# Insights (/docs/product-analytics/insights)

An insight is a saved chart. Start one from **Insights → New insight**, then pick the type. The chart updates as you edit.

| Type | Answers |
| --- | --- |
| [Trend](#trends) | How much, over time? |
| [Funnel](#funnels) | How many get from A to B? |
| [Retention](#retention) | Do people come back? |
| [Paths](#paths) | What do people do before or after something? |
| [Lifecycle](#lifecycle) | Is growth coming from new people, or are we losing the ones we have? |
| [Stickiness](#stickiness) | How much of a habit is it? |

## Trends [#trends]

A trend plots one or more **series** over time. Each lettered row (A, B, C…) is a series: an event (or **All events**) and a measure.

| Measure | Counts | Example |
| --- | --- | --- |
| **Total count** | Events | Pageviews per day |
| **Unique people** | People. Someone using two devices counts once. | Daily active users |
| **Unique sessions** | Sessions | Sessions per week |
| **Unique groups** | Companies, workspaces… | Weekly active companies |
| **Sum / Average / Min / Max** of a property | A number on the event | Revenue: sum of `amount` on `order_completed` |
| **Median / 90th / 99th percentile** | A number on the event | Typical basket size: median of `amount` |

Above the chart:

- **Date range**: from the last 24 hours to all time.
- **Interval**: hour, day, week (weeks start on Monday) or month.
- **Compare → Previous period** draws the period before as a muted line and shows the change.
- **Display**: line, area, bar, a big number, or a table.

The total for a series covers the whole date range. For *unique people*, that's distinct people over the whole range, not the sum of the daily numbers.

<Callout type="info" title="Hatched means still collecting">
  When the range ends now, the last day (or week) isn't complete. Its part of the line is dashed over a hatched band, and the summary leaves it out, so a half-finished day never looks like a drop.
</Callout>

## Funnels [#funnels]

A funnel counts how many people complete a series of steps in order, within a time limit:

- **SaaS activation**: `signed_up` → `project_created` → `invite_sent`
- **Checkout**: `product_viewed` → `product_added` → `checkout_started` → `order_completed`

Add 2–12 **Steps**. Under **Options**:

- **Convert within … days**: how long people have from the first step to finish.
- **In order** lets other events happen between steps. **Strict** ends the attempt if a later funnel step happens too early.
- **Count people** or **Count companies** (or another group type). Counting companies moves an account forward when *any* teammate does the next step, which is what you want for B2B activation.

The chart shows each step with the lost share hatched, and the table lists counts, conversion and drop-off. Each person counts once, using their furthest attempt.

<Callout type="info" title="Leave room at the end of the range">
  Every step must fall inside the date range. Someone who signs up on the last day hasn't had time to convert yet. For final numbers, end the range one conversion window ago.
</Callout>

## Retention [#retention]

Retention answers *of the people who started in a given week, how many came back?* Pick two events:

- the **cohortizing event** that marks the start, such as `signed_up` or the first `order_completed`;
- the **returning event** that counts as coming back, such as `project_opened` or `order_completed`.

Under **Periods**, choose day, week or month and how many columns you want.

Each row is a group of people who started in the same period. Column *N* is the share who did the returning event *N* periods later. Some things to know:

- A person lands in the row of their **first** start event within the date range. For "new users" retention, use an event that happens once, like `signed_up`.
- The **average** row weighs big cohorts more, and leaves out cohorts that haven't reached a column yet. If one cohort is much bigger (a launch or a campaign), look at the rows as well.
- Hatched cells haven't happened yet.

A curve that flattens out instead of dropping to zero means you have a core of retained users.

## Paths [#paths]

The most common routes people take through your pages or events, drawn as a flow. Each column is a step, and the width of a band is how many sessions (or people) went that way.

- **Steps are** pages, custom events, or both.
- **Starting at** shows what happens after a page or event, such as everything people do after `/pricing`.
- **Ending at** shows what leads up to it, such as the three steps before `signed_up`.
- Choose up to 8 steps, and one path per **session** or per **person**.

Reloads are merged, and the most common 7 destinations per step are shown. The table under the chart lists the top transitions.

## Lifecycle [#lifecycle]

Pick what counts as being active, such as `board_opened`. Each day, week or month, the people who did it are split into:

| | Means |
| --- | --- |
| **New** | Did it for the first time ever |
| **Returning** | Also did it the period before |
| **Resurrecting** | Did it before, but not the period before: came back after a gap |
| **Dormant** | Did it the period before, but not this one (drawn below the line) |

If new and resurrecting bars outweigh dormant, the active group is growing. The current period has no dormant bar until it ends.

## Stickiness [#stickiness]

How many days (or weeks) in the date range each person did something. With the last 30 days, the bar at **1** is people who did it once and the bar at **20** is people who did it on 20 different days. Bars leaning right mean a habit. Compare up to five events, such as two features.

## Filters and breakdowns [#filters-and-breakdowns]

**Filters** narrow a series, a step or the whole chart. You can filter on:

- an **event property**, such as `utm_source = newsletter`;
- a **person property**, such as `plan is one of pro, business`;
- a **group property**, such as the company's `industry`;
- a **cohort**: in, or not in, a [cohort](/docs/people/cohorts).

**Breakdown** splits the chart into one line (or one funnel) per value: per browser, per plan, per country, or per cohort to compare *power users* with *new users*. Only the most common values are shown (10 by default). Empty values show as **(none)**. In funnels, the value comes from the first step, so someone who signed up on mobile counts as mobile even if they bought on desktop. Segments with fewer than 30 people are stippled: treat them as anecdotes.

Retention has no breakdown. To compare groups, save one retention insight per cohort.

## Save and reuse [#save-and-reuse]

Give the insight a name, or leave it empty and Datahop names it from the query. Click **Save insight**, or **Add to dashboard** to save it and place it on a dashboard in one step.

- Add a **description** to say what's included ("excludes test accounts"). Teammates and agents search descriptions before building a new chart.
- Saved insights are shared with the whole project. The list shows who made each one, including agents.
- Under the chart, **Summary · what agents see** shows the plain-text summary an agent gets for the same chart.
- Results update when you open the insight. Numbers can be up to a minute old.