# Introducing Product Analytics

Last updated: 2026-09-18
Canonical source: https://confidence.spotify.com/blog/product-analytics
Owner: Spotify AB
Authors: Guillaume Perchais (Senior Product Manager)

> If this file and the page at the canonical URL disagree, the page is authoritative.

Product Analytics in Confidence lets you ask questions of your data in plain language and get answers you can explore with more depth and flexibility than a dashboard.

Having easy access to data that reveals how a product is used is the product builder's dream. It's key to understanding where your users are succeeding, where they hit blockers, and what they can't do that they should be able to. That knowledge shapes what teams choose to build.

Data has always been core to how we design and build at Spotify. We've written a lot about our investments in [experimentation and learning](https://engineering.atspotify.com/2025/9/spotifys-experiments-with-learning-framework), and one of the ways this shows up in our product org is the time we put into dashboards. (In 2023 alone, Spotifiers created [more than 4,900 dashboards](https://engineering.atspotify.com/2024/08/unlocking-insights-with-high-quality-dashboards-at-scale).)

Dashboards are great for answering recurring questions, but they cannot anticipate every line of inquiry. For example, if you're trying to investigate why monthly active users dropped in a certain market and your dashboard doesn't let you break down the country you're looking at by device, plan, or age, you don't have the data you need to make an informed decision.

AI is also obviating the need to build and maintain dashboards. People have quickly grown accustomed to the flow of asking a question and getting an answer in plain language instead of having to manipulate data or dig through charts.

So we decided to build our own data agent: you ask a question in plain text and the agent finds the relevant data, writes the query, and returns the answer (with its query and sources). As soon as you have a hypothesis, you can validate it without having to build a new dashboard. The agent pulls from all the data it has access to and if the data it needs to answer confidently doesn't exist, it tells you so you can start gathering it.

Since we launched it internally, our data agent has helped Spotify's product builders create and test hypotheses faster and with stronger evidence. Now we're making this capability available in [Confidence](https://confidence.spotify.com/), Spotify's product intelligence platform.

## Introducing Product Analytics in Confidence

Analytics is at the heart of our product practice, and it's always been a goal of ours to bring it into Confidence. The data agent we built for internal use is now the core of our new Product Analytics suite that lets you do what Spotifiers do: ask a question in plain language and get an answer you can explore with more depth and flexibility than a dashboard.

For example, let's ask: "How did monthly active users change in France last month?"

Confidence charts the evolution and lets you break it down by age group, compare the affected segment with the current rollout, and review recent experiment results in the same conversation.

You can ask in Confidence, or you can stay in Slack, if that's easier for your workflow. Here's what the answer looks like depending on where you start from:

*Figure: Slack conversation with the Confidence agent showing a bar chart of monthly orders over the past six months*

*Figure: Confidence chat interface showing a stacked bar chart of daily transactions by product category over ninety days*

*Figure: Sankey diagram showing a user funnel from 1,137 users through payment stages to 544 returning to the home page*

## Exploration: Strengthen analytics with built-in product context

Being able to ask questions of your data is useful on its own, but asking questions within Confidence adds valuable context because analytics and delivery live in the same platform. Product Analytics reads your warehouse directly alongside Confidence's knowledge of your product strategy, experiments, and rollout.

Movement within a metric means something different during a launch, halfway through a rollout, or after an experiment. A standalone insights tool that queries your warehouse directly has no idea what you shipped last Tuesday. It shows you only what happened within the product, leaving you with dots you still have to connect across separate tools.

Confidence connects data signals with your broader product strategy and release timeline. For example, you can see how a drop in a metric relates to a specific launch, or dig into why an experiment shows changes only in a specific segment. And the evidence for why you made a change (the A/B tests, the results of monitoring the rollout), is tied to the decision, ready to shape the next turn of your [product loop](/blog/the-product-loop).

In practice, it looks like this: Activation drops eight percent in Germany. You ask why. Confidence charts the drop and flags that a rollout reached fifty percent of German users the same day. It pulls the experiment readout, which shows that the treatment arm was flat.

In one conversation, with no context switching or manually assembling data from different tools, no having to rely on someone to remember what you shipped last week, you know what happened and have a data-driven hypothesis for how to fix it. That's the value of Product Analytics in Confidence.

*Figure: Confidence exploration showing a transaction drop investigation with a country breakdown chart highlighting Germany as the outlier*

*Figure: Continuation of the exploration linking the transaction decline to a Germany activation-page rollout that started the same day*

## Pulse: Keep important questions alive

Exploring one-off questions is only one aspect of Product Analytics. Some questions need to be answered every day or week, so we designed Confidence's **Pulse**, which turns a useful result into a recurring conversation about a metric that matters.

Think of a Pulse as a regular health check for your product. Pulses keep your vital signs, like monthly active users, activation, or rollout health, close at hand, in a form you can explore. Each time the Pulse refreshes, Confidence sends it to your team in Slack. You can question it there: reply in the thread to ask for a deeper breakdown, compare the metric with a rollout or experiment.

A scheduled report tells you something moved, but then it stops talking. A Pulse, on the other hand, kicks off another revolution of your product loop.

*Figure: Confidence Insights page showing a Pulse dashboard with charts for daily unique users, stock price, sales distribution, and customer segmentation*

*Figure: Slack notifications dialog for configuring scheduled Pulse report delivery to channels and direct messages*

## Good answers start with the right tables

There's such a thing as too much data: connecting every available table makes it harder for an agent to choose the right source.

We learned this lesson when we first built our internal product agent: Spotify has more than 70,000 datasets, and we found that ad hoc tables, debugging queries, and one-off analyses often taught the agent the wrong patterns. So we built in the ability for domain experts to choose the [fact tables](https://confidence.spotify.com/docs/metrics/fact-tables) Product Analytics can use.

You can start with product events, business measurements, and useful aggregations with clear definitions, skipping modeling tables and unfinished analysis. You can also explain how to handle the data, telling an agent, for example, to use the latest partition for current-state questions, follow the company's definition of an active user, or avoid forecasting or domains that are not ready to share.

*Figure: Data agent admin page showing a list of fact tables with toggles to control which tables the agent can query*

*Figure: Guardrails document editor where domain experts write rules that steer how the agent interprets and queries data*

## Expert human judgment with better evidence and faster iteration

Product Analytics in Confidence helps builders explore hypotheses and decide where to look next. But your hands are still on the wheel.

A chart can reveal a change, but it cannot establish the cause, fix the source data, or resolve an ambiguous metric. Every consequential finding still needs expert reviews, and often, you still need to experiment to understand the true cause of a metric movement.

Product Analytics lets you do it all faster, with more comprehensive data at hand. It's currently in beta for selected Confidence accounts. If you want to test it, reach out to us. We'll help you choose the fact tables and context the agent needs.