Why did our ecommerce conversion rate suddenly drop? Seven causes to rule out first

When an ecommerce conversion rate drops suddenly, check the measurement before the shopper. This is the order we rule causes out in: broken or changed tracking, a shift in traffic mix between devices or channels, a change in how traffic is counted, a site release, paid traffic landing on unavailable products, promotions, and only then a genuine change in shopper behavior. The first three are the fastest to test, and when one of them is the cause it saves a team from redesigning pages that were never broken.

Conversion rate is a ratio, orders divided by sessions, and a ratio can move because either side moved, because the mix underneath it moved, or because the way one side is counted changed. That is why the obvious theory is so often wrong.

1. Is the drop real, or did tracking break?

Start by asking whether the number can be trusted. The fastest test is to read related metrics together and look for a combination that cannot all be true at once.

At one large retailer, clicks on product recommendations appeared to be down 85% year over year, and the number had been carried through several weekly business reviews. Attributed revenue from the same placements was down only about 10%, and the conversion rate those two numbers implied had more than tripled. No change in shopper behavior produces that pattern. The cause was a tracking variable that had stopped firing on recommendation interactions. Several weeks of reporting were discounted instead of acted on, and the redesign an 85% decline would reasonably have triggered never started. (The full case study.)

What to check: did exposure, clicks, revenue and implied conversion move together? If one of them moved and the others did not, suspect measurement first.

2. Did the traffic mix shift between devices or channels?

A blended conversion rate can fall while every individual segment holds steady or even improves, because traffic moved from a high-converting surface to a lower-converting one.

At the same retailer, a product manager asked whether push-notification traffic was making her app conversion rate read low. It was a reasonable theory, but two quick checks ruled it out: the app's share of visits had moved by less than a tenth of a point, and site-wide conversion had fallen almost exactly as much as web-only conversion. The real decline was elsewhere. Desktop accounted for 88% of it, and it had been hit twice: its conversion rate slipped week over week and its share of web visits slipped with it. On a year-over-year view the picture inverts, with desktop conversion up close to a full point while its share of visits fell from roughly 23% to 17%. Traffic reallocating from the best-converting surface to the worst was swamping a real rate gain. (The full case study.)

What to check: break the rate into each device and channel's own conversion rate and its share of traffic. If shares moved more than rates, it is a mix shift.

3. Did the way traffic is counted change?

Changes to consent banners, bot filtering, attribution rules or channel definitions change the denominator without changing a single shopper.

The same retailer had an earlier example running the other way. A privacy-consent change suppressed about 15% of daily web traffic for six weeks. Orders held steady, so conversion and click-through improved on every dashboard. It was a stable numerator over a shrunken denominator, and their own team took about a month, across five functions, to diagnose it.

Channel reclassification is the quieter version. A parallel analysis at the same retailer looked like a year-over-year decline coming from one paid channel, until a check of the channel's history showed that two paid channels had swapped a large volume of visits in a single month: a reporting change, not a real shift in media. That analysis was parked rather than delivered.

What to check: consent, tagging, bot-filter and channel-grouping changes in the window. If sessions moved and orders did not, look here.

4. Did a release or tag change go out?

Front-end releases break tracking and checkout flows alike, and they are the change most often missed because the release log lives in a different system from the analytics.

At the same retailer, every web placement for recommendations stepped down by about two-thirds overnight while the mobile app barely moved. That web-only signature pointed at a site-wide front-end framework release the previous night, recorded in the retailer's own change-management system. Nobody had connected the release to the metric until the investigation did.

What to check: deployments, tag-manager publishes and feature flags in the 48 hours before the drop, and whether the drop is confined to the platform that changed.

5. Are you paying to send shoppers to products they cannot buy?

This one hides because it lives between three teams. Between 7% and 9% of the same retailer's product-page visits were landing on items that were fully unavailable, up to 282,000 sessions a week. Merchandising owned it as a chronic inventory metric. Meanwhile mobile-web add-to-cart was falling and being investigated as a page-design problem. The link was automated bidding buying clicks on items that had just gone out of stock, and mobile-web shoppers recovered from that dead end far less often than desktop shoppers did. The first fix, fresher availability on the retailer's own category pages, addressed about 30% of affected sessions and was live two days after the investigation started. (The full case study.)

What to check: the share of product-page sessions landing on unavailable items, split by traffic source, set against add-to-cart on each device.

6. Did a promotion start, end or overlap?

Promotions pull demand forward and push it back. A strong promo last year makes this year's same week look weak, and an offset between two promotions (one running in the comparison week and not in this one) can explain a dip that no single dashboard does. It is one of the most common "mysteries" in a weekly review, and one of the quickest to settle once someone puts the two calendars side by side.

What to check: the promotional calendar for both periods you are comparing, including competitor-matched and sitewide offers.

7. Did shopper behavior actually change?

Only after the first six are ruled out is it worth investigating a real change in behavior: checkout friction, pricing, delivery promises, page performance or a competitor's move. These are real causes, and when they are the answer they are worth a lot, but starting here is how teams spend a month on the wrong fix.

A note on the last two checks. Checks 1 to 5 come from investigations we have run and can point to. Promotions and a genuine change in behavior are standard practice, included because the list would be incomplete without them, not because our own work has turned one up yet. We would rather say which is which than dress a textbook answer up as experience.

A quick order of checks

Why this takes weeks

None of these checks is hard on its own. The difficulty is that the evidence sits in different systems owned by different teams: analytics, paid media, inventory, the release log, the promo calendar. Each team can half-explain the drop from its own dashboard, so the question bounces between them.

That is the problem Augmeta was built for. An AI Operator watches a KPI like conversion rate continuously, reads across those systems, writes each theory in a form that can be disproved, and tests them in parallel. It reports a refuted theory as clearly as a confirmed one, and it says "inconclusive" when the evidence does not settle it. We wrote about how that works in our post on Popper, and about the category in What is Agentic KPI Ops?

Key takeaways

• Check whether the drop is real before explaining it. Broken tracking is common and it is cheap to rule out.
• A blended conversion rate can fall while every segment improves. Separate rate from traffic share.
• Consent, bot filtering and channel definitions change the denominator without changing any shopper.
• Out-of-stock paid traffic is a leak that sits between marketing, merchandising and product, which is why nobody owns it.
• Genuine behavior change is the last thing to investigate, not the first.

Key terms

• Conversion rate: orders divided by sessions for a given period.
• Mix shift: a change in a blended rate caused by traffic moving between segments rather than any segment changing.
• Denominator effect: a change in a ratio caused by a change in how the bottom number is counted.
• PDP (product detail page): the page for a single product.
• Implied conversion: the conversion rate you get by dividing two reported numbers, used as a sanity check on both.

Frequently asked questions

Why did my ecommerce conversion rate drop suddenly?

The causes worth ruling out first are broken or changed tracking, a shift in traffic between devices or channels, a change in how sessions are counted, a site release, paid traffic landing on unavailable products, and promotions. Check those before assuming shoppers behaved differently.

How do I tell a tracking problem from a real drop?

Read related metrics together. If clicks fell sharply but revenue from the same placements barely moved, the implied conversion rate becomes impossible, which points to measurement rather than behavior.

Can conversion rate fall even if every segment improved?

Yes. If traffic moves from a high-converting device or channel to a lower-converting one, the blended rate falls even when each segment's own rate holds or rises.

Can a consent banner change affect conversion rate?

Yes. If a consent change reduces the sessions you record while orders stay the same, conversion rate rises on paper without any change in shoppers, and the reverse happens when it is removed.

How do out-of-stock products hurt conversion?

When paid campaigns or category pages send shoppers to items they cannot buy, those sessions convert at close to zero. At one large retailer this was 7% to 9% of product-page visits.

How fast should a conversion drop be explained?

Fast enough to act before the next review. At one large retailer, question-to-resolution went from about seven days to two once the investigation was run by an AI Operator.

What should I check first?

Whether the number is real. Then traffic mix, then counting changes, then releases.

Next step: If a conversion question has been open in your weekly review for more than a week, book a demo and bring it. We will show you how an Operator would work it.