This article will help you:
When working with product analytics, understanding why something is happening is arguably more important than understanding what is happening in the first place. This is especially true when Amplitude is showing anomalous data—i.e., events and properties that are out of the ordinary, and to a significant extent. With anomalous data, you need to be able to determine if what you're seeing is just a random blip, or the beginning of a shift in the way your users interact with your product.
Historically, that insight has not always been easy to come by. For example, it would take a decent amount of guesswork to navigate down the hierarchy from Platform → OS → Device family to finally discover that an observed change is driven by users on a specific type of device.
Amplitude's Root Cause Analysis (RCA) feature addresses this by analyzing the properties of the anomalous events for you, while also pulling in external context like country-specific holidays and new releases of your product. In this way, it can potentially explain the anomaly or rule out the obvious. It's designed to streamline your workflow and help you quickly understand the “why?” of a change, so that you can easily answer questions like “Which user groups best explain this change?” or “How are other correlated metrics affected?”
NOTE: This feature is currently in beta testing for customers on Enterprise plans, and those with the Insights add-on.
Analyze an anomalous data point
In order to use RCA, you must first have an analysis in Amplitude that's displaying anomalous data.
For example, this Event Segmentation chart has an unusual-looking spike for November 20th:
To examine this anomalous data point using Root Cause Analysis, follow these steps:
- Click Anomaly + Forecast to confirm that the result you're interested in is actually a statistical anomaly. Amplitude will enhance the chart to display the statistically-expected values for each day, as well as a range of values that would not be considered anomalous (in other words, values that could be attributable to random chance).
In this example, the November 20th data point is outside the range of statistically "normal" values.
- Click on the data point to bring up the Microscope. Then click Investigate Anomaly. Amplitude will automatically navigate to the Root Cause Analysis tab.
Amplitude will scan the properties of the anomalous event and compare them to a baseline date. You can find the baseline date Amplitude is using in the Anomaly Details panel, along with a list of any holidays occurring in the date range covered by the analysis (listed under the Holidays tab), and descriptions of recent product changes that may have affected the anomalous data (the Annotations tab).
After each scan, Amplitude will place each property into one of three groups, based on how much of the anomaly Amplitude believes it can explain: High significance, moderate significance, and low significance. In this example, Amplitude has already identified '[Amplitude] City' as a property that may be able to explain what's happening here.
You can get more details on these events by switching the Show supporting metrics toggle to the right:
In this example, it appears the anomaly is being driven by users in the cities of McAllen and Redwood City—Amplitude received 720 and 115 events with those values in the 'city' property, respectively, against a baseline of none.
You can also view this analysis as a time-series chart, showing fluctuations in the prevalence of each value for this event property over time:
Configure your analysis
Note, however, that this analysis is based on only fifteen event properties. RCA scans event properties in batches of fifteen, in order to present you with results for the most relevant properties first.
To scan more properties:
- Click Configure Analysis.
- In the modal that appears, click Continue analyzing.
If you'd like a specific event property to be included in the next scan, click Select property... and select the property from the list.