Continuous Metrics

A binary metric measures whether something happened — a click, a signup, a purchase. A continuous metric measures how much — revenue per session, average order value, dollars per signup. Continuous metrics give you a richer signal but require slightly different statistical handling. This guide covers when to use them and what changes when you do.

Binary vs Continuous

Question	Metric type
”Did the visitor convert?”	Binary — Beta-Binomial model
”How much revenue did the visitor produce?”	Continuous — Normal model

Both can run on the same goal at the same time. If you fire a conversion event with a value, Split Test Pro records it as a binomial count (the visitor converted) AND as a continuous data point (the visitor produced $X). You’ll see both modes reflected in the Results dashboard.

How to Trigger Continuous Mode

Continuous mode is opt-in per event fire, not per goal definition. To produce continuous data for a goal, pass a value (or amount — they’re aliases) when firing the event:

window.SplitTestPro.trackConversion("purchase", {
  value: 49.0,
  currency: "USD",
});

Without value, the event is recorded as a binary conversion only.

You don’t need to declare anything in the goal config — the value field on the event is what flips the engine into continuous mode for that data point. A goal that sometimes has value and sometimes doesn’t will produce both binary and continuous results, with the continuous one based on the subset of events that included a value.

What Changes Statistically

The Bayesian engine handles the two cases differently. (See Bayesian Stats Explained for the full treatment.)

Binary metric

• Models each variant’s conversion rate as a Beta distribution with a uniform Beta(1, 1) prior.
• Probability to be best is computed via Monte Carlo sampling across all variant distributions.
• Credible interval is the central 95% of the posterior conversion rate.

Continuous metric

• Models each variant’s mean value as a Normal distribution parameterized by the sample mean and standard error.
• Probability to be best is computed the same way (Monte Carlo).
• Credible interval represents the range of plausible values for the per-visitor mean (e.g., “we’re 95% sure the average revenue per session for Variant B is between $1.20 and $1.45”).

The continuous credible interval is reported in the Results view as a percent improvement vs control, e.g., “+8% to +18%.”

Reading Continuous Results

Continuous metrics show up on the experiment’s Results page in a Custom Metrics card (HTML) or in the built-in revenue panels (Shopify — Revenue per Session, AOV, Revenue per Purchaser).

For each continuous metric, you’ll see:

• Mean — the per-session/per-visitor average for that variant.
• Total — the sum across all events.
• Trials — how many events contributed (i.e., events that actually had a value).
• Probability — the probability this variant beats control on the continuous metric.
• Credible interval — the range of plausible lift, expressed as a percent.

When to Use Continuous Metrics

Use continuous when:

• Revenue varies meaningfully per conversion — testing a free-shipping threshold that might lift conversion rate but lower AOV. Binary alone misses this.
• Engagement is the goal, not just a yes/no — testing a new content layout where you care about how long visitors stay or how many items they view.
• You’re comparing high-margin vs low-margin outcomes — a discount variant that converts more sessions but at lower revenue per conversion.

Don’t use continuous when:

• The action is binary by nature — newsletter signup, account creation. There’s nothing to count beyond the binary “did it happen.”
• Values are noisy and don’t correlate with the test variable — testing button color and tracking the value of whatever they happened to buy. The variant doesn’t influence the order size; you’ll get noise.

Continuous Metrics on Shopify

Shopify workspaces get Revenue per Session and Average Order Value as built-in continuous metrics — no setup needed. The Web Pixel reads order totals and feeds them in automatically. Both are selectable as primary metrics.

If you need a custom continuous goal on Shopify (e.g., revenue from a specific product line), the path today is to fire a custom event with value from theme JavaScript — same pattern as HTML.

Trade-Offs

Continuous metrics are richer but they’re also more variable — a few outlier high-value conversions can swing the result. The Normal model handles this in principle (variance is part of the parameters), but in practice:

• Continuous metrics often need more data to reach the same probability-to-be-best as a binary metric on the same goal.
• A handful of huge orders early can give Variant B a misleadingly large lead. Run for at least a full week to wash this out.
• Outliers — like a $5,000 order during a B2B test — can dominate. Consider whether to clip or trim values before reporting.

The Bayesian engine doesn’t currently auto-trim outliers. If you have a known anomaly, the cleanest move is to wait until the cumulative data shifts the average back to representative values.

Common Mistakes

• Reporting on a continuous metric with very few trials. A continuous mean with N=8 isn’t telling you anything reliable. Wait for at least 100+ trials per variant before drawing conclusions.
• Forgetting to pass value. If you wanted continuous tracking but called trackConversion("purchase") without value, you only get binary. Fix it forward and look at the data from the fixed point onward.
• Using continuous as the primary metric on a low-trial goal. Pick a binary metric as primary in this case; treat the continuous metric as a secondary signal.

Next Steps

• Understand the Bayesian model that powers continuous metrics: Bayesian Stats Explained.
• Set up the events that produce continuous data: Custom Events.
• See how Shopify’s Revenue per Session works under the hood: Shopify Funnel Tracking.