Conversion
Your Shopify conversion rate dropped. Here's how to find out why.
Most conversion-rate drops are not conversion problems. Before you touch the checkout, work through the diagnostic order that separates a real decline from a traffic-mix change, a seasonality artefact, or ordinary noise.
August 4, 2026 · 13 min read
Diagnose a conversion-rate drop in a fixed order: confirm the drop is larger than normal variation, check whether traffic mix changed, isolate which segment moved, and only then examine the funnel. Most apparent drops are a change in who is visiting rather than a change in how well the store sells, and fixing the checkout when the cause was a traffic-mix shift wastes the week.
Conversion rate fell from 2.4% to 1.9% and the instinct is immediate: something broke in checkout. Sometimes it did. More often the store sells exactly as well as it did last month to the people who were already going to buy, and the denominator changed underneath you.
This is a diagnostic order rather than a list of fixes. Each step either explains the drop or rules out a cause, and doing them out of order is the reason so much conversion work produces no measurable result.
Step zero: is this even a drop?
Conversion rate is a ratio of two numbers that both vary, which makes it noisier than either. A store doing three hundred sessions a day has a conversion rate that swings meaningfully on ordinary randomness — and the smaller the store, the more of any given move is noise.
Two disciplines make this tractable. First, compare like with like: the same weekday, against a trailing median of several recent instances of that weekday, rather than against yesterday or against last week's average. Tuesday and Saturday are different businesses in most stores.
Second, use a median rather than a mean for the baseline. A single day with a bulk order or a viral post will drag a mean far enough to make the following normal week look like a collapse.
If the move is inside normal variation, stop. The correct action is to do nothing and keep watching, which is unsatisfying and is also the action that does not waste a week.
Step one: did the traffic mix change?
This is the highest-yield question and the most frequently skipped. Conversion rate is orders divided by sessions. If you add sessions that were never going to convert, the rate falls even though every existing visitor behaves identically.
The usual sources of a mix shift:
- A new campaign optimising for clicks or reach rather than purchases, delivering high volume at low intent.
- A post, video or press mention driving curiosity traffic — people looking at the thing, not shopping for it.
- Broadened geographic targeting into markets where you do not ship, or where shipping cost makes purchase irrational.
- A shift in device mix, since mobile converts lower than desktop in most stores; more mobile traffic lowers the blended rate without either segment changing.
- Bot or crawler traffic newly counted as sessions, inflating the denominator with nobody attached.
The test: hold the mix constant. Compute conversion rate per channel, per device and per country for the drop period and the baseline period. If each individual segment held steady but the blended rate fell, the mix moved and your store did not get worse. That is a media question, not a merchandising one.
Step two: which segment actually moved?
If the mix is stable, the drop is real and located somewhere specific. Slice in this order, because each is progressively less likely and more expensive to investigate:
| Segment | If the drop is concentrated here | Likely cause |
|---|---|---|
| One device type | Mobile only, or one browser only | A rendering or interaction bug from a recent theme or app change |
| One country | Single market fell | Shipping cost or availability change, currency display, a payment method disappearing, or a local competitor promotion |
| One channel | Paid social fell, others flat | Creative or audience change upstream, not a site problem |
| New vs returning | New visitors fell, returning steady | A first-impression problem: landing page, trust, price presentation |
| One landing page | A single page collapsed | A page-level change: a broken element, a price update, an out-of-stock hero product |
| One product | Individual product fell | Stock status, a review appearing, price change, or a competitor undercutting |
| Everywhere, evenly | All segments fell together | Checkout, payment provider, or a sitewide script error — now check the funnel |
The last row is the only one that justifies going straight to checkout, and it is the least common. A genuinely broken checkout produces an unmistakable pattern: it fails uniformly across segments and usually starts at a precise moment rather than drifting.
Step three: what changed, and when exactly?
Once you know where the drop lives, find the change that coincides with it. Get the start time as precisely as your data allows — an hour, if possible, not a week — because a precise onset time converts a list of suspects into one.
- 1
Pin the onset
Plot the affected segment hourly or daily and identify the first period that broke pattern. A sharp cliff means a discrete change — a deploy, an app install, a setting. A gradual slope means something accumulating: inventory depletion, price drift, competitive pressure, seasonality.
- 2
Check the change log around that moment
Theme publishes, app installs and uninstalls, price or inventory bulk edits, shipping rate changes, payment provider settings, discount codes expiring. Shopify records much of this; your team's calendar records the rest.
- 3
Reproduce on the affected segment
If mobile Safari fell, complete a purchase on mobile Safari yourself. This finds a surprising share of problems in minutes — a cookie banner covering the buy button, a payment method not rendering, a form field refusing input.
- 4
Check stock on your top sellers
A bestseller going out of stock lowers overall conversion rate without any page changing. It looks like a site problem and is an inventory problem. Check the products that carry a disproportionate share of your orders first.
- 5
Check what shipping now costs at checkout
Shipping surprise is a leading cause of abandonment, and rate changes often arrive from a carrier app rather than from a person. Compare the rate a customer sees today with the one they saw before the drop.
Step four: the funnel, finally
Only now is funnel analysis worth the time, because you know which segment and roughly when. Compare the affected segment's step-to-step rates against its own baseline — not against a benchmark from another store, which tells you nothing about your change.
The stage where the drop concentrates points at the cause: fewer product-page views from collections suggests merchandising or stock; fewer add-to-carts suggests price, reviews, or a product-page element; fewer checkout starts suggests cart friction or a shipping estimate; fewer completions suggests payment, address validation, or a required field newly failing.
Then fix one thing. Fixing three simultaneously means the recovery cannot be attributed and you will not know what to do the next time this happens.
Verifying the fix
A fix is not proven by the number recovering the next day. Traffic varies, and the day after an intervention is exactly when you are most inclined to see what you hoped for.
The defensible version: record the affected segment's baseline before you change anything, apply one change, exclude the day of the change from measurement because it is contaminated by partial exposure, then compare a fixed window after against the same-weekday baseline before. State the result as a range rather than a point, and say plainly if the observed change is within normal variation.
This is the loop Nomu automates — capturing the before-state when a fix is applied, waiting out the contaminated period, then re-measuring the affected segment against its own seasonality-aware baseline and reporting the delta with the arithmetic attached. But the loop is what matters, not the tool. A fix with no before number is a story.
Frequently asked questions
What is a good Shopify conversion rate?
The benchmark question is less useful than it looks, because rates vary enormously by category, price point, traffic mix and market. A high-consideration product at a high price converts far lower than an impulse purchase and may be far more profitable. Compare against your own trailing baseline, segmented, rather than against a cross-industry average.
How big a drop is worth investigating?
It depends on your volume, because smaller stores have noisier rates. The practical rule is to compare against the variation you normally see on the same weekday: if the move is inside your usual range, watch it; if it is clearly outside and persists for several days, investigate.
Could it be seasonality?
Frequently, and it is under-diagnosed outside the obvious Q4 peak. Regional calendars matter enormously — Ramadan and Eid shift both traffic volume and purchase behaviour in ways that make a naive month-over-month comparison meaningless in affected markets.
Should I run an A/B test to fix it?
Not as the first move. Testing is for optimising something that works; a drop is a regression to be diagnosed. Find what changed first — testing your way back to a previous state is slow and expensive when reverting the change is available.
My conversion rate dropped but revenue is flat. Does it matter?
Usually not, and this is the clearest sign of a mix change. More sessions at a lower rate producing the same orders means you bought traffic that does not convert. The question becomes whether that traffic was worth its cost, which is a media efficiency question rather than a conversion problem.
How long should I wait before concluding a fix worked?
Long enough to cover a full weekly cycle at minimum, so weekday effects average out, and longer if your daily volume is low. Excluding the day of the change is important — it contains both pre-fix and post-fix traffic and belongs to neither period.
The verdict
Most conversion-rate drops are not conversion problems, and most conversion work fails because it started at step four. The order matters more than the tactics: confirm the drop is real against a same-weekday median, check whether the mix moved, isolate the segment, find the coinciding change, and only then open the funnel.
Worked in that order, a large share of drops resolve within an hour and turn out to be an out-of-stock bestseller, a campaign change, a shipping rate, or a theme update that broke one browser. None of those are fixed by redesigning the checkout.