Product Analytics Market

We analyze three overlapping markets that we can think of as three rings.
Ring 1 — Product Analytics (Pendo, Amplitude, Mixpanel): What are users doing? tools that track in-app user behavior, feature adoption, retention, and engagement.
Ring 2 —Digital Adoption Platforms (DAP) / Onboarding (Pendo, Appcues, Userpilot, WalkMe): How do we guide users? tools that deliver in-app guidance, walkthroughs, tooltips, surveys, and onboarding flows to help users get value from software faster.
Ring 3 — Experimentation / A/B Testing (Optimizely, LaunchDarkly, VWO, Statsig): Which version of X works better?

Market size

Product analytics is broader by definition. The product analytics market encompasses retail, e-commerce, healthcare, financial services, and more — essentially any company with a digital product needs it. DAP and A/B testing are more specialized use cases within that universe. The DAP market is actively consolidating. WalkMe got acquired by SAP, Heap by Contentsquare, and there's rapid M&A pressure across the space = a sign the standalone DAP market may be getting absorbed into larger platforms rather than growing independently. The DAP market is growing at ~19% CAGR and product analytics at ~15% CAGR — so the DAP market is actually growing faster from a smaller base. But in raw dollar terms, product analytics is clearly the bigger market

A/B testing has the slowest projected growth of the three — which makes intuitive sense for a few reasons:
1. It's a more mature category. Optimizely has been around since 2010, VWO since 2009. The basic concept of split-testing a webpage is well understood and well-penetrated in enterprise marketing teams. There's less greenfield to capture.
2. It's getting commoditized from below. Tools like PostHog, GrowthBook, and Statsig offer free or very cheap experimentation, which compresses pricing power across the whole category.
3. It's being absorbed into larger platforms. Amplitude added Amplitude Experiment. Optimizely expanded into a full DXP. Adobe has Adobe Target. As experimentation becomes a module inside bigger suites rather than a standalone purchase, the standalone A/B testing market naturally grows more slowly — spend shifts to the broader platform.
4. DAP has a stronger AI tailwind. DAPs are uniquely positioned to help enterprises adopt generative AI tools — since AI copilots require behavioral change, prompt engineering education, and output validation guidance that DAPs are built to provide — and Gartner noted in early 2026 that 63% of enterprises plan to deploy generative AI tools broadly, yet fewer than 15% have structured adoption programs in place. That's a massive unmet need that's fueling DAP's faster growth rate.

So we have Product Analytics (biggest, fastest absolute growth) > DAP (small but accelerating, strong AI tailwind) > A/B Testing (similar size to DAP today, but slowest growth and most commoditization pressure).

The product analytics and user onboarding market

In this broader market companies have many niches. Let’s use Pendo as an example. Pendo positions itself as an integrated product experience platform, unifying analytics, in-app guidance, user feedback, and roadmapping into a single solution. Its core differentiator is combining behavioral analytics with the ability to directly act on that data through in-app experiences — without needing two separate tools. It acts at the intersection of two markets: Product analytics and digital adoption platform (DAP).

Pendo’s product analytics rivals:

• Amplitude, Mixpanel, Heap, FullStory, and Smartlook are among the top alternatives in the product analytics space.
• PostHog is also popular, especially among developer-led teams.

User Onboarding / In-App Guidance rivals:

• Appcues, WalkMe, Userpilot, and UserGuiding are Pendo's main competitors in the user onboarding category. Appcues leads head-to-head migrations, while Userpilot has also gained customers from Pendo. WalkMe (acquired by SAP in 2024) is most frequently evaluated for large enterprise use cases, given its desktop coverage and workflow automation capabilities.
• Whatfix is another strong player, particularly for enterprise employee training.
• In the broader customer experience category, Pendo also competes with Zendesk, Intercom, and Trustpilot.

The A/B Testing / Experimentation Market

The A/B Testing Software market was valued at $1.5 billion in 2025, projected to reach $1.67 billion in 2026, and forecast to expand to $4.82 billion by 2036 at an 11.2% CAGR. Key players break into two camps:

1. Marketing/CRO-Focused (no-code, visual editor, website testing): Optimizely, VWO + AB Tasty
2. Engineering/Feature Flag-Focused (developer-first, full-stack): LaunchDarkly, Statsig.

How A/B Testing Relates to the Previous Companies
The product analytics tools we discussed do have A/B testing features, but it's generally secondary or bolt-on, not their core identity:

• Pendo & Appcues/Userpilot — can test different onboarding flows, guide variants, and UI patterns, but they're not full experimentation platforms. It's lightweight A/B testing within the context of in-app experiences.
• Amplitude — has a dedicated product called Amplitude Experiment (feature flagging + A/B testing), which is actually a genuine competitor in this space, though analytics remains the core.
• Mixpanel — has basic A/B test analysis reporting but doesn't run the tests itself; you'd use it to analyze results from another tool.

So the dedicated A/B testing / experimentation market is a related but distinct category — focused on conversion rate optimization (CRO), feature flagging, and multivariate testing at scale, often across marketing pages, not just in-app product flows.

The key strategic tension: as product analytics tools add experimentation (Amplitude Experiment) and experimentation tools add analytics (Optimizely's NetSpring acquisition), the categories are converging — which is why you're seeing rapid M&A consolidation across both spaces simultaneously.

There’s the progression of: Observe → Test → Optimize Automatically
Think of it as a maturity curve. Most companies are at level 1 or 2. Very few are at level 3.
Level 1 — Observe only (pure analytics): Mixpanel, Heap/Contentsquare Level 2 — Observe + manual A/B test: Amplitude, Pendo, Optimizely, VWO Level 3 — Observe + adaptive experimentation:: the interesting ones below:

Statsig is probably the most native here. Statsig's Autotune feature adds contextual bandits on top of its experimentation layer — you choose what the model personalizes on, it handles feature engineering, model selection, and training automatically, and it supports both binary and continuous outcome optimization. Because Statsig also has full product analytics, it's one of the only platforms where the feedback loop is genuinely closed — analytics data feeds directly into bandit decisions.

Eppo (now Eppo by Datadog) was the most statistically rigorous pure-play here before its acquisition. Eppo's "contextual bandit" system automatically identifies new variants of websites, apps, or AI models and actively explores their performance by routing increasing traffic to promising variants — which is the true explore/exploit loop. Datadog acquired Eppo in May 2025 specifically to close the loop between feature flags, experimentation, and product analytics into one unified platform, creating arguably the most complete engineering-side stack for this.

Amplitude Experiment is at level 2.5 — it has MAB-style traffic allocation but it's not as sophisticated as a true contextual bandit. The strength is that it draws on Amplitude's behavioral analytics data to define experiment segments and success metrics, so the feedback loop between analytics and experimentation is tighter than most.

Kameleoon is a lesser-known but technically sophisticated player. Kameleoon has introduced contextual bandits specifically for experimentation and personalization — unlike static A/B tests, contextual bandits allocate traffic dynamically based on user context, enabling faster insights and more impactful outcomes.

Braze is doing this specifically in the marketing/messaging layer. Braze's MAB framework uses reward-based reinforcement learning where each campaign interaction feeds back a reward signal, allowing the model to update confidence levels across variations and adapt within seconds rather than hours or days — combining statistical efficiency with the flexibility needed for fast-moving campaigns. The key difference is Braze is focused on messaging/lifecycle, not product UI.

Most companies can't actually do true contextual bandits at scale because it requires:

1. Context features available at serving time — user attributes, session info, device, cohort tags, etc. These can be pre-computed and cached, not streamed.
2. A reward signal that can be logged and joined back — which can happen asynchronously and in batch.
3. A model serving layer that's fast enough to make a per-user decision at page/feature load time.
4. Periodic model retraining — which can be batch, scheduled, or even manual at lower traffic volumes.
5. Feature flags tightly coupled to bandit decisions — for consistency and atomicity of variant assignment
6. Statistical rigor — specifically variance reduction, non-stationarity handling, and guardrail metric monitoring. For variance reduction, techniques like CUPED (which Eppo pioneered commercially) reduce noise in your reward estimates, meaning the bandit converges to the right arm faster with less data. This is especially valuable at lower traffic volumes. Non-stationarity handling: if user preferences shift over time (seasonality, product changes, etc.), the bandit needs a mechanism to down-weight old observations so it doesn't exploit a stale winner. This is often overlooked. Guardrail metrics: pure bandit optimization can over-exploit a variant that wins on one metric (e.g. clicks) while harming another (e.g. retention). You need the statistical framework to simultaneously monitor guardrail metrics and pump the brakes if they degrade.

This is why the Datadog/Eppo combination is strategically significant — Datadog explicitly positioned it as enabling engineers to track code changes with feature flags, while data scientists design and measure experiments, and business analysts understand product usage — all on one platform. That end-to-end loop is what makes true explore/exploit possible.

The first four are the data infrastructure. The last two are more about correctness — without them you can run a bandit, but you'll either get inconsistent experiences (#5) or optimize toward the wrong thing and not know it (#6). The real differentiator between companies is actually #3 — the model serving infrastructure — not the data pipeline speed. That's where companies like Statsig and Eppo have invested heavily, and where traditional analytics tools like Amplitude or Mixpanel have the biggest gap to close.

The combined table:

Things that stand out when you see it all together:

• Valuation vs. ARR mismatches. LaunchDarkly is valued at ~$3B at 300M ARR. Optimizely meanwhile has more ARR ($400M+) but a lower estimated valuation (~$600M), weighed down by its $1.1B debt restructuring in late 2024. Amplitude is the only publicly traded pure-play, but its market cap has compressed significantly relative to its ARR.
• Pendo and Amplitude sit at the center and touch all three rings. Optimizely has been actively expanding from Ring 3 into an entirely different orbit on CMS, marketing content and personalization (it's now a full Digital Experience Platform DXP) and 52% of its ARR comes from those. LaunchDarkly and Statsig are squarely in Ring 3 but from the engineering side — feature flags first, experimentation second.
• The pricing philosophies split cleanly into two camps. The engineering-first tools (LaunchDarkly, Statsig) offer transparent, low entry-point pricing to get developers hooked. The enterprise-first tools (Pendo, Optimizely, WalkMe) are fully sales-gated with no public pricing — a deliberate strategy to control deal size and prevent self-serve downmarket erosion. Pendo commands a much higher sales-gated price than its user onboarding peers. And at the lower end, Appcues and Userpilot offer nearly identical entry pricing ($249/mo) but are an order of magnitude smaller in scale.
• Amplitude is the only public company in this entire landscape, making it the clearest benchmark for how the market values this category — and at ~2x EV/Revenue, the market is currently being quite skeptical of growth prospects across the board.