Kissmetrics V2: Analytics That Sets Itself Up

A.The install was never the problem

Ask any team how long their analytics took to set up and you will get two answers. The official one is “about an hour”, which is the snippet. The real one, if you keep asking, is somewhere between six weeks and never.

The gap between those two numbers is the work nobody counts: deciding what to track, agreeing what to call it, getting a developer to instrument it, discovering three months later that half the events fire twice, and re-doing it. The tool was installed on day one. The first trustworthy funnel arrived in Q2.

Analytics products compete on the last step and lose customers on the middle three. Every vendor demos the report. Nobody demos the eight weeks of arguing about whether “signup” means the form submit or the confirmed email.

B.Why the usual fix makes it worse

The industry’s answer to this has been the warehouse: pipe everything somewhere central, model it later, and let a data team resolve the naming. It works, and it moves the bottleneck rather than removing it. Now you need the warehouse, the pipeline, the modelling layer, and a person who understands all three.

The other answer is autocapture: record everything and figure it out afterwards. That solves instrumentation and creates a second problem, because a product that captured four thousand undifferentiated click events has not told you which one is a purchase.

Both are reasonable. Both leave the same question unanswered: who decides what these events mean? Every existing answer is “you do, later”.

A.It captures the data itself

One snippet, and Kissmetrics records every event on your site, resolved to a person across devices and sessions. There is no warehouse to stand up and no connector reading somebody else’s summary.

That last clause is the one that matters and it is easy to skim past. A connector-based tool reads what another platform already aggregated, which means it inherits that platform’s definitions, its sampling and its identity model. We own the ingestion, so when we say a person did five things over three weeks on two devices, that is one record rather than a join across three vendors who each counted differently.

B.It configures itself, which is the actual bet

This is the part that is genuinely new, and the part most worth being sceptical about.

Kissmetrics reads the events that arrive, works out what kind of business you run, assigns a role to every event, and builds your first metrics, funnel and dashboard before you touch anything. Give it a URL and it scans the site first, so the inference starts from your actual pages rather than from a blank schema.

The output is not a pile of raw events. It is a product view: this is a product view, this is an add to cart, this is the purchase, and here is the funnel between them. What used to be six weeks of a tracking plan arrives as a first draft you edit.

C.Then you just ask

Ask in plain English. It writes the query, picks the fastest way to answer it, gives you the report, and saves the setup so the same question runs the same way next month.

That last clause is doing most of the work. Plenty of tools will answer a question once. The reason analytics practices decay is that the answer is not reproducible: someone asks in March, someone re-asks in June with a slightly different filter, the numbers disagree, and the meeting becomes about the numbers instead of the decision. Saving the query as a report means June and March are the same question.

A.The honest cost of configuring itself

A tracking plan written by hand has one real virtue: every line was argued over by someone who understands the business. Inference does not have that. It will occasionally decide that a newsletter signup is your primary conversion because it looks like one from the outside.

So the schema is visible and editable. After the scan, Kissmetrics shows you what it inferred: which events it found, what role it gave each one, what business model it thinks you run. Then you correct it. The bet is not that the inference is right. It is that reviewing a wrong draft takes an afternoon and writing a right one from scratch takes a quarter.

If we ever hide that schema to make the demo smoother, we will have built the thing we set out to replace.

B.What we are not

We are not a warehouse, and if your analytics problem is joining six internal systems together, we are the wrong tool. We are not a session-based web analytics product either; if you want pageviews by page, that already exists and it is free.

The whole product assumes you care about people: who they are, what they did over months, and what they were worth. That assumption is the same one Kissmetrics made in 2008, and it is why the V2 rebuild was possible at all: the identity model did not need replacing, only everything on top of it.

It is also why we can say something most analytics vendors cannot. We have six years of our own historical data showing which content produced customers at the level of the individual post, because the blog and the product shared an identity layer. We published the teardown, including the parts that do not flatter us.