Explainer · Case studies
Five anomalies, in detail
What detection actually feels like on the day it matters
Each case below walks through one real failure mode end to end — what the team was doing that morning, what the data did, what ainpulse caught, and what a good response looks like. Five cases. Five different failure modes.
Case 01 · Channel composition
Organic regressed. Paid covered. Nobody saw it
The setup. Monday morning standup. SEO tracks organic in their weekly report, paid media watches their campaign numbers daily, and the top-line report everyone else sees shows total sessions for the property — steady as it gets. Comparing the two trends against each other is nobody's explicit job.
What happened. For seven days, organic search traffic fell 30% on a key property. Paid budget had been increased the same week for an unrelated campaign launch, and paid sessions lifted to roughly the same magnitude. The total stayed almost perfectly flat. The two trends moved silently against each other inside the aggregate.
Sessions by channel · acme-store.com
Illustrative. Two trends inside a flat aggregate.
Detected: 2026-05-12 Organic sessions are −30% below baseline, sustained 7 days. Total sessions appear stable due to compensating paid lift. Composition shift surfaced inside the aggregate.
What you do. Open the property's channel breakdown — not the aggregate. Confirm organic regression isn't being masked by a paid lift. Investigate organic root cause: content, ranking, site changes.
Case 02 · Traffic volume
Sessions dropped 35% on Tuesday. Nobody looked until Friday
The setup. You manage 14 properties. Most look stable on Monday morning. You move on. Tuesday a UTM tagging change shipped on one of them — nobody on the analytics team knew it had gone out.
What happened. Sessions on that property dropped 35% overnight, then held flat at the new low. The data was in GA4 the whole time — anyone who opened that property would have seen it. But with 14 properties on the roster, nobody had a reason to open this one until Friday's weekly review. By then it had been four days, and the UTM tagging update had been silently breaking a major referral attribution path the entire time.
Daily sessions · store-acme.com
Illustrative. Sustained 3-day pattern triggers the alert.
Detected: 2026-05-08 Sessions are −35% below baseline, sustained 3 days. Pattern began 2026-05-05. Magnitude and duration match a structural change, not a one-day blip.
What you do. Check what shipped recently — UTM rules, redirect rules, tag manager publishes. The drop's shape suggests a structural break, not a temporary dip.
Case 03 · Conversions
Paid CVR dropped 57% overnight. Bid strategy kept buying
The setup. You manage paid acquisition for a client. Friday afternoon they pushed a landing page redesign — supposedly minor copy changes. The bid strategy is tCPA, optimizing automatically against expected conversion rate.
What happened. Saturday morning paid CVR halved. From 4.2% to 1.8%. Sessions stayed normal — people were still clicking the ads — but few of them were converting. The bid strategy didn't react: it takes days of data to recalibrate. By Sunday night, two full days of paid budget had run against the broken landing page. ROAS regressed proportionally.
Paid CVR · acme-store.com
Illustrative. Rate-based metric, same baseline logic.
Detected: 2026-05-09 Paid sessions converting at 1.8% vs baseline 4.2% (−57%), sustained 2 days. Pattern began 2026-05-07 — a recent landing page deploy may be the cause.
What you do. Pause paid spend until the landing page is rolled back or fixed. Test the conversion path manually first. Confirm tag firing, then resume.
Three cases in. The next one could be from your data
Connect one property in under five minutes — your baseline starts forming today.
Case 04 · Revenue & AOV
Conversions still firing. Daily revenue regressed anyway
The setup. You manage e-commerce monitoring for a client. Daily revenue is the metric leadership watches. Order count is the metric ops watch. Both are looking close to normal — orders are coming in at expected rate.
What happened. For three days, average order value dropped 22%. A bug in the cart system was permanently applying a "first-time customer" discount code to every order — even repeat customers. Order count looked normal. Revenue looked slightly down, easy to dismiss as seasonal noise. The AOV deviation surfaced the real cause.
Average order value · acme-store.com
Illustrative. AOV holds the signal that volume hides.
Detected: 2026-05-12 AOV at $47 vs baseline $60 (−22%), sustained 3 days. Order count is normal — the regression is per-order, not in volume. Recent cart or pricing changes likely.
What you do. Check for discount logic changes — cart, promotion engine, code application. Volume looks fine; the leak is per-order.
Case 05 · Conversions · Inflated
Conversions doubled overnight. The data was wrong
The setup. Tuesday morning standup. Yesterday's purchase conversions looked extraordinary — up 110% over the property's normal pattern. The team's first instinct was to look at what marketing did right.
What happened. ainpulse flagged the spike as anomalous before anyone could celebrate. Conversions outside the baseline zone in either direction get attention — up or down. A GTM container update the day before had introduced a duplicate purchase trigger. Every successful checkout was firing two purchase events. The "good week" was an artifact.
Daily purchases · acme-store.com
Illustrative. Spike anomalies surface as fast as drops.
Detected: 2026-05-13 Purchase events firing at 105/day vs baseline 50 (+110%), sustained 3 days. Pattern is unusually high — magnitude and duration suggest a tracking change, not real performance lift.
What you do. Check recent GTM publishes for the conversion event. Confirm single-fire behavior. If duplicate trigger, deprecate it and re-backfill the metric.
Questions
Common questions
Five cases shown.
Yours start the day you connect
Connect one property in under five minutes. Your baseline forms within hours. The first alert that lands is from your data — not a story.