Metrics

  1. Defining a Metric
  2. Metrics From Your Database
  3. Google Analytics
  4. Creating Your Own Metric
  5. Digging Deeper

A good starting point for improving — on anything — is measuring. Vanity allows you to measure multiple metrics, best way to tell how well your experiments are doing.

Startup metrics for pirates: AARRR!

  1. Acquisition
  2. Activation
  3. Retention
  4. Referral
  5. Revenue

Defining a Metric

Vanity always loads metrics defined in the experiments/metrics directory. A metric definition is a Ruby file that looks like this:

metric "Signup (Activation)" do
  description "Measures how many people signed up for our awesome service."
end

That’s a basic metric and you feed it data by calling the track! method. For example:

class AccountsController < ApplicationController

  def create
    @person = Person.new(params[:person])
    if @person.save
      Vanity.track!(:signup)      # track successful sign up
      UserSession.create person
      redirect_to root_url
    else
      render :action=>:new
    end
  end

end

The metric identifier is the same as the file name. The above example defines the metric :signup in the file experiments/metrics/signup.rb.

You can call track! with a value to track. This example tracks how many items were bought during the day:

def checkout
  Vanity.track!(:items, @cart.items.count)
  . . .
end

Calling track! with no value is the same as calling with one, and for convenience you can pass zero and negative numbers, both will be ignored.

Define, track, and you’re ready to roll.

Metrics From Your Database

If you already have the data, why not use it?

This example defines a metric for signups, based on the number of Account records created each day:

metric "Signup (Activation)" do
  description "Measures how many people signed up for our awesome service."
  model Account
end

You don’t need to call track! with this metric, all the data already exists. It’s a simple query to count the number of records created, grouped by their timestamp (created_at). And since it’s querying the database, you’ll immediately see historical data for the last 90 days.

Even though the metric itself doesn’t store any information, it needs to update experiments whenever new records are created. To do that, it registers itself as an after_create callback on the model.

Some metrics measure values, not occurrences. For example, this metric measures user satisfaction by calculating average value from the column rating:

metric "Satisfaction Survey" do
  description "Measures how many people signed up for our awesome service."
  model Survey, :average=>:rating
end

The aggregates you can use this way are: :average, :minimum, :maximum and :sum.

You can use a condition when the metric only applies to some records. Here’s a metric that only measures unlimited accounts:

metric "Signups to Unlimited" do
  description "Signups to our All You Can Eat and Burp Unlimited plan."
  model Account, :conditions=>{ :plan_type=>'unlimited' }
end

If you have named scopes, you’ll want to use them instead:

metric "Signups to Unlimited" do
  description "Signups to our All You Can Eat and Burp Unlimited plan."
  model Account.unlimited
end

When you view this metric, it calculates the number of accounts created on any given day that are currently unlimited plans. So, if ten accounts were created over the past week, and today five of them upgraded to unlimited plan, the metric will show five unlimited accounts (current state) but spread over the past week (their created_at timestamp).

If your metric uses aggregates or conditions, and the aggregate/conditional attributes change over time, and you need to know when the change took place, consider tracking the event.

This example tracks when accounts were created or upgraded to unlimited plan:

metric "Signups (Unlimited)" do
  description "Signups to our All You Can Eat and Burp Unlimited plan (including upgrades)."
  Account.after_save do |account|
    track! if account.plan_type_changed? && account.plan_type == 'unlimited'
  end
end

The model specifier can also take an :identity option. :identity: should be a Proc that specifies how to fetch the identity of the experiment participant. This is useful when constructing objects outside the ActionController context (perhaps in a background task). However, note that your experiment participants may still be identified by their anonymous cookie identifier, if you started any experiments before the user was identifiable.

This example will record conversions for the Account to which the Subscription belongs:

metric "Subscriptions finished on backend" do
  description "These signups were actually created in a background task."
  model Subscription, :identity => lambda { |record| record.account_id }
end

Google Analytics

You can easily include Google Analytics metrics in your Vanity dashboard. You’ll need, in addition to Vanity, to use Garb, a Ruby wrapper for the Google Analytics API.

Login to Google Analytics using either username and password, or OAuth authentication token. Here’s a sample config/environment snippet:

Rails::Initializer.run do |config|
  gems.config "vanity"
  gems.config "garb"

  . . .
  config.after_initialize do
    require "garb"
    ga = YAML.load_file(Rails.root + "config/ga.yml")
    Garb::Session.login(ga['email'], ga['password'], account_type: "GOOGLE")
  end
end

To define a metric that corresponds to the Google Analytics daily visitors:

metric "Acquisition: Visitors" do
  description "Unique visitors on any given page, as tracked by Google Analytics"
  google_analytics "UA-1828623-6", :visitors
end

The first argument is the GA profile, the second argument the GA metric name (defaults to pageviews).

You can use the full Garb API by accessing the report directly, for example:

metric "Activation: Signups" do
  google_analytics "UA-1828623-6"
  report.filters do
    eql(:page_path, 'welcome')
  end
end

See the Garb documentation and Google Analytics API for more details.

Creating Your Own Metric

Got other ideas for metrics? Writing your own metric is fairly simple.

The easiest way to create your own metric is by adding your own values method, for example:

metric "Hours in a day" do
  description "Measures how many hours in each day."
  def values(from, to)
    (from..to).map { |i| 24 }
  end
end

This example is based on Vanity::Metric. You can, of course, base your metric on any other class.

For simplicity, a metric is any object that implements these two methods:

A metric may also implement these methods:

If you wrote your own metric implementation, please consider contributing it to Vanity so we can all benefit from it. Thanks.

Digging Deeper

All metrics are listed in Vanity.playground.metrics, a hash that maps metric identifier to metric object. Methods like track! and metrics (see A/B tests) reference metrics using their identifier.

On startup, Vanity loads all the metrics it finds in the experiments/metrics directory. The metric identifier is the same as the file name, so experiments/metrics/coolness.rb becomes :coolness.

You can always populate the hash with your own metrics.

When Vanity loads a metric, it evaluates the metric definition in a context that has two methods: metric and playground. The metric method creates a new Vanity::Metric object, and evaluates the block in the context of that object, so when you see the metric definition using methods like description or model, these are all invoked on the metric object itself.

A Vanity::Metric object responds to track! and increments a record in the database (an O(1) operation). It creates one record for each day, accumulating that day’s count. When generating reports, the values method fetches the values of all these keys (also O(1)).

You can call track! with a value higher than one, and it will increment the day’s count by that value.

Any time you track a metric, the metric passes its identifier, timestamp and count (if more than zero) to all its hooks. A/B tests use hooks to manage their own book keeping. When you define an experiment and tell it which metric(s) to use, the experiment registers itself by calling the hook method.

When you call model on a metric, this method changes the metric definition by rewriting the values method to perform a query, rewriting the track! method to update hooks but not the database, and register an after_create callback that updates the hooks.

How about some tips & tricks for getting the most out of metrics (you might call them “best practices”)? Got any to share?