First, split the drop to find where it lives
Start with the business system. If orders, leads or bookings held steady while the reported rate fell, the cause is measurement or traffic rather than lost sales, and the first and last causes below come first. If the business figures fell too, the loss is real and the question is where it happened.
Then build a comparison in Google's Explorations: the days since the drop against the same number of days before it, with device category, browser, session source and landing page as breakdowns, and the steps of the conversion FunnelModel of the customer journey stages. laid out in order. Read the conversion rate (CVRShare of visits or clicks that convert.) and the step rates in each slice, not the totals. A drop that sits in one browser or device points at a bug; one that sits in one source points at the traffic mix; one that sits at the payment step points at checkout; one that is spread evenly across every slice points at measurement, price or offer. Keep a list of everything that changed in the days around the drop, including releases, tag container versions, campaigns launched or paused, prices, stock and payment provider notices; the conversion rate troubleshooting hub covers the separate case of a drop that followed a redesign. When the rate moves the other way, jumping with no change anyone can name, the diagnosis for a conversion rate that jumped overnight checks for double counting, spam entries and a shrunken denominator before crediting the site. For a fall that matches none of the causes below, the general method in explaining a change in a marketing metric shows how to separate a shift in traffic mix from a change in the rate inside each segment. When the fall shows mainly in Google Ads reporting rather than across the site, the diagnosis for Google Ads conversions that dropped suddenly checks the account's own conversion settings, windows and bidding before the page.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Reported conversions fell to near zero, or by a round share, while business figures held | Analytics against the order or lead system for the same days; recent tag and site releases | Tracking stopped recording | Run the outage procedure; restore and verify the event |
| Sessions rose on the same day the rate fell; one source or campaign grew sharply | Rate by source and campaign before and after; what launched that day | Traffic mix changed | Judge the new source on its own; keep it off the core rate |
| The fall sits in one browser, device or operating system | Funnel step rates by browser and device across the change date | Bug in one browser, device or step | Reproduce, fix and test on the affected combination |
| Checkout starts held but payments fell, or one payment method went quiet | Step rate at payment; payment provider declines and errors; method mix | Checkout or payment failing for some customers | Fix the failing method or step; restore alternatives |
| Product views held but add-to-basket or enquiry fell across all segments | Price, stock, delivery charge and offer changes against competitors' current terms | Price, stock, shipping or offer changed | Reverse or explain the change; fix availability |
| Sessions jumped with near-zero engagement, odd locations or repeated patterns | Engagement, location, hostname and timing of the new sessions | Bot traffic diluting the rate | Exclude the traffic; report the rate on clean data |
Fix, verify and prevent each cause
Conversion tracking stopped recording
A release that changed a button, a confirmation page or the data layer, a new consent banner, or a paused tag can stop a Key EventA GA4 event marked as important to the business, the property's measure of success. from being sent while customers keep buying. Google documents that Analytics receives interactions as events, so when the event stops, the conversion disappears from every report built on it.
Fix: follow the tracking outage response procedure to confirm the outage, find the layer that broke, protect automated bidding and deploy the smallest fix; for a missing purchase event specifically, the purchase event diagnosis narrows it further.
Verify: real conversions, not test ones, appear again in analytics and each ad platform at their usual ratio to business figures.
Prevent: set an alert on the daily ratio of reported conversions to business orders or leads, and add a conversion test to every release checklist.
A new source changed who arrives
A broad campaign, a viral mention, a new referral partner or a change in an ad platform's targeting can add a crowd of visitors who were never close to buying. The rate falls because the denominator changed, not because the site got worse. The intent signals guide explains how to read which stage a visit sits at from its source, query and behaviour.
Fix: report the new source separately, give it a next step that fits its intent, and keep the site's core rate on the sources it was measured on before.
Verify: with the new source separated, the rate for the established sources matches its earlier level.
Prevent: annotate every campaign launch and read the conversion rate by source as standard, never only as a blended total.
A bug that only some visitors meet
A browser update, a new script or a small release can break a date picker, a button or a form on one browser or device while everything works on the team's own machines. The drop then sits almost entirely in that segment.
Fix: use the device, browser and step breakdowns from the funnel analysis tutorial to find the step where that segment's rate fell, reproduce the fault on the same combination, fix it and treat it as a bug rather than a hypothesis to test.
Verify: the affected segment's step rate returns to the level of the other segments within days.
Prevent: keep a short list of the browser and device combinations that carry most conversions and walk the journey on each after every release.
Checkout or payment failing for some customers
A payment provider change, an expired certificate, a new fraud rule, a timeout or a missing wallet on one device can turn away customers at the very last step, often silently.
Fix: compare the payment step's rate and the mix of payment methods before and after the date, read the provider's decline and error reports, and restore the failing method or its alternative, following the payment and error-handling sections of the checkout optimisation guide.
Verify: payment-to-confirmation completion and the share of each payment method return to normal.
Prevent: monitor completion at the payment step daily and subscribe the team to the payment provider's change and incident notices.
A price, stock, delivery or offer change
A price rise, a best seller going out of stock, a new delivery charge or the end of a promotion changes the decision every visitor makes, so the drop appears across all segments at once, usually on the day of the change. The pricing and packaging guide explains why costs that appear late, unclear terms and hidden charges damage both conversion and trust.
Fix: list commercial changes made in the days around the drop, compare them with competitors' current terms, and reverse, soften or explain the change; restock or replace unavailable items on high-traffic pages.
Verify: add-to-basket or enquiry rates recover after the change is reversed or explained, without a fall in order value.
Prevent: give marketing advance notice of price, stock and delivery changes and record them in the same change log as site releases.
Automated traffic swelling the count
Bots, scrapers and referral spam add sessions that never convert, so the rate falls while sales do not. They tend to show as sudden bursts with near-zero engagement, unusual locations or identical behaviour.
Fix: confirm the pattern with the bot traffic spike diagnosis, then apply the exclusions and filters described in the data quality guide so the rate is reported on human sessions.
Verify: the cleaned rate returns to its previous level and the suspicious sessions no longer appear in source and location reports.
Prevent: include engagement, location and hostname checks in the weekly data quality routine.
A prevention routine for sudden drops
Sudden drops are cheapest when they are caught in hours rather than weeks, so the routine is mostly about alerts and a change log. Keep one dated log of everything that can move conversion: site releases, tag container versions, consent changes, campaigns launched or paused, prices, stock, delivery terms and payment provider changes. Set alerts on three ratios: reported conversions against business orders or leads, sessions against their usual range, and payment-step completion. Each week, read the conversion rate by device, browser and source rather than as one figure, so that a fall in one slice is visible before it drags the total down. After every release, walk the journey on the browser and device combinations that carry most sales, including the payment step with each common method. The tutorial on checking conversion paths after a release shows how to choose those combinations, script the walk and follow each result to the order or lead. When an alert fires, start with the change log entry closest to the time of the drop, because it explains most sudden falls, and record what the cause turned out to be so the next investigation starts faster.
A homeware shop's Tuesday drop
A homeware shop's conversion rate fell from 2.8% to 1.9% overnight on a Tuesday, with sessions up 30%. Orders in the shop system were down only from around 410 to 380 a day. Splitting the drop showed two things. A social campaign launched on Monday evening had brought 4,000 extra sessions a day converting at 0.3%; with it separated, the rate for other sources was 2.6%. The remaining fall sat in one mobile browser, where the payment step completion had dropped from 88% to 61% after the payment provider updated its embedded form. The shop reported the campaign separately, switched that browser to the provider's hosted payment page until a fix arrived, and within four days the rate for established sources was back at 2.8%.
Common mistakes
Redesigning pages before checking the tracking. Reading only the blended rate. Assuming every drop is a site problem when the visitors changed. Testing the fix for a bug instead of shipping it. Forgetting that price and stock are part of conversion. Investigating without a change log.
