Amplitude bought Command AI in October 2024 and shipped it as Guides and Surveys in February 2025. That closed the last real gap between the two halves of this category. As of August 2026, Amplitude, Pendo, Appcues, and Userpilot all ship in-app guides, all ship survey capture, and three of the four ship a module for measuring the AI agents running inside your product.
So “buy an analytics tool and an adoption tool” is now a way to pay twice for the same capability — and the second copy targets its own definition of the same cohort. This stack is not four purchases. It is one decision about where the event taxonomy lives, one guidance layer chosen against that decision, and a named route from a behavioral cohort to a CSM.
The shape
- Amplitude is the event taxonomy and the surface everyone is allowed to look at. Every Amplitude plan includes unlimited seats, and the meter is event volume rather than people. Free is $0 with 2M events per month, no credit card and no expiry, and it carries basic product analytics, cohorts, session replay, Guides and Surveys, and AI Agents with MCP. Plus starts at $0 with the first 2M events free and scales by volume to 70M events per month. Growth and Enterprise are quote-only and add advanced behavioral exploration, SSO and project permissions, and 20K then 50K monthly session replays. The seat model is the reason this layer goes at the bottom: the people who need to read adoption data — CSMs, support, PM, sales — outnumber the people who build flows several times over, and every other tool here bills for the readers.
- Exactly one guidance layer sits on top of it. Pendo, Appcues, Userpilot, or Amplitude’s own Guides and Surveys — one of them, chosen against the fork below. The count matters more than the choice: two guidance layers is the most common way this stack goes wrong, and it usually arrives by acquisition of a team, not by decision.
- The CS route is a cohort, not a report. A cohort in Amplitude — “admin signed up, has not invited a second user by day 14” — is the same object that targets the in-app guide and fires the CSM alert. One definition, two consumers. That is the whole mechanism, and it is what makes the difference between CS acting on usage and CS acting on a feeling about the account.
Named handoffs
- A user fails to reach an activation event by its deadline → the Amplitude cohort updates → the guidance layer targets that same cohort in-app. The guide is aimed by the analytics tool’s definition, never by a second one written in the guide builder.
- Cohort membership crosses a threshold at account level → the CS system of record and Slack get the alert. This is the usage-drop alert, and it is the handoff that turns instrumentation into a CSM task with a name and a date on it.
- A guide is shown → its outcome event fires back into Amplitude → the guide’s effect is measured where the baseline lives. A guidance tool grading its own guides reports engagement with the guide. The analytics tool reports whether the behavior changed.
- A survey response lands → the AI Feedback Agent maps the theme to what that user actually did. Amplitude’s agent line, announced 17 February 2026, is a Global Agent plus four specialized agents including Web Experimentation and AI Feedback, with MCP reaching Claude, Cursor, Figma, Notion, and GitHub. Verbatims tied to behavior beat verbatims in a spreadsheet.
- Quarterly → someone counts live experiences and retires the dead ones. On Appcues this is enforced by the meter itself: Start allows 10 published experiences, Grow 25, Enterprise 100. On the others nothing stops the sprawl, which is why it needs a calendar entry.
The rule that keeps this from costing twice
Cohort definitions live in one place. If the in-app guide targets “not activated” as Userpilot defines it while the CSM alert fires on “not activated” as Amplitude defines it, the two drift apart inside a quarter — different session windows, different event names, different handling of the user who did the thing once and stopped. The CSM sees an alert for an account that the product team’s dashboard says is healthy, and after the third time, the CSM stops opening the alert.
The practical form of the rule: the guidance layer consumes cohorts, it does not author them. Pick the tool that can accept the cohort from your analytics tool over the one with the nicer segment builder, because you are not going to use the segment builder.
The fork that decides cost
Three compositions are defensible. Buying all four is not.
- Amplitude alone. Analytics, cohorts, session replay, and Guides and Surveys under one contract with unlimited seats. Correct when your guide volume fits the base tier — unlimited live guides, A/B testing for nudges, custom CSS, and advanced theming are a Growth and Enterprise add-on, not part of the base. Cheapest by a wide margin, and fastest to stand up: no second SDK, no cohort sync to maintain.
- Amplitude plus Appcues. Correct when the person building experiences is a CSM or PM who will not be given an analytics seat’s worth of training. Appcues meters on MAU, number of installations, and published experiences, and includes implementation services, onboarding support, and a named experience manager on every plan. It publishes no price as of August 2026 — Start, Grow, and Enterprise all sit behind a call — so budget a quote cycle before committing to this path.
- Pendo alone. Correct when analytics depth and the in-app layer need to be one contract, when you want journey orchestration and warehouse sync, or when you want to query product data from Claude over MCP. Vendr reports a $49,000/year median across 572 anonymized transactions, from $17,815 to $151,590, with 12.1% average negotiated savings. Budget a named taxonomy owner alongside it; Pendo deployments without one degrade into dashboards nobody opens.
- Userpilot alone is the budget floor: $299/mo Starter for up to 2,000 MAU and 3 seats, from $849/mo Growth from 2,000 MAU with 15 seats. The routing argument against Pendo is in Pendo vs Userpilot, and against Appcues in Appcues vs Userpilot.
Cost reality
A B2B SaaS product at 25,000 MAU, with 8 CSMs and support who need to read the data and 3 people who build experiences:
- Amplitude-alone: the free ceiling of 2M events/month works out to 80 events per user per month at that MAU count, so a real product crosses it and lands on Plus, which scales by event volume from $0. Run the estimator on your own event count before assuming a number; the seat line stays $0 regardless of whether it is 11 readers or 60.
- Amplitude plus Userpilot: the Userpilot line alone is roughly $10.2K/year at the $849/mo Growth floor, and 15 seats covers 3 builders with room. Above 2,000 MAU that floor moves with your MAU, so price it at 25,000, not at the headline.
- Amplitude plus Appcues: unpriced until you call. The published constraint that will bite is the experience cap, not the MAU band — 25 live experiences on Grow goes fast once onboarding, feature launches, and NPS all run at once.
- Pendo-alone: the $49K/year median is the honest planning number, and roughly 5x the Userpilot floor. What that buys is session replay one tier earlier, orchestration, warehouse sync, and no seat ceiling on the builder side.
The two costs nobody puts in the model: the taxonomy owner, and the quarterly audit. Both are people, not licenses, and skipping either one is what turns a $10K stack into shelfware faster than picking the wrong vendor does.
Variations and when to swap
- Add a dedicated guidance layer to an Amplitude-only stack the moment you need A/B testing on nudges, custom CSS, or more live guides than the base tier allows — check the add-on price first, because the add-on sometimes closes the gap that motivated the second vendor.
- Swap Appcues for Userpilot when the published-price ladder matters more than the no-feature-gate promise, and you have three or fewer builders on Starter or up to 15 on Growth. Swap the other way when your team is larger than 15 builders or when you want every capability from day one without tier negotiation.
- Collapse to Pendo when your analytics questions routinely end in “watch the replay of the accounts that churned” and you are already staffing a product-ops owner. Do not collapse to Pendo to save money; on this evidence it costs more.
- Mobile changes the answer. Userpilot’s mobile support is a paid add-on on Growth and Enterprise, unavailable on Starter, with no mobile session replay. If the adoption problem is in a mobile app, get that quoted before treating $849/mo as the number.
What this stack does not replace
- It is not a customer success platform. No health scoring, no renewal forecasting, no CSM task management. That is Gainsight, ChurnZero, or Vitally — see the customer retention stack.
- It is not an onboarding delivery system. Kickoff plans, customer-facing portals, and implementation project tracking are the customer onboarding stack. This stack instruments what happens after login, not what happens in the shared plan.
- It does not decide what “adopted” means. Someone has to define the activation event and defend it. Start from activation and feature adoption rate; an instrumented product with a vague activation definition produces precise numbers about nothing.
- It does not write the in-app copy. Every tool here ships an empty builder.
Watch-outs, each with a guard
- The same capability arrives in three invoices. All four vendors now sell in-app guidance, and the overlap is invisible on a renewal spreadsheet that lists them under different categories. Guard: before renewing any of these, write down which single tool owns guides, which owns the taxonomy, and cancel the third.
- Two cohort definitions for one concept. Guard: the guidance layer consumes cohorts and never authors them; make that a written rule before the first flow ships, because retrofitting it means rebuilding every live experience.
- Experience sprawl outlives its authors. A guide written for a UI that shipped 18 months ago is now a lie your product tells new users. Guard: a quarterly audit with a named owner, and on Appcues let the published-experience cap force the conversation rather than raising the tier reflexively.
- MAU-metered guidance bills you for succeeding. Adoption work grows MAU, which grows the bill on Pendo, Appcues, and Userpilot alike. Guard: price the guidance layer at your 12-month MAU projection, not today’s, and check whether the analytics half of the stack — metered on events with unlimited seats — should absorb more of the work.
- Agent-measurement modules are new and priced separately. Pendo reports measuring more than 350 agents and 2.5 million prompts a week; Amplitude gates AI Visibility prompts by tier at 500, 2,500, and 5,000. Guard: if you are instrumenting an AI agent in your product, count prompts before picking a tier — that meter is not the MAU meter and it will not scale with it.
Match rules
Right pick when: you run a self-serve or hybrid product where usage is visible before the renewal conversation, CS is being asked to explain why an account is at risk with evidence rather than a call summary, and someone can be named as the owner of the activation definition. It fits best from roughly 5,000 MAU up, where the free analytics tier stops covering you and the first real vendor decision arrives.
Wrong pick when: your product is high-touch enterprise with 40 accounts and a named CSM on each — the CSM already knows what the customer is doing, and instrumentation adds a maintenance job with no new signal. Also wrong when nobody will own the taxonomy: the failure mode is not that the stack underperforms, it is that the numbers quietly go wrong and a CS org acts on them for two quarters.
If you can only do one thing: stand up Amplitude Free and define one activation event. It costs nothing, it carries unlimited seats, and it forces the argument about what “adopted” means before any vendor is in the room to answer it for you.