Category: Practices

We have decided to start running A/B Testing in Firefox for Android. These experiments are intended to optimize specific outcomes, as well as, inform our long-term design decisions. We want to create the best Firefox experience we can, and these experiments will help.

The system will also allow us to throttle the release of features, called staged rollout or feature toggles, so we can monitor new features in a controlled manner across a large user base and a fragmented device ecosystem. If we need to rollback a feature for some reason, we’d have the ability to do that, quickly without needing people to update software.

Technical details:

• Mozilla Switchboard is used to control experiment segmenting and staged rollout.
• UI Telemetry is used to collect metrics about an experiment.
• Unified Telemetry is used to track active experiments so we can correlate to application usage.

What is Mozilla Switchboard?

Mozilla Switchboard is based on Switchboard, an open source SDK for doing A/B testing and staged rollouts from the folks at KeepSafe. It connects to a server component, which maintains a list of active experiments.

The SDK does create a UUID, which is stored on the device. The UUID is sent to the server, which uses it to “bucket” the client, but the UUID is never stored on the server. In fact, the server does not store any data. The server we are using was ported to Node from PHP and is being hosted by Mozilla.

We decided to start using Switchboard because it’s simple, open source, has client code for Android and iOS, saves no data on the server and can be hosted by Mozilla.

Planning Experiments

The Mobile Product and UX teams are the primary drivers for creating experiments, but as is common on the Mobile team, ideas can come from anywhere. We have been working with the Mozilla Growth team, getting a better understanding of how to design the experiments and analyze the metrics. UX researchers also have input into the experiments.

Once Product and UX complete the experiment design, Development would land code in Firefox to implement the desired variations of the experiment. Development would also land code in the Switchboard server to control the configuration of the experiment: On what channels is it active? How are the variations distributed across the user population?

Since we use Telemetry to collect metrics on the experiments, the Beta channel is likely our best time period to run experiments. Telemetry is on by default on Nightly, Aurora and Beta; and Beta is the largest user base of those three channels.

Once we decide which variation of the experiment is the “winner”, we’ll change the Switchboard server configuration for the experiment so that 100% of the user base will flow through the winning variation.

Yes, a small percentage of the Release channel has Telemetry enabled, but it might be too small to be useful for experimentation. Time will tell.

What’s Happening Now?

We are trying to be very transparent about active experiments and staged rollouts. We have a few active experiments right now.

• Onboarding A/B experiment with several variants.
• Easy entry points for accessing History and Bookmarks on the main menu.
• Experimenting with the awesomescreen behavior when displaying search results page.

You can always look at the Mozilla Switchboard configuration to see what’s happening. Over time, we’ll be adding support to Firefox for iOS as well.

Building and shipping a successful product takes more than raw engineering. I have been posting a bit about using Telemetry to learn about how people interact with your application so you can optimize use cases. There are other types of data you should consider too. Being aware of these metrics can help provide a better focus for your work and, hopefully, have a bigger impact on the success of your product.

Active Users

This includes daily active users (DAUs) and monthly active users (MAUs). How many people are actively using the product within a time-span? At Mozilla, we’ve been using these for a long time. From what I’ve read, these metrics seem less important when compared to some of the other metrics, but they do provide a somewhat easy to measure indicator of activity.

These metrics don’t give a good indication of how much people use the product though. I have seen a variation metric called DAU/MAU (daily divided by monthly) and gives something like retention or engagement. DAU/MAU rates of 50% are seen as very good.

Engagement

This metric focuses on how much people really use the product, typically tracking the duration of session length or time spent using the application. The amount of time people spend in the product is an indication of stickiness. Engagement can also help increase retention. Mozilla collects data on session length now, but we need to start associating metrics like this with some of our experiments to see if certain features improve stickiness and keep people using the application.

We look for differences across various facets like locales and releases, and hopefully soon, across A/B experiments.

Retention / Churn

Based on what I’ve seen, this is the most important category of metrics. There are variations in how these metrics can be defined, but they cover the same goal: Keep users coming back to use your product. Again, looking across facets, like locales, can provide deeper insight.

Rolling Retention: % of new users return in the next day, week, month
Fixed Retention: % of this week’s new users still engaged with the product over successive weeks.
Churn: % of users who leave divided by the number of total users

Most analysis tools, like iTunes Connect and Google Analytics, use Fixed Retention. Mozilla uses Fixed Retention with our internal tools.

I found some nominal guidance (grain of salt required):
1-week churn: 80% bad, 40% good, 20% phenomenal
1-week retention: 25% baseline, 45% good, 65% great

Cost per Install (CPI)

I have also seen this called Customer Acquisition Cost (CAC), but it’s basically the cost (mostly marketing or pay-to-play pre-installs) of getting a person to install a product. I have seen this in two forms: blended – where ‘installs’ are both organic and from campaigns, and paid – where ‘installs’ are only those that come from campaigns. It seems like paid CPI is the better metric.

Lower CPI is better and Mozilla has been using Adjust with various ad networks and marketing campaigns to figure out the right channel and the right messaging to get Firefox the most installs for the lowest cost.

Lifetime Value (LTV)

I’ve seen this defined as the total value of a customer over the life of that customer’s relationship with the company. It helps determine the long-term value of the customer and can help provide a target for reasonable CPI. It’s weird thinking of “customers” and “value” when talking about people who use Firefox, but we do spend money developing and marketing Firefox. We also get revenue, maybe indirectly, from those people.

LTV works hand-in-hand with churn, since the length of the relationship is inversely proportional to the churn. The longer we keep a person using Firefox, the higher the LTV. If CPI is higher than LTV, we are losing money on user acquisition efforts.

Total Addressable Market (TAM)

We use this metric to describe the size of a potential opportunity. Obviously, the bigger the TAM, the better. For example, we feel the TAM (People with kids that use Android tablets) for Family Friendly Browsing is large enough to justify doing the work to ship the feature.

Net Promoter Score (NPS)

We have seen this come up in some surveys and user research. It’s suppose to show how satisfied your customers are with your product. This metric has it’s detractors though. Many people consider it a poor value, but it’s still used quiet a lot.

NPS can be as low as -100 (everybody is a detractor) or as high as +100 (everybody is a promoter). An NPS that is positive (higher than zero) is felt to be good, and an NPS of +50 is excellent.

Go Forth!

If you don’t track any of these metrics for your applications, you should. There are a lot of off-the-shelf tools to help get you started. Level-up your engineering game and make a bigger impact on the success of your application at the same time.

Firefox 31 for Android is the first release where we collect telemetry data on user interactions. We created a simple “event” and “session” system, built on top of the current telemetry system that has been shipping in Firefox for many releases. The existing telemetry system is focused more on the platform features and tracking how various components are behaving in the wild. The new system is really focused on how people are interacting with the application itself.

Collecting Data

The basic system consists of two types of telemetry probes:

• Events: A telemetry probe triggered when the users takes an action. Examples include tapping a menu, loading a URL, sharing content or saving content for later. An Event is also tagged with a Method (how was the Event triggered) and an optional Extra tag (extra context for the Event).
• Sessions: A telemetry probe triggered when the application starts a short-lived scope or situation. Examples include showing a Home panel, opening the awesomebar or starting a reading viewer. Each Event is stamped with zero or more Sessions that were active when the Event was triggered.

We add the probes into any part of the application that we want to study, which is most of the application.

Visualizing Data

The raw telemetry data is processed into summaries, one for Events and one for Sessions. In order to visualize the telemetry data, we created a simple dashboard (source code). It’s built using a great little library called PivotTable.js, which makes it easy to slice and dice the summary data. The dashboard has several predefined tables so you can start digging into various aspects of the data quickly. You can drag and drop the fields into the column or row headers to reorganize the table. You can also add filters to any of the fields, even those not used in the row/column headers. It’s a pretty slick library.

Acting on Data

Now that we are collecting and studying the data, the goal is to find patterns that are unexpected or might warrant a closer inspection. Here are a few of the discoveries:

Page Reload: Even in our Nightly channel, people seem to be reloading the page quite a bit. Way more than we expected. It’s one of the Top 2 actions. Our current thinking includes several possibilities:

1. Page gets stuck during a load and a Reload gets it going again
2. Networking error of some kind, with a “Try again” button on the page. If the button does not solve the problem, a Reload might be attempted.
3. Weather or some other update-able page where a Reload show the current information.

We have started projects to explore the first two issues. The third issue might be fine as-is, or maybe we could add a feature to make updating pages easier? You can still see high uses of Reload (reload) on the dashboard.

Remove from Home Pages: The History, primarily, and Top Sites pages see high uses of Remove (home_remove) to delete browsing information from the Home pages. People do this a lot, again it’s one of the Top 2 actions. People will do this repeatably, over and over as well, clearing the entire list in a manual fashion. Firefox has a Clear History feature, but it must not be very discoverable. We also see people asking for easier ways of clearing history in our feedback too, but it wasn’t until we saw the telemetry data for us to understand how badly this was needed. This led us to add some features:

1. Since the History page was the predominant source of the Removes, we added a Clear History button right on the page itself.
2. We added a way to Clear History when quitting the application. This was a bit tricky since Android doesn’t really promote “Quitting” applications, but if a person wants to enable this feature, we add a Quit menu item to make the action explicit and in their control.
3. With so many people wanting to clear their browsing history, we assumed they didn’t know that Private Browsing existed. No history is saved when using Private Browsing, so we’re adding some contextual hinting about the feature.

These features are included in Nightly and Aurora versions of Firefox. Telemetry is showing a marked decrease in Remove usage, which is great. We hope to see the trend continue into Beta next week.

External URLs: People open a lot of URLs from external applications, like Twitter, into Firefox. This wasn’t totally unexpected, it’s a common pattern on Android, but the degree to which it happened versus opening the browser directly was somewhat unexpected. Close to 50% of the URLs loaded into Firefox are from external applications. Less so in Nightly, Aurora and Beta, but even those channels are almost 30%. We have started looking into ideas for making the process of opening URLs into Firefox a better experience.

Saving Images: An unexpected discovery was how often people save images from web content (web_save_image). We haven’t spent much time considering this one. We think we are doing the “right thing” with the images as far as Android conventions are concerned, but there might be new features waiting to be implemented here as well.

Take a look at the data. What patterns do you see?

Here is the obligatory UI heatmap, also available from the dashboard: