Prism AI continuously tests your popup variants and automatically directs more traffic toward the best performers. Here's how to read the analytics on your Prism AI dashboard.
Optimization States
Each Prism AI element goes through two phases:
Learning — Prism AI is gathering initial data across all variants. Traffic is split evenly so each variant gets a fair chance. No performance estimates are shown yet.
Optimizing — Prism AI has collected enough data to start identifying winners. Traffic is now being actively directed toward the best-performing variants, and performance estimates are displayed.
Performance Metrics
Once a Prism enters the Optimizing state, you'll see three key metrics:
Email submit rate — The percentage of popup viewers who submit their email address.
SMS submit rate — The percentage of popup viewers who submit their phone number.
Attributed order rate — The percentage of popup viewers who go on to place an order.
For each metric, two values are shown:
The optimized rate (displayed as a percentage) — This is the estimated overall rate your visitors are experiencing right now, with Prism AI directing traffic toward the best variants.
The lift (displayed as an arrow with a percentage) — This is how much better the optimized rate is compared to what you'd see if all variants received equal traffic. A green upward arrow means Prism AI is improving performance.
Attribution window: Prism AI uses a 30-day attribution window for conversion tracking.
Summary Statistics
At the top of the Prism AI page, you'll see aggregate stats across all of your Prism AI elements:
Prism AI elements — The total number of active Prism elements.
Total variants — The total number of variants across all Prisms.
Total popup views — The total number of visitors who have seen a Prism-powered popup.
Avg. email signup lift / Avg. SMS signup lift / Avg. attributed order lift — The average lift across all Prisms that have enough data to report.
How Lift Is Calculated
The lift percentage represents the improvement Prism AI is delivering by intelligently allocating traffic. It compares two scenarios:
Baseline: What would happen if every variant received an equal share of traffic (no optimization).
Optimized: What's actually happening now, with Prism AI sending more traffic to the better-performing variants.
The lift is scaled by Prism AI's confidence in its assessment. When there's a clear winner, the full estimated lift is shown. When performance differences are less certain, the lift is reported more conservatively. This means the numbers you see are a realistic, confidence-adjusted estimate — not an optimistic projection. That said, early results can be noisy — small sample sizes mean the numbers may fluctuate significantly before settling.
Wait for at least a few hundred popup views per variant before drawing conclusions from the lift metrics.
Dimensions
If you've added dimensions (such as landing page or device type), Prism AI optimizes independently for each visitor segment. For example, one variant might perform best for mobile visitors while another wins on desktop.
Prism AI handles this automatically — the lift metrics you see reflect the combined benefit of per-segment optimization.
Why reporting only shows "Overall" and "Other"
If your dimension breakdown shows only Overall and Other categories, Prism AI has not yet gathered enough data to optimize across additional pages or segments. More specific breakdowns appear as more data accumulates over time.
Why results may be limited or slow to appear
Common reasons a Prism experiment produces little usable insight:
Low traffic per variant — the element isn't receiving enough views to distinguish performance.
Too many variants — traffic is split too thinly across options.
Placement on a later step — fewer users reach it, so it accumulates data more slowly than an element on the first step.
The fix is either more time or fewer variants.
Deleted elements still appearing in analytics
Deleted Prism elements can remain visible in analytics. This is expected: they may still exist within reusable components, or could be reintroduced later. They are not removed automatically, but can be deleted manually if you no longer want them reported.
Tips
Give it time. Prism AI needs sufficient data before it can confidently identify winners. The Learning phase typically lasts until each variant has received at least 100 popup views.
More variants = more data needed. Each additional variant increases the amount of data required before optimization kicks in. Start with 2–3 variants for the fastest results.
Lift will grow over time. As Prism AI gathers more data and becomes more confident, it will allocate traffic more aggressively toward winners, increasing the lift.
Multiple Prism elements are fine, with enough traffic. Using several Prism elements across one popup works, but only if traffic supports it. Introduce additional elements after the first has gathered sufficient data.