Set Up Metrics

Why set up metrics

An experiment without metrics is just a coin flip. Metrics define what success looks like — whether that is more clicks, higher revenue, or fewer support tickets. Choosing the right metrics before you start ensures you measure what matters and avoid p-hacking after the fact.

Metric types

Metric type	What it measures	Example
Conversion	Whether an event occurred (binary)	“Clicked sign-up button”
Numeric	A count or sum per visitor	”Number of pages viewed”
Revenue	Monetary value per visitor	”Total purchase value”

Each experiment should have one primary metric that drives the decision, plus optional secondary metrics for additional insight.

	Primary	Secondary
Purpose	The single metric that decides the winner	Guards against unexpected side effects
Count	Exactly one per experiment	Two to four recommended
Decision weight	Determines ship/no-ship	Advisory — flags tradeoffs

Example setup for a checkout experiment:

• Primary: Revenue per visitor (increase)
• Secondary: Cart abandonment rate (decrease), page load time (decrease), support tickets (decrease)

Create an event

Events are the raw signals that metrics are built on. You can create events through the UI or by sending them from code.

Create a click event (Web Experimentation)

1. Navigate to Implementation → Events
2. Click Create New Event
3. Select Click event
4. Enter a descriptive name (e.g., “CTA button click”)
5. Use the visual selector to click the target element on your page, or enter a CSS selector
6. Click Save

Create a custom event (Feature Experimentation)

Custom events are sent from your application code when a meaningful action occurs.

Track a custom event

JavaScript Python C# (.NET)

javascript Copy

// After the user completes a purchase
const user = optimizely.createUserContext('user-123', {
  country: 'US',
});

// Simple conversion event
user.trackEvent('purchase_complete');

// Revenue event with tags
user.trackEvent('purchase_complete', {
  revenue: 4999,  // Value in cents
  quantity: 2,
});

python Copy

# After the user completes a purchase
user = client.create_user_context('user-123', {
    'country': 'US',
})

# Simple conversion event
user.track_event('purchase_complete')

# Revenue event with tags
user.track_event('purchase_complete', {
    'revenue': 4999,  # Value in cents
    'quantity': 2,
})

csharp Copy

// After the user completes a purchase
var user = optimizely.CreateUserContext("user-123", new UserAttributes
{
    { "country", "US" },
});

// Simple conversion event
user.TrackEvent("purchase_complete");

// Revenue event with tags
var tags = new EventTags
{
    { "revenue", 4999 },   // Value in cents
    { "quantity", 2 },
};
user.TrackEvent("purchase_complete", tags);

Important: The revenue tag must be an integer in cents (or the smallest currency unit). A $49.99 purchase is 4999.

Add metrics to an experiment

1. Open your experiment and navigate to the Metrics tab
2. Click Add Metric
3. Select or search for the event you created
4. Configure the metric:
   • Metric name — A clear label (e.g., “Purchase conversion rate”)
   • Aggregation — Unique conversions (binary), total conversions (count), or revenue (sum)
   • Winning direction — “Increase” for positive metrics, “Decrease” for metrics like bounce rate
5. Mark one metric as the Primary metric — this is the metric used to determine the winner
6. Add additional secondary metrics as needed

Configure pageview events (Web Experimentation)

Pageview events track when a visitor reaches a specific page. These are useful for funnel-based metrics.

1. Navigate to Implementation → Events
2. Click Create New Event
3. Select Pageview event
4. Enter the URL match — use substring, exact, or regex matching
5. Name the event (e.g., “Reached checkout confirmation”)
6. Attach it to your experiment as a metric

Set statistical significance

1. In the experiment settings, navigate to Statistical Settings
2. Set the significance level — 90% for low-risk tests, 95% for most decisions (default), 99% for high-impact changes
3. Set the minimum detectable effect (MDE) — the smallest improvement worth detecting. A 2% MDE needs more traffic than a 10% MDE.
4. Enable multiple comparisons correction when testing three or more variations to reduce false positive risk

Lower MDE values require more traffic. Balance sensitivity with how long you can run the experiment.

Best practices

• Decide metrics before starting — Adding metrics mid-experiment introduces bias
• Limit secondary metrics — 3 to 5 additional metrics is usually sufficient
• Use revenue metrics for e-commerce — They capture magnitude, not just frequency
• Name events descriptively — “purchase_complete” is better than “event_7”
• Track events close to the decision point — If you are testing the checkout button, track the purchase, not just the button click

Troubleshooting

Issue	Cause	Fix
Metric shows zero conversions	Event not firing or name mismatch	Check browser dev tools (Web) or SDK logs (Feature) for the event
Revenue shows as whole numbers	Revenue not sent in cents	Multiply by 100 before sending
Metric not attached to experiment	Event created but not added as metric	Open experiment and add the event in the Metrics tab
Results seem delayed	Events batch before sending	Wait 5-10 minutes; SDK batches events for efficiency

Related content

Continue your journey with these related pages.

📖Understand the concepts

• How Experimentation Works

🔧Follow a guide

• Create an Experiment
• Custom Metrics and Events
• Analyze Results