Run advanced analysis in Metric Explorer

Advanced analysis in Metric Explorer changes how a metric's values accumulate over time, compares the date range with earlier periods, and flags values outside the expected range. Cumulative and rolling measurements apply to one event metric at a time, so you can smooth a noisy series while the other series on the chart stay raw.

Add a cumulative or rolling measurement to a metric

1. On an event metric, open the ⋮ menu and choose Advanced.
2. Choose a measurement:
   • Cumulative: Shows a running total from the start of the date range. With Unique Users, each point counts distinct users from the start of the range through that point, so a user who returns counts once.
   • Rolling Average: Shows the unweighted mean of the last N intervals at each point. Use Rolling Average to smooth weekly cycles, such as weekday peaks and weekend troughs.
   • Rolling Window: Aggregates the last N intervals into each point. Use Rolling Window for metrics such as 7-day active users on a daily chart.
3. Set the number of intervals, for example 7 days.

The metric shows the measurement next to its name, for example "rolling average," and the other metrics on the chart don't change.

A rolling measurement reads data from before the start of the date range. With a 7-day rolling average on a February chart, the point for February 6 averages January 31 through February 6.

The Advanced options depend on the metric's measurement. For example, Revenue offers all three, and Average of Property Value offers Rolling Window. When you switch a metric to a measurement that doesn't support its advanced measurement, Amplitude removes the advanced measurement from that metric.

Compare to a past period

Period comparison adds a series for an earlier period next to each metric, so you can read this period's change against the last one.

1. Click Compare above the chart and choose Previous time periods.
2. Choose a preset, such as Previous week or Previous year, or set a custom period that starts or ends on a date you pick.
3. To compare against two earlier periods at once, click Add Comparison.

The comparison series appears in the legend as [Previous 1], and the takeaways above the breakdown table show the current and previous values side by side. For preset definitions, custom periods, and the Rolling option, go to Period-over-period comparison.

Detect anomalies and forecast

Anomaly + Forecast draws an expected range around each metric and marks values outside it. Anomaly + Forecast can also project a metric forward.

1. Open Chart Settings and select Anomaly & Forecast.
2. Enable anomaly detection and choose a mode:
   • Agile: Adjusts quickly to recent trends, with a 95% confidence interval and 120 days of training data.
   • Robust: Uses a full extra year of training data and accounts for seasonality. Use Robust for stable metrics.
   • Custom: Set the confidence interval and training duration yourself.
3. To project the metric, enable the forecast and set how far ahead to forecast.

The chart shows a confidence band and a dashed Expected Value series, and the breakdown table adds the expected, low, and high values for each point. For how Amplitude trains the model and interprets results, go to Anomaly + Forecast.

Mark events on the chart with annotations

Annotations mark dates such as releases or campaigns on the time axis. To add one, click + below the chart's X-axis, then set the date, description, and visibility. To choose which annotation categories the chart shows, open Chart Settings and select Annotations. Annotations appear on every Metric Explorer, Event Segmentation, and User Sessions chart in the project. For permissions and date ranges, go to Releases and annotations.

Use real-time and excluded-day intervals

• Realtime: Set the interval to Realtime to see the latest events in five-minute buckets.
• Excluded days: In the date picker, click Exclude and choose days of the week, such as Saturday and Sunday. Amplitude drops those days from the chart, the breakdown table, row calculations, and CSV exports.

Speed up queries with sampling

On large projects, query time sampling runs the chart on a random 10% of users and scales the results up to the full population. To turn it on, open Chart Settings, select Query Time Sampling, and click Enable Sampling. The chart shows "Sampling 10%" while sampling is on. Sampled values are estimates. Turn sampling off before you report exact counts. For which measurements scale and which features sampling turns off, go to Optimize performance with query time sampling.

Query by server upload time

By default, Amplitude places each event at the time it happened on the user's device. To place events at the time Amplitude received them, open Chart Settings, select Query by Server Upload Time, and enable it. The chart shows "Querying by Server Upload Time." Use this view to check ingestion, such as events that arrive late from offline mobile devices.