Product analytics for B2B SaaS: a B2B product analytics platform with account-level product analytics, not just user counts

You need account-level analytics the moment a single customer contains more than one user, which in B2B is essentially day one. User-level metrics hide the distribution that matters. Ten active users at one company is one active account, and a dashboard that reports ten looks four times healthier than one reporting from a base of forty accounts.

The practical test is whether you can answer this without exporting anything: how many of our accounts used feature X at least once last month, and how many used it in three separate weeks. If getting there means a spreadsheet, you are running user-level analytics on a B2B product and your adoption numbers are inflated by your largest customers.

There is a second reason that shows up later. Renewal risk is concentrated. A consumer product losing 3 percent of users loses 3 percent of revenue. A B2B product can lose 1 percent of users and 20 percent of revenue if those users were the champions inside two large accounts. Account-level reporting is what makes that concentration visible while there is still time to act on it.

User-level analysis stays useful for the things it is good at: onboarding funnels, individual session paths, feature discovery. Keep it. Just stop reading it as a measure of customer health.

Affordability in this category is decided by the billing unit, not the entry price, and the units are not comparable to each other.

Event-based billing, which is what Amplitude and Mixpanel use, tracks how busy your product is. That is usually kind to B2B, because a small number of accounts can generate a lot of value without generating much volume, and seats do not cost extra on Amplitude. It punishes chatty instrumentation, so an over-eager tracking plan can cost real money.

Monthly active user billing, which is what Pendo uses, is the one to think hardest about. It charges you for exactly the outcome your product team is paid to produce. A successful quarter of adoption work arrives as a larger invoice.

Session billing, used by Heap and Fullstory, is hard to forecast in B2B because your heaviest accounts log in constantly. Fullstory compounds this by not publishing its session definition, so the meter cannot be modeled before you commit.

For a genuinely small team, the cheapest honest path is usually Amplitude's free tier, because unlimited seats mean the whole company can look without raising the bill, plus a decision to delay the account-level add-on until you actually have enough accounts for the distinction to matter. Below roughly fifty customers you can still read the list by hand, and there is no shame in doing that until it stops working.

Pick tools that collect before anyone decides what to measure, and pick a small number of metrics you will not change.

Autocapture is the part that removes the engineering dependency. Heap, Fullstory and Pendo all record behavior first and let you define what counted as an event afterwards, which means a new question does not become a ticket in the engineering backlog. Amplitude, Mixpanel and PostHog all assume a tracking plan exists before the data is worth anything, and building that plan is a project rather than an afternoon.

The second half is discipline, and it is the half people skip. Choose three or four features that a customer could not leave you without rebuilding something. For each one, track how many accounts touched it at all, how many used it in three separate weeks, and how many tried it and stopped. Report the same three numbers the same way every month. The third number is the strongest early churn signal in most B2B products and almost nobody looks at it.

What you do not need is a data scientist to run that. What you do need is somebody who refuses to change the definitions every quarter, because a metric that gets redefined twice a year cannot show you a trend.

Four decisions, in this order, and getting them wrong is why adoption numbers get argued about instead of acted on.

First, who could have adopted. Only count accounts on a plan that includes the feature and that have been live long enough to reach it. Including accounts that cannot use the feature makes every adoption rate look bad and hides the ones that are genuinely stalling.

Second, what counts as adoption. One click is not adoption. Pick a threshold that means the account got value: three uses, or use in two separate weeks, or a completed outcome. Write it down, because you will be asked in three months and the answer has to be the same.

Third, which entity you are counting. This is the account-level question, and it is where the group analytics add-on becomes unavoidable. Counting users will overstate adoption wherever one enthusiastic team hides a quiet company.

Fourth, how long they get. Adoption without a window is a cumulative number that only goes up and therefore says nothing. Thirty days from account creation, or thirty days from feature release, both work. Pick one.

Then keep the reason attached. An account that adopted a feature and abandoned it has told you something specific, and the specifics are usually in a support ticket rather than in the event stream.

Several tools will surface anomalies automatically, and it is worth being precise about what that does and does not give you.

Anomaly detection is mature. Most platforms in this category will tell you that a metric moved more than expected, and some will point at the segment where it moved. That is genuinely useful and it is what the phrase usually means when a vendor says automated insights.

What none of the event-based tools can do is tell you why, because the why is not in their data. The event stream records that a set of accounts stopped using the import feature. The reason those accounts stopped is sitting in three support tickets about a file format, one review, and a survey comment nobody read. Joining those to the usage drop is a different kind of product to a query builder, and it is what UserInsight is being built to do.

The realistic expectation to hold, whatever you buy, is that automation gets you to the right question much faster and rarely hands you a finished answer. A tool that claims to skip the reasoning step entirely is either narrow enough that the answers are obvious or is guessing. Treat automatic themes as a shortlist to check, not a conclusion to act on.

It depends almost entirely on which meter you land on, and the same company can see very different numbers.

On events, a few hundred B2B accounts with normal instrumentation will often sit inside or just above a free tier. Amplitude publishes 2 million events a month free with unlimited seats and Mixpanel publishes 1 million. What breaks that is a tracking plan that fires on every scroll and hover, which is a decision you control.

On monthly active users the math is different and less forgiving. A few hundred accounts with ten users each is a few thousand MAU, well past Pendo's published 500 free, and every successful adoption campaign moves you further up the scale.

On sessions it is unpredictable. Heap publishes 10,000 sessions free with SSO included and Fullstory publishes 30,000, but a B2B product used daily inside working hours burns sessions quickly, and Fullstory does not publish what a session is.

Then add the account-level add-on, which is the part most budgets miss, because it is a separate line at Amplitude, Mixpanel and PostHog rather than a tier feature. Ask for that number specifically during the sales conversation, and ask on PostHog how subscribing changes billing for identified events across the whole project. Every figure here was read from a vendor page and is dated in our pricing guides, but this category changes prices often, so confirm before you commit.