Most e-commerce sites take three quarters of their visits from phones and convert them at roughly half the desktop rate. Contentsquare's 2026 Digital Experience Benchmark, built on 99 billion sessions across more than 6,500 sites, puts desktop conversion at 3.4% and reports that as 74% higher than mobile web. In Turkey the exposure is larger: StatCounter recorded mobile at 75.9% of web traffic in August 2026.
That gap usually lands on a dashboard as a mandate to redesign the mobile experience. Often it is not an interface problem at all. This guide is for the analyst or product owner who has to decide what to fund. Work the six gates below in order, stopping at the first condition that is met. You will finish knowing whether the gap is a measurement artifact, a traffic-mix effect, a delivery problem or a real interface failure, and which funnel step it sits on.
The six gates, in order
Each gate has a stop condition. If it is met, the gap is explained and you stop. Working them out of order is how a team redesigns a product page to fix a tagging bug.
| Gate | Question | Stop condition |
|---|---|---|
| 1 | Is the number real? | Raw recomputation closes half the gap |
| 2 | Are these the same people? | Like-for-like gap under ~15% relative |
| 3 | Is it delivery, not design? | Mobile fails two vitals, desktop passes |
| 4 | Where does the gap open? | One step carries most of the gap |
| 5 | Which surface is failing? | Any surface scoring zero there |
| 6 | Is the fix worth building? | Gain below cost to ship |
Gate 1: confirm the number before you explain it
Device-split conversion in a standard analytics report is not raw data. Under consent mode, Google models users who declined analytics storage, subject to thresholds: at least 1,000 events a day with analytics_storage='denied' for seven consecutive days, and 1,000 daily consented users on seven of the previous 28. Modelling covers users, sessions and key event rates, but not event counts and not the BigQuery export. A modelled report and your raw export can therefore disagree about the same week, and neither is broken. Consent rates also differ by device, so the effect lands unevenly on your comparison.
Three checks, in order:
- Recompute the gap from raw events or the back-office order table, not the modelled report.
- Confirm both device groups count the same conversion event at the same point, with app sessions and tablets excluded from both.
- Reconcile orders by device against the order table. A discrepancy on one device only is a tagging fault, not a UX finding.
Stop if the raw recomputation closes more than half the gap. Your problem is reporting.
Gate 2: check whether the two populations are the same people
The device dimension is session-scoped. Someone who browses on a phone at lunch and buys on a laptop that evening is two sessions, and without User-ID two users: a mobile non-converter and a desktop converter. That one behaviour manufactures a device gap out of nothing. GA4's blended reporting identity evaluates user ID, device ID and modelled data, but data collected before User-ID was implemented is not reprocessed, so you cannot answer this about last quarter by turning it on today.
Traffic mix does the same damage from the acquisition side. Paid social skews mobile and cold; branded search, email and direct skew desktop and toward people who already decided. Recompute the gap inside a single channel, then a single campaign type, new and returning separated. If it narrows sharply, the work belongs to media buying.
Stop if the like-for-like gap is under roughly 15% relative. A residual that size is device ergonomics, not a redesign programme.
Gate 3: rule out delivery before you blame design
Core Web Vitals thresholds are LCP at 2.5 seconds or less, INP at 200 milliseconds or less and CLS at 0.1 or less, each assessed at the 75th percentile and segmented separately for mobile and desktop. That segmentation is the point: compare your mobile 75th percentile against your own desktop one, not only the pass mark.
Use field data, not a lab run on office wifi. A mobile LCP of 4.1 seconds against a desktop 1.9 is a delivery gap that will imitate a UX gap in every session replay you watch. People rarely describe a slow page as slow; they describe it as confusing.
Stop if mobile fails two of three vitals in the field while desktop passes all three. Fix delivery, wait for field data to refresh, then re-measure before touching layout.
Gate 4: find the step where the gap opens
A device-level gap averages five or six step transitions and is almost never spread evenly. Build a per-step continuation table and compute a mobile-to-desktop ratio for each step. The numbers below are illustrative:
| Step transition | Desktop | Mobile | Ratio | Mobile sessions at step |
|---|---|---|---|---|
| Session to product page | 62% | 58% | 0.94 | 420,000 |
| Product page to cart | 11% | 10% | 0.91 | 243,600 |
| Cart to checkout start | 68% | 64% | 0.94 | 24,360 |
| Checkout start to payment | 77% | 48% | 0.62 | 15,590 |
| Payment to order | 91% | 88% | 0.97 | 7,483 |
Two reading rules. The step carrying the gap is the one with the lowest ratio, and any step with too little traffic is ignored however bad its ratio looks. Above, one transition carries nearly all the gap and still sees over 15,000 mobile sessions a month.
If every ratio sits between 0.90 and 0.97 you do not have a leak. You have a uniformly harder surface, which is slower and costlier to fix, and Gate 6 usually says not to fund it.
Gate 5: score the three surfaces that fail most on phones
Baymard's mobile research covers more than 20,000 hours of usability testing and a benchmark of 138 large mobile sites, of which 62% rate mediocre or worse and none rate good. Assume your site has issues; this gate decides which one sits on the leak step. Score each row 0 (absent), 1 (partial) or 2 (done properly), for that step only.
| Surface | What earns a 2 | Score |
|---|---|---|
| Input and keyboards | Correct inputmode and autocomplete on every field, no zoom on focus, single column, inline errors visible without scrolling on a 390px viewport | 0 / 1 / 2 |
| Selection and choice | Variant selection reachable without pinch, price and delivery promise visible at the decision point, persistent primary action | 0 / 1 / 2 |
| Payment and trust | Wallet and saved-card options before manual card entry, guest checkout available, full total including shipping shown before the payment step | 0 / 1 / 2 |
Anything scoring zero at the leak step is your first fix. Payment and trust earns its own row because the number underneath is already large: Baymard's rolling average of 50 studies puts documented cart abandonment at 70.22%. A step most people leave anyway responds well to friction removal and badly to persuasion.
One regional adjustment: if you sell in Turkey, instalment payment is part of how a card purchase gets evaluated, not a payment-screen detail, so where instalment information appears is a product-page decision. No generic mobile checklist raises that.
[INTERNAL DATA NEEDED: which of these three surfaces carries the leak most often across Switas mobile audits, split by e-commerce and travel accounts]
Gate 6: size the fix before you build it
Take the leak step's mobile sessions, the ratio deficit and average order value, and calculate the annual value of closing half the deficit. Half, not all: desktop is not the ceiling for a phone, and treating it as one is how these projects miss.
Then choose ship or test. A defect, a compliance fix or a missing wallet option ships untested, because there is no plausible world where the broken version wins. A design opinion gets tested, and sized before it starts. At 15,000 monthly sessions you can detect a large effect and nothing subtle, which is itself useful: make one substantial change, not four small ones.
[INTERNAL DATA NEEDED: Switas benchmark for realistic recovery of a device ratio deficit after one mobile checkout fix, to replace the 50% planning assumption]
Where this breaks down
- Low volume. Below a few hundred mobile orders a month the ratios are noise. Run moderated sessions on a phone.
- Apps. This is web diagnostics; in-app funnels need their own instrumentation.
- Considered purchases. In travel, health and B2B a long cross-device path is structural, and the honest mobile metric is a micro-conversion such as a quote request.
- Benchmarks are not targets. Cross-industry aggregates say whether your gap is unusual, not what it should be.
- Everything rests on Gate 1. If the conversion event does not fire reliably on both devices, the other five gates are decoration.
Frequently asked questions
What is a normal mobile-to-desktop conversion ratio?
In the Contentsquare 2026 benchmark desktop converts at 3.4%, reported as 74% higher than mobile web, putting the typical ratio near 0.57. Baymard describes many mobile sites as converting less than half as many visitors as desktop. That range is ordinary, not alarming.
Is "mobile converts 74% lower" the same claim?
No. Desktop being 74% higher than mobile means mobile is about 43% lower than desktop. Reversing the base of a relative comparison inflates the gap by more than half and changes what size of fix looks justified.
Does consent mode make device-level data unusable?
No, but it makes the standard report a modelled estimate rather than a count. Modelling covers users, sessions and key event rates, not event counts or the BigQuery export. Reconcile against raw data first.
How do I separate a cross-device effect from a mobile UX problem?
Implement User-ID and compare logged-in users only, where one person is one user across devices. If that gap is much smaller than the site-level one, you were looking at cross-device behaviour. It works going forward only.
Is Core Web Vitals a ranking question or a conversion question here?
Treat it as a conversion question. Gate 3 uses the thresholds as a device comparison, not an SEO checklist. A slow mobile page produces behaviour that looks like confusion in session replay.
How much traffic do I need before per-step ratios mean anything?
Several hundred mobile sessions a month per transition at minimum, preferably thousands at the step you intend to change. Below that the table is a hypothesis generator, to be validated qualitatively.
Run it on your own funnel
Take your last complete month and work the gates in order. If you would rather have the scored worksheet, the leak step and the sized fix handed to you, talk to us about a mobile UX audit.
Sources
- Contentsquare, mobile analytics statistics (2026 benchmark)
- Contentsquare, Digital Experience Benchmark 2026
- StatCounter, platform share in Turkey, August 2026
- Baymard, mobile e-commerce usability benchmark
- Baymard, cart abandonment rate statistics
- web.dev, Web Vitals thresholds
- Google, behavioural modelling with consent mode
- Google, User-ID and cross-device reporting
