tree: 02321f9cf7e5cb5b00e73022d7202d2c16b653c7 [path history] [tgz]
  1. internal/
  2. public/
  3. test/
  5. DEPS
  8. components_unittests.filter
  9. features.gni

Feature Engagement

The Feature Engagement component provides a client-side backend for displaying feature enlightenment or in-product help (IPH) with a clean and easy to use API to be consumed by the UI frontend. The backend behaves as a black box and takes input about user behavior. Whenever the frontend gives a trigger signal that in-product help could be displayed, the backend will provide an answer to whether it is appropriate to show it or not.


We often add new features, but some are hard to find. Both new and old features could benefit from being surfaced for users that we believe would be the ones who benefit the most. This has lead to the effort of providing direct in-product help to our end users that should be extremely context aware to maximize the value of the new information.

Conceptually one could implement tracking whether In-Product Help should be displayed or not through a single preference for whether it has been shown before. However, that leads to a few issues that this component tries to solve:

  • Make showing In-Product Help context aware.
    • If a user is continuously using a feature, there is no reason for Chrome to display In-Product Help for it.
    • Other events might be required to have happened first that would make it more likely that the end user would be surprised and delighted when the In-Product Help in fact does show up.
      • Example: Having seen the Chrome offline dino 10 times the last week, the user might be happier if they are informed that they can download web pages exactly as a page successfully loads.
  • Tackle interactions between different In-Product Help features.
    • If other In-Product Help has been shown within the current session, we might not want to show a different one.
    • Whether we have shown a particular In-Product Help or not might be a precondition for whether we should show different one.
  • Users should be able to use try out a feature on their own for some time before they see help.
    • We should show In-Product Help only if they don't seem use it, but we believe it would be helpful to them.
  • Share the same statistics framework across all of Chrome.
    • Sharing a framework within Chrome makes it easier to track statistics and share queries about In-Product Help in a common way.
  • Make it simpler to add new In-Product Help for developers, but still enabling them to have a complex decision tree for when to show it.


Each In-Product Help is called a feature in this documentation. Every feature will have a few important things that are tracked, particularly whether the in-product help has been displayed, whether the feature the IPH highlights has been used and whether any required preconditions have been met. All of these are tracked within daily buckets. This tracking is done only locally on the device itself.

The client-side backend is feature agnostic and has no special logic for any specific features, but instead provides a generic API and uses the Chrome Variations framework to control how often IPH should be shown for end users. It does this by setting thresholds in the experiment params and compare these numbers to the local state.

Whenever the triggering condition for possibly showing IPH happens, the frontend asks the backend whether it should display the IPH. The backend then compares the current local state with the experiment params to see if they are within the given thresholds. If they are, the frontend is informed that it should display the IPH. The backend does not display any UI.

To ensure that there are not multiple IPHs displayed at the same time, the frontend also needs to inform the backend whenever the IPH has been dismissed.

In addition, since each feature might have preconditions that must be met within the time window configured for the experiment, the frontend needs to inform the backend whenever such events happen.

To ensure that it is possible to use whether a feature has been used or not as input to the algorithm to decide whether to show IPH and for tracking purposes, the frontend needs to inform whenever the feature has been used.

Lastly, some preconditions might require something to never have happened. The first time a user has IPH available, that will typically be true, since the event was just started being tracked. Therefore, the first time the Chrome Variations experiment is made available to the user, the date is tracked so it can be used to require that the IPH must have been available for at least N days.

The backend will track all the state in-memory and flush it to disk when necessary to ensure the data is consistent across restarts of the application. The time window for how long this data is stored is configured server-side.

All of the local tracking of data will happen per Chrome user profile, but everything is configured on the server side.

Developing a new In-Product Help Feature

You need to do the following things to enable your feature, all described in detail below.

Declaring your feature

You need to create a base::Feature that represents your In-Product Help feature, that enables the whole feature to be controlled server side. The name should be on the form:

  1. kIPH prefix
  2. Your unique CamelCased name, for example MyFun.
  3. Feature suffix.

The name member of the base::Feature struct should match the constant name, and be on the form:

  1. IPH_ prefix
  2. Your unique CamelCased name, for example MyFun.

There are also a few more places where the feature should be added, so overall you would have to add it to the following places:

  • //components/feature_engagement/public/

    const base::Feature kIPHMyFunFeature{"IPH_MyFun",
  • //components/feature_engagement/public/feature_constants.h:

    extern const base::Feature kIPHMyFunFeature;
  • //components/feature_engagement/public/

    • Add to const base::Feature* kAllFeatures[].
  • //components/feature_engagement/public/feature_list.h:


If the feature will also be used from Java, also add it to: org.chromium.components.feature_engagement.FeatureConstants as a String constant.

Using the feature_engagement::Tracker

To retrieve the feature_engagement::Tracker you need to use your platform specific way for how to retrieve a KeyedService. For example for desktop platforms and Android, you can use the feature_engagement::TrackerFactory in //chrome/browser/feature_engagement/tracker_factory.h to retrieve it from the Profile or BrowserContext:

feature_engagement::Tracker* tracker =

That service can be first of all used to notify the backend about events:


In addition, it can tell you whether it is a good time to trigger the help UI:

bool trigger_help_ui =
if (trigger_help_ui) {
    // Show IPH UI.

If feature_engagement::Tracker::ShouldTriggerHelpUI return true you must display the In-Product Help, as it will be tracked as if you showed it. In addition you are required to inform when the feature has been dismissed:


Inspecting whether IPH has already been triggered for a feature

Sometimes additional tracking is required to figure out if in-product help for a particular feature should be shown, and sometimes this is costly. If the in-product help has already been shown for that feature, it might not be necessary any more to do the additional tracking of state.

To check if the triggering condition has already been fulfilled (i.e. can not currently be triggered again), you can call:

Tracker::TriggerState trigger_state =

Inspecting this state requires the Tracker to already have been initialized, else NOT_READY is always returned. See IsInitialized() and AddOnInitializedCallback(...) for how to ensure the call to this is delayed.

A note about TriggerState naming

Typically, the FeatureConfig (see below) for any particular in-product help requires the configuration for event_trigger to have a comparator value of ==0, i.e. that it is a requirement that the particular in-product help has never been shown within the search window. The values of the TriggerState enum reflects this typical usage, whereas technically, this is the correct interpretation of the states:

  • HAS_BEEN_DISPLAYED: event_trigger condition is NOT met and in-product help will not be displayed if Tracker is asked.
  • HAS_NOT_BEEN_DISPLAYED: event_trigger condition is met and in-product help might be displayed if Tracker is asked.
  • NOT_READY: Tracker not fully initialized yet, so it is unable to inspect the state.

Inspecting whether IPH would have been triggered for a feature

Another way to check the internal state of the Tracker is to invoke feature_engagement::Tracker::WouldTriggerHelpUI which is basically the same as invoking feature_engagement::Tracker::ShouldTriggerHelpUI, but being allowed to ignore the state. It is still required to invoke feature_engagement::Tracker::ShouldTriggerHelpUI if in-product help should be shown.

WARNING: It is not guaranteed that invoking ShouldTriggerHelpUI(...) after this would yield the same result. The state might change in-between the calls because time has passed, other events might have been triggered, and other state might have changed.

Configuring UMA

To enable UMA tracking, you need to make the following changes to the metrics configuration:

  1. Add feature to the histogram suffix IPHFeatures in: //tools/metrics/histograms/histograms.xml.
    • The suffix must match the base::Feature name member of your feature.
  2. Add feature to the actions file at: //tools/metrics/actions/actions.xml.
    • The suffix must match the base::Feature name member.
    • Find the <action-suffix> entry at the end of the file, where the following <affected-action>s are listed:
      • InProductHelp.NotifyEvent.IPH
      • InProductHelp.NotifyUsedEvent.IPH
      • InProductHelp.ShouldTriggerHelpUI.IPH
      • InProductHelp.ShouldTriggerHelpUIResult.NotTriggered.IPH
      • InProductHelp.ShouldTriggerHelpUIResult.Triggered.IPH
      • InProductHelp.ShouldTriggerHelpUIResult.WouldHaveTriggered.IPH
    • Add an alphebetically sorted entry to the list of <suffix>es like: <suffix name="MyFunFeature" label="For MyFunFeature feature."/>

Adding a local field trial testing configuration

For each in-product help feature, it is required to also configure the expected launch configuration as the main testing configuration. See Field Trial Testing Configuration for details.

Basically this requires you to add a new section to //testing/variations/fieldtrial_testing_config.json for your feature. The format is described in the documentation linked above, but it will probably look something like this:

  "MyFunFeatureStudy": [
      "platforms": ["android"],
      "experiments": [
          "name": "MyFunFeatureLaunchConfig",
          "params": {
            "availability": ">=30",
            "session_rate": "<1",
            "event_used": "name:fun_event_happened;comparator:any;window:360;storage:360",
            "event_trigger": "name:fun_feature_iph_triggered;comparator:any;window:360;storage:360",
            "event_1": "name:related_fun_thing_happened;comparator:>=1;window:360;storage:360"
          "enable_features": ["IPH_MyFunFeature"],
          "disable_features": []

Demo mode

The feature_engagement::Tracker supports a special demo mode, which enables a developer or testers to see how the UI looks like without using Chrome Variations configuration.

The demo mode behaves differently than the code used in production where the chrome Variations configuration is used. Instead, it has only a few rules:

  • Event model must be ready (happens early).
  • No other features must be showing at the moment.
  • The given feature must not have been shown before in the current session.

This basically leads to each selected IPH feature to be displayed once. The triggering condition code path must of course be triggered to display the IPH.

How to select a feature or features is described below.

Enabling all In-Product Help features in demo-mode

  1. Go to chrome://flags
  2. Find “In-Product Help Demo Mode” (#in-product-help-demo-mode-choice)
  3. Select “Enabled”
  4. Restart Chrome

Enabling a single In-Product Help feature in demo-mode

  1. Go to chrome://flags
  2. Find “In-Product Help Demo Mode” (#enable-iph-demo-choice)
  3. Select the feature you want with the "Enabled " prefix, for example for IPH_MyFunFeature you would select:
    • Enabled IPH_MyFunFeature
  4. Restart Chrome

Using Chrome Variations

Each In-Product Help feature must have its own feature configuration FeatureConfig, which has 4 required configuration items that must be set, and then there can be an arbitrary number of additional preconditions (but typically on the order of 0-5).

The data types are listed below.



  "availability": "{Comparator}",
  "session_rate": "{Comparator}",
  "session_rate_impact": "{SessionRateImpact}",
  "event_used": "{EventConfig}",
  "event_trigger": "{EventConfig}",
  "event_???": "{EventConfig}",
  "tracking_only": "{Boolean}"
  "x_???": "..."

The FeatureConfig fields availability, session_rate, event_used and event_trigger are required, and there can be an arbitrary amount of other event_??? entries.

  • availability REQUIRED
    • For how long must an in-product help experiment have been available to the end user.
    • The value of the Comparator is in a number of days.
    • See Comparator below for details.
  • session_rate REQUIRED
    • How many other in-product help have been displayed within the current end user session.
    • The value of the Comparator is a count of total In-Product Help displayed in the current end user session.
    • See Comparator below for details.
  • session_rate_impact
    • Which other in-product help features showing the current IPH impacts.
    • By default, a feature impacts every other feature.
    • Defaults to all.
    • See SessionRateImpact below for details.
  • event_used REQUIRED
    • Relates to what the in-product help wants to highlight, i.e. teach the user about and increase usage of.
    • This is typically the action that the In-Product Help should stimulate usage of.
    • Special UMA is tracked for this.
    • See EventConfig below for details.
  • event_trigger REQUIRED
    • Relates to the times in-product help is triggered.
    • Special UMA is tracked for this.
    • See EventConfig below for details.
  • event_???
    • Similar to the other event_ items, but for all other preconditions that must have been met.
    • Name must match /^event_[a-zA-Z0-9-_]+$/ and not be event_used or event_trigger.
    • See EventConfig below for details.
  • tracking_only
    • Set to true if in-product help should never trigger.
    • Tracker::ShouldTriggerHelpUI(...) will always return false, but if all other conditions are met, it will still be recorded as having been shown in the internal database and through UMA.
    • This is meant to be used by either local tests or for comparisons between different experiment groups.
    • If you want to later transition users with this flag set to true to in fact display in-product help, you might want to use a different EventConfig::name for the event_trigger configuration than the non-tracking configuration.
    • Defaults to false.
    • See Boolean below for details.
  • x_???
    • Any parameter starting with x_ is ignored by the feature engagement tracker.

    • A typical use case for this would be if there are multiple experiments for the same in-product help, and you want to specify different strings to use in each of them, such as:

      "x_promo_string": "IDS_MYFUN_PROMO_2"
    • Failing to use an x_-prefix for parameters unrelated to the FeatureConfig will end up being recorded as FAILURE_UNKNOWN_KEY in the InProductHelp.Config.ParsingEvent histogram.


  "availability": ">=30",
  "session_rate": "<1",
  "event_used": "name:download_home_opened;comparator:any;window:90;storage:360",
  "event_trigger": "name:download_home_iph_trigger;comparator:any;window:90;storage:360",
  "event_1": "name:download_completed;comparator:>=1;window:120;storage:180"


Format: name:{std::string};comparator:{COMPARATOR};window:{uint32_t};storage:{uint32_t}

The EventConfig is a semi-colon separate data structure with 4 key-value pairs, all described below:

  • name
    • The name (unique identifier) of the event.
    • Must match what is used in client side code.
    • Must only contain alphanumeric, dash and underscore.
      • Specifically must match this regex: /^[a-zA-Z0-9-_]+$/
    • Value client side data type: std::string
  • comparator
    • The comparator for the event. See Comparator below.
  • window
    • Search for this occurrences of the event within this window.
    • The value must be given as a number of days.
    • For value N, the following holds:
      • 0 Nothing should be counted.
      • 1 |current_day| should be counted.
      • 2+ |current_day| plus |N-1| more days should be counted.
    • Value client side data type: uint32_t
  • storage
    • Store client side data related to events for this event minimum this long.
    • The value must be given as a number of days.
    • For value N, the following holds:
      • 0 Nothing should be stored.
      • 1 |current_day| should be stored.
      • 2+ |current_day| plus |N-1| more days should be stored.
    • The value should not exceed 10 years (3650 days).
    • Value client side data type: uint32_t
    • Whenever a particular event is used by multiple features, the maximum value of all storage is used as the storage window.




Format: {COMPARATOR}[value]

The following comparators are allowed:

  • < less than
  • > greater than
  • <= less than or equal
  • >= greater than or equal
  • == equal
  • != not equal
  • any always true (no value allowed)

Other than any, all comparators require a value.




Format: [true|false]

The following values are allowed:

  • true
  • false

The value must be quoted (like all the other values).




Format: [all|none|comma-separated list]

  • all means this feature impacts every other feature regarding their session_rate calculations. This is the default.
  • none means that this feature does not impact any other features regarding the session_rate. This feature may therefore be shown an unlimited amount of times, without making other features go over their session_rate config.
  • [comma-separated list] means that this feature only impacts the particular features listed. Use the base::Feature name of the feature in the list. For features in the list, this feature will affect their session_rate conditions, and for features not in the list, this feature will not affect their session_rate calculations.
    • It is NOT valid to use the feature names all or none. They must only be used alone with no comma, at which point they work as described above.



Using Chrome Variations at runtime

It is possible to test the whole backend from parsing the configuration, to ensuring that help triggers at the correct time. To do that you need to provide a JSON configuration file, that is then parsed to become command line arguments for Chrome, and after that you can start Chrome and verify that it behaves correctly.

  1. Create a file which describes the configuration you are planning on testing with, and store it. In the following example, store the file DownloadStudy.json:

      "DownloadStudy": [
          "platforms": ["android"],
          "experiments": [
              "name": "DownloadExperiment",
              "params": {
                "availability": ">=30",
                "session_rate": "<1",
                "event_used": "name:download_home_opened;comparator:any;window:90;storage:360",
                "event_trigger": "name:download_home_iph_trigger;comparator:any;window:90;storage:360",
                "event_1": "name:download_completed;comparator:>=1;window:120;storage:180"
              "enable_features": ["IPH_DownloadHome"],
              "disable_features": []
  2. Use the field trial utility to convert the JSON configuration to command line arguments:

    python ./tools/variations/ DownloadStudy.json android shell_cmd
  3. Pass the command line along to the binary you are planning on running or the command line utility for the Android platform.

    For the target chrome_public_apk it would be:

    ./build/android/adb_chrome_public_command_line "--force-fieldtrials=DownloadStudy/DownloadExperiment" "--force-fieldtrial-params=DownloadStudy.DownloadExperiment:availability/>=30/event_1/name%3Adownload_completed;comparator%3A>=1;window%3A120;storage%3A180/event_trigger/name%3Adownload_home_iph_trigger;comparator%3Aany;window%3A90;storage%3A360/event_used/name%3Adownload_home_opened;comparator%3Aany;window%3A90;storage%3A360/session_rate/<1" "--enable-features=IPH_DownloadHome<DownloadStudy"

Printf debugging

Several parts of the feature engagement tracker has some debug logging available. To see if the current checked in code covers your needs, try starting a debug build of chrome with the following command line arguments:


Development of //components/feature_engagement


To compile and run tests, assuming the product out directory is out/Debug, use:

ninja -C out/Debug components_unittests ;
./out/Debug/components_unittests \

When adding new test suites, also remember to add the suite to the filter file: //components/feature_engagement/components_unittests.filter.