How to use Interaction Analysis
Interaction analysis measures which UI elements your shoppers actually use — filters, sorts, search, carousels — so you can prune dead weight and double down on what drives conversion.
Interaction Analysis
Interaction analysis quantifies which UI elements users actually engage with — and which sit unused on the page.
Interaction analysis is the practice of measuring every meaningful click, tap, hover, scroll-into-view and input on a page, then aggregating those events by element so you can see which features earn their place. It answers questions like "does anyone use the colour filter on this PLP?" or "how often does the related-products carousel get a swipe past slide one?"
It sits inside the broader discipline of behavioral analytics, but where behavioral analytics zooms out to journeys and cohorts, interaction analysis zooms in to the component level. The output is a ranked list of UI elements by engagement rate, dwell, and downstream conversion contribution — the evidence base for what to keep, fix, or kill.
Most stores ship UI on instinct. The product team adds a wishlist icon because competitors have one, a size guide because returns are creeping up, a "customers also bought" carousel because the theme came with one. Two years later nobody knows which of those elements actually moves the needle — and the page weighs 4MB.
Interaction analysis replaces that instinct with evidence. Instead of asking "do shoppers like our filters?", you ask "what share of PLP sessions touch a filter, which filter values get picked, and do filter-users convert at a different rate?" The answers tend to be uncomfortable — and very useful.
What interaction analysis actually measures
Three numbers do most of the work. First, reach: of all sessions that saw the element, what share interacted with it at least once? Second, depth: among those who interacted, how far did they go — one filter or five, slide one or slide six? Third, lift: do interactors convert at a higher rate than non-interactors on the same page?
Reach tells you whether the element is discoverable. Depth tells you whether it's useful once discovered. Lift tells you whether it's actually contributing to revenue, or whether interactors were already high-intent shoppers who would have converted anyway.
The trap is stopping at reach. A search-as-you-type bar with 8% reach sounds low — until you discover that those 8% convert at 4× the site average and account for 22% of revenue. That's a load-bearing element, not a niche feature.
Reach without lift is vanity
A carousel can show 60% reach simply because it autoplays into the viewport. If those swipers convert at the same rate as non-swipers, the element is decoration. Always pair engagement with a downstream conversion delta before declaring a feature a winner.
How to instrument it without dev tickets
The legacy approach is one custom event per element — a JIRA ticket per filter, per sort dropdown, per accordion. Stores end up with 40 half-named events, no schema, and an analytics setup nobody trusts. A single lightweight snippet that auto-captures clicks, form inputs, and visible scroll depth removes that bottleneck.
What you want is element-level capture by default, with the option to label specific components for cleaner reporting. Filters on a Shopify PLP should appear in your dashboard the day you ship them — not three sprints later when someone finally writes the tracking spec.
Typical engagement rate by UI element (apparel PLP, mobile sessions)
The chart above is the typical shape we see on mobile PLPs in apparel: swatches and size filters dominate, sort dropdowns are almost ignored, and quick-view sits in the awkward middle. Your numbers will differ — but the long-tail distribution rarely does.
Benchmarks: what "good" looks like by element type
Benchmarks are blunt instruments — your category and traffic mix shape engagement more than any best-practice guide — but ballparks help you spot the elements that are obviously broken. The table below is what we see across Shopify and WooCommerce stores in the €1M-€15M revenue band.
Read the conversion-lift column carefully. A 1.4× lift means interactors convert 40% more often than non-interactors on the same page view — not that the element causes the lift. Causation needs an A/B test; this just identifies the candidates worth testing.
Engagement and conversion lift by UI element (DTC apparel & beauty, mobile)
| UI element | Reach (% of sessions) | Median depth | Conversion lift vs non-interactors |
|---|---|---|---|
| Faceted filters (size, colour) | 15-25% | 2.1 filters applied | 1.8×-2.4× |
| Sort dropdown | 3-7% | 1 sort applied | 1.1×-1.3× |
| Search-as-you-type | 6-12% | 1.7 queries | 3.0×-4.5× |
| Related-products carousel | 8-14% | 2.3 slides viewed | 1.2×-1.6× |
| Quick view modal | 5-10% | 1 product opened | 1.4×-1.9× |
| Size guide link (PDP) | 4-8% | n/a | 1.5×-2.1× |
| Wishlist / save icon | 3-6% | 1.4 saves | 0.9×-1.2× |
Two patterns stand out. Search bars punch far above their weight — low reach, huge lift — because they self-select for high-intent shoppers. And wishlist features tend to hover around parity, which means they're either a retention play (worth keeping) or genuine dead weight (worth removing). Interaction data alone can't tell you which; you need to cross-reference with returning-visitor cohorts.
Turning findings into experiments
Interaction analysis on its own is diagnostic, not prescriptive. The output is a hypothesis backlog. A sort dropdown with 4% reach and no conversion lift becomes: "removing the sort dropdown will not hurt conversion and may speed up the PLP — let's A/B test it." A search bar with 4× lift but 6% reach becomes: "making search more prominent will lift overall conversion — let's test a larger placement."
The discipline is to rank candidates by potential impact, not by how curious the data is. An element that's used by 25% of sessions and adds 5% conversion lift is a bigger prize than an element used by 3% with 4× lift, even though the second number is more dramatic. Multiply reach by lift to get a rough revenue-at-stake figure before you queue the test.
The kill-it test
When you suspect an element is dead weight, the cheapest experiment is to remove it for 50% of traffic for a week. If conversion holds, ship the removal — you've reduced page weight, JS execution, and cognitive load in one move. Most stores have three or four of these wins hiding in plain sight.
Frequently asked questions
Heatmaps visualise where clicks land on a page; interaction analysis aggregates those clicks by element and ties them to downstream outcomes. A heatmap tells you the filter bar is hot. Interaction analysis tells you 18% of sessions use it, applying 2.1 filters on average, and converting 2× the page baseline.
It can, if you stack a separate tag for every feature — heatmap tool, session replay, custom GTM events, A/B tester. A single auto-capture snippet typically adds under 30KB and runs after first paint, so checkout speed and Core Web Vitals stay intact. Audit your current tag list first; most stores find they can remove more than they add.
Behavioral analytics is the parent discipline — it covers journeys, funnels, cohorts, and segments across the whole site. Interaction analysis is the component-level slice: which buttons, filters, and widgets get used on a given page. You need both. Journeys tell you where shoppers drop off; interaction analysis tells you which UI elements on those pages are pulling weight.
On mobile apparel PLPs, 15-25% of sessions typically touch at least one filter, with size and colour leading. Below 10% usually points to a discoverability problem — filters hidden behind a drawer with a vague label, or a category page that's already narrow enough that filtering feels redundant.
Not automatically. Check conversion lift first — a search bar with 8% reach can drive 20%+ of revenue. Remove elements that have both low reach AND no conversion lift. A 50/50 removal test for a week is the safest way to confirm before shipping the change site-wide.
For a PLP with steady traffic, two to four weeks usually gives enough volume to see stable engagement rates and lift estimates per element. Seasonal categories or campaign-driven traffic need longer — ideally a full purchase cycle plus a week of buffer — so you don't mistake a sale-week spike for normal behaviour.
Partially. Click and scroll data is only collected from the moment you instrument, so element-level interaction history starts on day one of capture. But pageview, session, and conversion history from GA4 can be imported retroactively, so you have funnel context immediately even if element-level depth needs a few weeks to build up.
Yes — auto-capture is DOM-driven, so it works on any HTML interface including single-page Shopify Hydrogen builds and headless WooCommerce frontends. The one caveat is that virtual route changes need to be tracked as page events explicitly, which most modern snippets handle automatically through history API hooks.
You can't, from observational data alone. Interaction analysis identifies candidates worth testing — "users of feature X convert higher" — but the only way to prove the feature causes the lift is an A/B test that hides or alters it. Use the analysis to rank your experiment backlog by potential revenue impact, not to declare winners.
AI is useful for two things: clustering similar elements automatically (treating all "add to cart" buttons across templates as one entity) and proposing hypotheses from the data ("sort dropdown has low reach and zero lift — propose removal test"). It speeds up the move from raw event data to an experiment-ready backlog, which is usually where teams stall.
Get an AI expert review of your site
Paste your URL — Metricuno's AI runs the same heuristic checks a senior CRO consultant would, scoring your page and prioritising the fixes that'll move conversion fastest.