Your conversion rate dropped. Here's the order to check things in

A conversion rate drop gets investigated backwards almost every time. Somebody notices the number, someone else proposes a redesign of the checkout, and three weeks later the real cause turns out to have been a consent banner change or a shift in where the traffic came from. The order you check things in matters more than how thoroughly you check any one of them, because the cheap explanations are also the likely ones.

First question: numerator or denominator?

Conversion rate is conversions divided by sessions. The rate can fall three ways: conversions fell, sessions rose, or both moved. Those are different investigations, and answering which one applies takes about two minutes.

• Conversions fell, sessions flat. Something broke, or demand changed. Genuine site or tracking investigation.
• Sessions rose, conversions flat. You did not get worse at converting; you acquired traffic that was never going to convert. This is the most common "drop" of all and the most misdiagnosed.
• Both fell, conversions faster. Usually a mix shift plus a real problem. Do the mix analysis first or you will chase the wrong half.

If sessions rose, stop and look at where they came from before touching anything on the site. A campaign launch, a broad match expansion, a piece of content that got picked up, or a bot wave can all raise the denominator without any human being becoming less likely to buy.

The signature table

Drops leave fingerprints. The shape of the drop — when it started, who it affected, whether it is confined — narrows the cause faster than any tool does. This is the table we work from.

Two rows in that table — the first and the last — account for a large share of the emergencies we get called into, and both are resolved by the same reconciliation: count the real orders or leads yourself and compare. Do that before you convene anybody.

Step 1: rule out measurement

Measurement goes first because it is the cheapest to check and, in 2026, the most likely to have silently changed. The mechanisms are well documented and they all fail quietly:

• Consent. If a visitor arrives on a paid click and does not grant consent immediately, the session's source can be lost and later re-attributed as direct. The banner still appears, the tag still fires, and the dashboard still looks green while attribution quietly degrades. A change to banner wording, position or default state is enough to move the numbers.
• Client-side blocking. Ad blockers stop tags for a substantial minority of desktop users — vendor estimates commonly put US desktop ad-blocker use around a third — and browser tracking prevention shortens cookie lifetimes further. None of this is new, but any change to how your tags load can change how much of it bites.
• Deploys and tag containers. A tag removed during a release, a duplicated container, or a conversion event renamed will produce a clean step change on a single date. That signature is nearly diagnostic on its own.

The ground truth test is always the same: your own record of orders, bookings or qualified enquiries, compared with what the analytics platform says for identical dates. If the two disagree, you have a measurement problem and there is nothing to fix on the site. We wrote about how this failed silently for merchants during Shopify's checkout extensibility deadline, where reporting broke without a single order being affected.

Step 2: check the traffic mix

If the counting is sound, the next most likely explanation is that you are converting the same as before, on a different audience. This is where the arithmetic surprises people, so here is the worked version.

Suppose desktop converts at 4% and mobile at 2%, and your traffic was an even split. Your blended rate is 3.0%. Now a campaign, a seasonal shift or a viral post moves the split to 35% desktop and 65% mobile. Nothing on the site changed. Neither segment got worse.

A 10% relative decline in the headline number, produced entirely by mix. The same pattern appears with geography, channel, new versus returning visitors and branded versus non-branded search. The rule: if every segment's rate is stable and only the blend moved, the site is not the problem and site changes will not fix it. What changes it is either acquiring a different mix or improving the weaker segment — which is a real project, and a different one from the emergency you thought you had. The mobile version of that project is covered in why mobile converts worse than desktop.

Two 2026-specific mix shifts are worth checking explicitly:

1. Automated traffic. Imperva's 2026 Bad Bot Report found automated traffic accounted for more than 53% of web requests in 2025 — roughly 40% bad bots and 13% good bots — with human traffic down to about 42.5%. Not all of it reaches your analytics, but a scraping wave or an AI crawler surge inflates sessions while contributing nothing to conversions. The same distortion wrecks experiments, which we covered in A/B tests that measure bots instead of customers.
2. AI assistant referrals. Visits arriving from AI answers behave differently from search visits: often better informed, often further along, sometimes arriving on a page chosen by a model rather than by intent. A growing share of these changes your blended rate in either direction, and it is worth segmenting rather than guessing.

Step 3: now look at the site

By this point the remaining candidates are genuine, and they are worth checking in a specific order because "what changed" is a much shorter list than "what could be wrong".

1. Your own deploys. Check the release log against the drop date. Obvious, and skipped surprisingly often.
2. Changes you did not deploy. Apps and plugins auto-update, payment providers ship their own releases, consent and analytics vendors update scripts from their own infrastructure, and Chrome has been on a two-week stable release cycle since 8 September 2026. A drop confined to one browser version is the signature here, and it is why segmenting client-side errors by browser is worth setting up before you need it. We wrote about this in why a code freeze does not freeze what breaks checkout.
3. Step-level funnel data. Completion by step, by hour, against the same hours the previous week. A single step collapsing is a specific, findable fault; a uniform decline across every step is usually mix or measurement, which sends you back to steps 1 and 2.
4. Payment method success rates, individually. One wallet or BNPL option failing while cards work is invisible in a blended number and is one of the most common provider-side faults.
5. Speed on the converting path. A third-party script that added several hundred milliseconds degrades conversion without ever throwing an error. Compare real-user performance before and after the drop date rather than running a synthetic test today.

When it is real and none of the above

Sometimes the counting is right, the mix is stable and nothing broke. The remaining explanations are commercial rather than technical: a competitor changed price or shipping terms, your promotion ended, demand moved seasonally, or the offer stopped being competitive. These are slower, broader and rarely start on a single date.

One benchmark worth holding onto for perspective: Baymard Institute's meta-analysis of 50 studies puts average documented cart abandonment at 70.22%. A high abandonment rate is the normal state of e-commerce, not evidence of a fault. The question that matters is always whether your number moved against your own baseline, segment by segment — which is why the baseline is worth maintaining while nothing is wrong.

The 60-minute triage

1. 0–10 min. Compare back-office orders or leads with analytics for the drop window. If they disagree, stop: it is measurement.
2. 10–20 min. Plot the drop by day. Single break point, or slope? A break point points at measurement or a deploy; a slope points at mix.
3. 20–35 min. Recalculate conversion rate within each channel, device and country separately. If every segment is stable, write "mix shift" on the board and stop the emergency.
4. 35–50 min. Check funnel step completion and payment method success individually.
5. 50–60 min. Check the release log, app and plugin update history, and client-side errors segmented by browser version.

Most investigations end in the first twenty minutes. The value of the sequence is that it ends them there, instead of a month into a redesign that was never going to change the number.

Frequently asked questions

Why did my conversion rate drop suddenly?

A sudden drop that starts on one date and affects every channel and device at once is almost always a measurement failure rather than a real change in behaviour — a deploy that removed a tag, a consent banner change, a republished tag container or a renamed conversion event. Confirm it in ten minutes by comparing your own order or lead count with what analytics reports for the same dates. If your back office is flat, nothing broke except the counting.

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

Reconcile against a source that does not depend on tags: orders in your e-commerce admin, bookings in your calendar, qualified enquiries in your CRM. Tracking problems show a gap between that count and the analytics count. Real problems show the drop in both. This single test resolves the majority of conversion rate emergencies before any site investigation starts.

Can a conversion rate fall even if nothing on the site changed?

Yes, and it is common. If desktop converts at 4% and mobile at 2%, moving the traffic split from 50/50 to 35/65 drops the blended rate from 3.0% to 2.7% with both segments performing exactly as before. The same effect occurs with channel, country and branded versus non-branded traffic. If every segment's rate is stable and only the blend moved, site changes will not recover the headline number.

Could bot traffic be lowering my conversion rate?

It can, by inflating sessions without adding conversions. Imperva's 2026 Bad Bot Report found automated traffic exceeded 53% of web requests in 2025, with about 40% classed as bad bots and human traffic down to roughly 42.5%. The signature is sessions rising sharply while conversions stay flat, usually with poor engagement metrics and an unusual geographic distribution.

Where should I look first when conversion rate drops?

In this order: measurement, traffic mix, then the site. Measurement first because it is the cheapest to rule out and the most likely to have changed silently. Mix second because it explains most gradual declines and cannot be fixed on the page. The site last, where the useful evidence is step-level funnel completion, payment method success rates by individual method, and client-side errors segmented by browser version.

How long should I wait before concluding my conversion rate really dropped?

Long enough for the comparison to be meaningful against a like-for-like period — same days of the week, same promotional conditions, and enough volume that a few conversions either way do not swing the percentage. Low-traffic sites see large swings from ordinary randomness. If a week-on-week change is within the range your site normally varies by, it is not yet a finding.

The takeaway

Diagnose in cost order, not in interest order. Reconcile your own order count against analytics first, because measurement changes silently and explains most single-day drops. Recalculate each segment separately second, because a stable segment rate with a moved blend is a mix shift that no site change will fix. Only then look at the page, and look at what changed rather than at what could be improved — your deploys, the updates you did not deploy, step-level completion, payment methods individually, and errors by browser version. The teams that lose a month to a conversion drop are almost never the ones who lacked tools. They are the ones who started with the redesign and worked backwards to the consent banner.