Five years of ad spend and a flat revenue line: where to look before you fire another agency

Three agencies in five years. Each one arrived with a deck, a keyword plan and a fresh set of creative. Each one moved cost per acquisition around for a quarter or two. Revenue stayed roughly where it was.

If that's close to your last few years, the conclusion you've probably landed on is that the marketing was wrong. Fair enough, and it's usually part of the answer. It's also incomplete, for a fairly boring reason: across all three engagements, one variable never changed.

The storefront.

Traffic is what gets tested. Audiences, channels, creative, bid strategy, landing page copy — all of it gets rebuilt every time a new team comes in. The theme, the catalogue structure and the page weight sit more or less where the last developer left them. Your site is the constant in every experiment you've run, which is exactly why nobody looks at it, and exactly why it's the place to look first.

"Your site is the constant in every experiment you've run — which is exactly why nobody looks at it."

Four things to check. None of them require firing anyone.

Why nobody re-tests the site

Two reasons, and neither is really anyone's fault:

• Cost asymmetry. Changing the ad account is a Tuesday afternoon. Changing the product template is a project with a scope, a quote and a QA round. Given that choice every quarter for five years, everyone keeps changing the Tuesday thing.
• The wrong reviewer, on the wrong device. A founder, a designer and an account manager look at the site on large monitors, on fast connections, with the cache warm and an ad blocker running. The customer looks at it on a phone, on mobile data, arriving cold from a search result. Those are two different websites.

So the ad account gets five years of iteration, the storefront gets none, and the flat revenue line gets blamed on the only part anyone was actually working on.

The mobile buying experience

Published Australian benchmarks put mobile at roughly 60 percent of web traffic in 2026, with mobile conversion running at around half the desktop rate. Some of that gap is genuine behaviour — people research on a phone and buy at a desk. Not all of it is, and the part that isn't is yours to fix.

The test takes two minutes. Pick up your own phone, turn the ad blocker off, and get to a top-selling product page the way a customer would, through a search result or an ad rather than a bookmark. Then measure how far you scroll before the add-to-cart button is on screen.

On a 390 by 844 viewport, which is an iPhone 12 through 15 and the width worth designing to, plenty of stores land somewhere past 1,400 pixels. That's about two full screens of hero image, breadcrumb, gallery, review carousel and variant swatches before the customer can do the one thing they came to do.

They are not reading all of that. They're scrolling, deciding it's taking too long, and going back to the results page.

"None of it shows up in your ad account, because the session looks identical to a badly targeted click."

Catalogue structure

Open your product list and read fifty titles the way a stranger would.

The pattern we see most often is titles that describe the pack rather than the product. “Value Bundle 3-Pack” tells a customer nothing about what's inside it, who it suits, or why it beats the thing sitting next to it. It tells a search engine less. And it tells an AI shopping assistant almost nothing, which matters more every quarter, because that assistant builds its shortlist from your title, description and structured data. It never sees your photography.

The structured-data description field has the same problem in a different shape. On most stores it's a packing list: dimensions, weight, materials, what's in the box. Genuinely useful at the point of purchase. Useless as a reason to be recommended, which is what that field is now doing.

This work is unglamorous and it does not need a developer. It needs someone who knows the products to sit down with a spreadsheet and rewrite titles, categories and attributes in priority order, best sellers first. Agencies rarely propose it because it doesn't bill like a project. That's a reason it gets skipped, not a reason it doesn't matter.

Page weight, and which numbers matter

The published thresholds are Largest Contentful Paint under 2.5 seconds, Interaction to Next Paint under 200 milliseconds, and Cumulative Layout Shift under 0.1. Each is judged at the 75th percentile of real visits, so the number that counts is the experience of your slowest quarter of customers, not the one you get on office wifi.

Two things account for most of the damage on a typical Shopify homepage:

• Oversized images. A 750 pixel image dropped into a 285 pixel slot, downloaded in full over mobile data. That's a different fix from “we compressed the images”, and usually a bigger one.
• Third-party scripts. Widgets pulling in their own stylesheets and fonts ahead of your content — which is how a support chat widget ends up being the slowest thing on the page.

The trap here is chasing a green score. A customised theme carrying a real app stack will not go green, and buying the last few points gets expensive quickly. Chase the two numbers that map to money: time until the main image is on screen, and time until the buy button actually responds to a tap. The composite score is a vanity metric.

What your site tells machines about itself

Shopify has handled most of the platform side of agentic commerce. Agent tools, schemas, the machine-readable plumbing. What breaks is the theme sitting on top of it.

Landmark roles applied to the wrong elements. A logo link with no accessible name. Form fields with no label attached to them. Layout that keeps moving while the page loads. Individually each one reads as accessibility housekeeping and gets deferred. Together they mean an AI agent can land on a page that has everything it needs to complete a purchase and still not be able to interpret it.

That last one costs you twice, because layout instability is also why customers tap the wrong thing. Someone reaches for a button, the hero image finishes loading, everything below it shifts down, and they land on a newsletter signup instead. That is the Cumulative Layout Shift number described in human terms.

None of this says your agencies were wrong. Most were probably right about the thing they were hired to look at, and diagnosis was never the hard part anyway.

The argument is narrower than that. If revenue has been flat across three genuinely different traffic strategies, the constant is more interesting than the variables, and the constant is the site.

So do one thing this week before you brief anyone else. Open your best-selling product page on your own phone, ad blocker off, arriving from a Google result rather than a bookmark. Count the screens of scrolling before the add-to-cart button appears. Write the number down.

If it's more than two, you've found something no media plan is going to fix.

At beCommerce, we help Australian Shopify stores find out what's actually happening between the ad click and the add-to-cart tap — the mobile experience, the catalogue structure and the Core Web Vitals that never show up in a campaign report. If you want a second pair of eyes on your storefront before you brief another agency, let's talk.