Kameleoon Review 2026: Features, Pricing, and Fit

Kameleoon is an enterprise experimentation and personalization suite that does something unusual for its tier: it publishes a starting price, and it documents its statistics in public. Both of those make it easier to evaluate honestly than most of its direct competitors, and both come with a catch worth understanding before a demo call. This review covers what the platform includes, what the published plan actually caps, which statistical capabilities live only in the quote-only tier, and a worked example, computed with this blog’s own engine, of how much experiment fits inside the entry plan. For the category-level picture, see our neutral comparison of CRO and A/B testing tools, and for the head-to-head, our AB Tasty vs Kameleoon comparison.

Disclosure before anything else: Donnu is an A/B testing tool and competes with Kameleoon on part of this scope. This piece is written to be useful even to a reader who ends up choosing Kameleoon.

What Kameleoon is, one line per piece

Piece	What it does	Plan where it lives
Web experimentation	A/B, split, and multivariate tests on pages, client-side through the snippet	Starter and up
Segmentation	40+ criteria for defining who sees what, plus unlimited audiences and metrics	Starter and up
Experiment hygiene	Mutually exclusive groups, SRM detection, multiple testing correction	Starter and up
Feature management and rollout	Flags and progressive releases tied to the same platform	Enterprise
Server-side and mobile app testing	SDK-based experimentation inside application logic and native apps	Enterprise
Advanced personalization and AI targeting	Predictive per-visitor targeting on top of the segmentation engine	Enterprise
Advanced statistics	Frequentist and Bayesian methods, CUPED, multi-armed and contextual bandits	Enterprise

The honest read of that table: the entry plan is a web experimentation product, and the suite is the Enterprise plan. That is not a criticism, it is a qualifying question. If the reason Kameleoon is on your shortlist is variance reduction, bandits, or server-side testing, you are shopping in the quote-only tier from the first call, and the published 495 dollar figure is not the number that will show up in your contract.

Pricing: what is published and what is not

Numbers checked on the official plans page on 14 August 2026:

Plan	Price	Experiments	Tested users per month	Notable inclusions
PBX Starter, free trial	Free for 30 days, no credit card	Up to 3	Not stated on the page	Prompt-based experimentation, sequential testing, 40+ segmentation criteria, Figma integration
PBX Starter	From 495 dollars/month	Up to 10	50,000 monthly tracked users	Everything in the trial, plus mutually exclusive groups, multiple testing correction, SRM detection, unlimited metrics and audiences
Enterprise	Quoted	Unlimited	Unlimited	Feature management and rollout, mobile app testing, server-side testing, advanced personalization, JS and CSS code editor, multi-armed and contextual bandits, CUPED, holdouts, frequentist and Bayesian methods, AI targeting, data warehouse integrations, SSO, dedicated CSM and TAM

Two things worth pulling out of that table.

First, publishing a floor is a real transparency win in a category where most enterprise vendors publish nothing at all. Our A/B testing tools pricing comparison shows how rare that is. It lets a buyer disqualify the tool in five minutes instead of five meetings, which is worth something even to the vendor.

Second, the published floor answers only the smallest version of the question. Unlimited traffic, personalization, feature management, and the advanced statistics are all Enterprise, and Enterprise carries no published number. So the useful way to read the plans page is as a qualification gate, not a price list: it tells you cheaply whether your requirements even fit inside Starter.

The statistics, and why the documentation matters

Kameleoon’s public documentation lists three selectable methods for analyzing an experiment, each with a stated trade-off:

• Fixed-sample frequentist. Requires a hypothesis and a sample size defined before launch. The documentation frames the trade directly: maximal statistical power, at the cost of a rigid predefined sample size and test duration. This is the classic approach our statistical significance guide walks through.
• Bayesian. Incorporates existing knowledge through priors. The documentation is candid that this can lead you in the wrong direction if the prior data is wrong, which is a more honest framing than most vendor material offers.
• Sequential. Decides dynamically as data accumulates, without estimating a sample size in advance, and provides a valid confidence interval at any point during the run. The stated cost is lower statistical power than fixed-sample methods, which matches the treatment in our guide to sequential testing.

Alongside the three methods, the same documentation describes complementary techniques that add to the chosen method rather than replace it: multiple testing correction, the necessary care when several variations or metrics are read at once, and CUPED, which reduces the sample size required by exploiting pre-experiment data, and which the documentation says works best with returning visitors, substantial history in the platform, and pre-experiment conversions correlated with the live experiment’s metric.

This is worth spelling out because it inverts a common assumption. A buyer who reads “sequential testing included on the entry plan” as the more advanced option has it backwards: sequential is what lets you look early without lying to yourself, and it pays for that with lower power, meaning more traffic for the same detectable effect, which interacts directly with the tested-user cap covered next.

Worked example: how much test fits in 50,000 tested users

The Starter plan’s binding constraint is not the 10-experiment limit, it is the 50,000 monthly tracked users. Here is the arithmetic, computed with the same sampleSizePerVariant function that runs the calculator below.

The scenario: an ecommerce product page converting at 4.5%, with 30,000 eligible visitors per week available to the test. Set your own baseline in the calculator:

Current conversion rate%Minimum detectable effectrelative (%)absolute (pp)Confidence909599%Power8090%Visitors per week (total)Testtwo-sidedone-sided

-Visitors per variation

-Total (2 variations)

-Estimated duration

Two-proportion normal approximation, 2 variations (50/50). Tweak the inputs and watch it update live.

Ambition (relative lift)	Users per variant	Total for a 2-arm test	Days at 30,000/week	Fits in 50,000 monthly tracked users?
20% (4.5% to 5.4%)	9,118	18,236	5	Yes, with room for a second test
15% (4.5% to 5.175%)	15,860	31,720	8	Yes, and it uses most of the month
10% (4.5% to 4.95%)	34,897	69,794	17	No, it exceeds the monthly cap by itself

All four numbers come out of the frozen formula at 95% confidence and 80% power, two-sided. The lesson is not that the cap is stingy, 50,000 tracked users at 495 dollars is a reasonable entry package. The lesson is that the plan silently sets your minimum detectable effect. On this baseline, the entry plan supports chasing 15% relative lifts and above. Chasing a 10% lift, which is exactly the range where most real page-level improvements live, needs either a bigger plan, a higher-traffic page, or two consecutive months of a single test, and the two-month version brings its own seasonality problems.

Note also how this compounds with the method placement above: sequential analysis, the method included in Starter, delivers lower power than the fixed-sample method by the vendor’s own documentation, so the sample requirements in that table are the optimistic version of the picture for a Starter user reading a test sequentially.

Reading the result

Say the 15% test ran a bit longer and closed at 25,000 users per variant, with 1,125 conversions on control (4.50%) against 1,290 on the variation (5.16%).

Control (A)

VisitorsConversions

Variation (B)

VisitorsConversions

Confidence90%95%99%

Control (A) · Rate-

Variation (B) · Rate-

Relative lift-

p-value-

95% CI of the difference-

Two-sided two-proportion z-test. "Not significant" almost always means not enough sample, not that the versions are equal.

Running those four numbers through this blog’s significance engine: z = 3.44, p-value ≈ 0.00058, a 95% confidence interval on the difference of +0.28 to +1.04 percentage points, and an observed relative improvement of +14.7%. That is a clean, well-powered result, and it fills the monthly tracked-user allowance exactly. Which is the practical point: on the entry plan, one decisive test per month is roughly the shape of the operation, and a program that wants weekly test velocity is an Enterprise conversation.

Real strengths

• A published entry price in a quote-only category. Rare, and genuinely useful for disqualifying fast.
• Statistics documented with their trade-offs. Three named methods, each with a stated cost, plus explicit conditions for when CUPED helps. That is a level of candor that makes an independent evaluation possible.
• Experiment hygiene included at entry. Mutually exclusive groups and SRM detection on the entry plan are the unglamorous features that keep a program honest, and they are often upsells elsewhere.
• One vendor across experimentation, personalization, and feature management. For teams that genuinely need all three, consolidation has real operational value.
• A 30-day trial with no credit card. Enough runway to instrument a real page and see the workflow before a purchase conversation.

Limitations and watch-outs

• The interesting statistics are Enterprise. CUPED, frequentist and Bayesian analysis, and bandits sit above the published tier. Confirm what your quote actually includes.
• The tracked-user cap sets your detectable effect. As the worked example shows, 50,000 monthly tracked users on a mid-single-digit baseline caps ambition at roughly 15% relative lifts.
• Enterprise pricing is unpublished. The floor tells you where the ruler starts, not where your contract lands. Treat any third-party price comparison as a dated estimate, not a current quote.
• Suite breadth is cost for the wrong team. If your actual need is testing a few pages honestly, most of what you would pay for stays switched off.
• Vendor structure is a live variable in this category. Kameleoon was independent at the time of this check, while its closest French competitor became part of a larger group in January 2026. Structure changes reprice packages, so confirm current status before a multi-year commitment.

How to evaluate it in a pilot

1. Do the cap arithmetic first. Put your real baseline and traffic into the calculator above and see which lift the plan lets you chase. If the answer is “only large ones”, the tool tier is wrong, not the tool.
2. Ask which plan each statistical feature belongs to, in writing. Specifically CUPED, the analysis methods, and bandits.
3. Run an A/A test during the trial. Confirm the split matches the configuration and that the platform reports the non-effect honestly.
4. Test the sequential workflow deliberately. If your team will look at results daily, decide upfront that sequential is the method, and accept the power cost instead of discovering it later.
5. Confirm every number on the vendor’s own pages. Plans and inclusions move; the value of this review is the date of the check, not a permanent guarantee.

Kameleoon review: final analysis and who it is for

It fits mid-market and enterprise teams that need experimentation, personalization, and feature management from one vendor, that have traffic well above the entry cap, and that value published documentation of the statistics over a black-box verdict. For a team with returning visitors and a real analytics history, the Enterprise-tier CUPED support is a legitimate reason to pay enterprise money, since variance reduction converts directly into shorter tests.

It fits less well for small teams whose need is honest A/B testing on a handful of pages, where a 495 dollar per month floor buys a lot of platform that will stay switched off, and for teams whose single must-have capability turns out to live only in the quote-only tier. For the direct rival comparison, see AB Tasty vs Kameleoon; for the client-side versus server-side question that decides much of the Enterprise scope, see client-side vs server-side A/B testing.

If your checklist is closer to “test properly and trust the result” than to “run a personalization program”, Donnu is a lighter option focused on web A/B testing with a conservative read of the result. It does not replace Kameleoon’s personalization depth, feature management, or enterprise services, and saying otherwise would be dishonest. If the lighter case is yours, start a free 14-day trial.

---

Read also: CRO Tools Compared · AB Tasty vs Kameleoon · A/B Testing Tools Pricing Compared · What Is CUPED

References

• Kameleoon. Plans. Official plans page: PBX Starter from 495 dollars per month with up to 10 experiments and 50,000 monthly tracked users, the 30-day no-card trial limited to 3 experiments, and the Enterprise inclusions (feature management, server-side, mobile, advanced personalization, bandits, CUPED, frequentist and Bayesian methods, AI targeting, SSO). Checked 14 August 2026. kameleoon.com/plans.
• Kameleoon. Choosing the right statistical method for A/B testing. Documentation of the three selectable methods and their trade-offs, plus multiple testing correction and CUPED as complementary techniques. docs.kameleoon.com.
• Deng, A., Xu, Y., Kohavi, R. and Walker, T. Improving the Sensitivity of Online Controlled Experiments by Utilizing Pre-Experiment Data. WSDM 2013, the original CUPED paper. exp-platform.com.
• Kohavi, R., Tang, D. and Xu, Y. Trustworthy Online Controlled Experiments: A Practical Guide to A/B Testing. Cambridge University Press, 2020. Chapters on platform choice, SRM, and variance reduction. Companion material at experimentguide.com.