applied to CAC, product investment and sales capacity.
Set one economic objective for each pricing cycle. A company seeking faster penetration should not judge an experiment by the same hurdle rate as a company trying to increase gross profit from an established customer base.
Align commercial incentives with the experiment's long-run measure. Sales and growth teams rewarded only for first-month bookings will naturally favour offers that close quickly, even when a different treatment produces better retained revenue.
Create a permanent pricing decision record. Preserve the hypothesis, eligible population, treatment, final effect and management decision. Three years of organised learning is far more valuable than three years of disconnected experiment dashboards.
Treat a pricing-model change as an operating-model decision. Test rates and packages aggressively, but do not migrate the primary meter until Product, Finance, RevOps, Sales and Engineering can meter, quote, bill and explain it consistently.
Amplitude can make SaaS pricing tests much more rigorous. Its greatest value, however, comes from imposing discipline on what happens after randomisation: preserving account-level exposure, following behaviour beyond conversion and linking product use to retained economics. The software supplies the evidence. The pricing strategy still has to decide what evidence is worth collecting.
Assumptions
The £99/£109/£119 scenario is a modelled SaaS case rather than observed Amplitude customer data; taxes, discounts, expansion, variable COGS and payment failure are excluded. Current vendor pricing was checked on 13 August 2026. Historical Wayback references below point to the stated archive periods and are cross-checked against Amplitude's contemporaneous investor disclosures and SEC filing because archive availability can vary by capture.
Footnotes
Monetizing Agentic AI: https://www.amazon.com/Monetizing-Agentic-AI-Handbook-Transformation/dp/B0H7Z13VKJ/
Amplitude, current pricing page, checked 13 August 2026: https://www.amplitude.com/pricing
Amplitude, Inc., 2025 Form 10-K: https://www.sec.gov/Archives/edgar/data/1866692/000119312526057847/ampl-20251231.htm
Amplitude, Inc., Q2 2026 earnings-call transcript, 5 August 2026: https://investors.amplitude.com/static-files/c501e2aa-38c5-4399-b395-0ef5f993325e
Amplitude, "Amplitude Brings Full Power of Digital Analytics to Every Team for Less", 17 October 2023: https://investors.amplitude.com/news-releases/news-release-details/amplitude-brings-full-power-digital-analytics-every-team-less
Internet Archive, Amplitude pricing archive, February 2024: https://web.archive.org/web/202402/https://www.amplitude.com/pricing
Internet Archive, Amplitude pricing archive, March 2026: https://web.archive.org/web/202603/https://www.amplitude.com/pricing
HubSpot, Marketing Hub pricing, checked 13 August 2026: https://www.hubspot.com/pricing/marketing
HubSpot, Q2 2026 earnings-call transcript, 5 August 2026: https://ir.hubspot.com/static-files/f868efed-4a81-4a8b-b0c9-a68cde86a345
Atlassian, Jira pricing, checked 13 August 2026: https://www.atlassian.com/software/jira/pricing
Datadog, pricing list, checked 13 August 2026: https://www.datadoghq.com/pricing/list/
Datadog, Q2 2026 earnings-call transcript, 6 August 2026: https://investors.datadoghq.com/static-files/2360a5cc-f17a-4731-b74a-fb6f4617baf5
Snowflake, pricing documentation, checked 13 August 2026: https://docs.snowflake.com/en/user-guide/snowflake-cortex/pricing
Alex Deng, Ya Xu, Ron Kohavi and Toby Walker, "Improving the Sensitivity of Online Controlled Experiments by Utilizing Pre-Experiment Data", WSDM 2013: https://doi.org/10.1145/2433396.2433413
David Holtz, Ruben Lobel, Inessa Liskovich and Sinan Aral, "Reducing Interference Bias in Online Marketplace Experiments Using Cluster Randomization: Evidence from a Pricing Meta-experiment on Airbnb", Management Science, published online 5 April 2024: https://doi.org/10.1287/mnsc.2020.01157