What an ecommerce growth agency is actually selling you

I am Moeez Abbas. I work on my own, remotely from Lahore in Pakistan, for businesses in the United Kingdom and the United States. There is no office anywhere and no team. The phrase “ecommerce growth agency” is the vaguest label in this industry. It survives because it is unfalsifiable: growth is a direction, not a deliverable, and a page that promises it has promised nothing a client can later hold anyone to.

So the argument of this page is simple. The phrase only earns its keep when it is attached to named mechanics — a specific attribute in a specific feed specification, a specific HTTP status code on a specific class of URL, a specific statutory instrument that constrains what a price may say. Where a number exists below it is sourced; where it does not, the point is made qualitatively rather than invented.

The honest framing of the market: the Office for National Statistics put online sales at 28.8% of all retail sales in Great Britain in August 2026, up from 28.4% in July 2026, in its Retail Sales Index bulletin — the continuous series is J4MC, so anybody can check it. That is the scale of the channel. It is not evidence that any particular shop will grow.

The businesses that come to a page like this sit in one of four groups: direct-to-consumer brands on Shopify or WooCommerce; mid-market multi-channel retailers on BigCommerce, Adobe Commerce, Magento Open Source or Shopware, running acquisition through Google Merchant Center and a Meta catalogue; marketplace-dependent sellers on Amazon Marketplace, eBay and TikTok Shop trying to build demand they own; and cross-border sellers exposed to the UK £135 consignment rule, the European Union’s Import One Stop Shop, and now a per-item customs duty on low-value parcels.

Each of those four has a different failure mode. The rest of this page is organised around the four places online revenue actually leaks — the product feed, the category page, the product detail page and the checkout — plus the regulatory layer that decides what may legally be advertised at all.

Google publishes a product data specification precise enough that feed work stops being a matter of opinion. Seven attributes are required for every product: id, title, description, link, image_link, availability and price. Three more are conditional and routinely misunderstood: condition only for used or refurbished items, brand for new products excepting movies, books and musical recordings, and mpn only where the product has no manufacturer-assigned GTIN — which is why identifier_exists is there for genuinely unbranded or handmade stock.

The limits are equally concrete. id is capped at 50 characters, title at 150, description at 5,000. price must be submitted in ISO 4217 format, so a UK feed states GBP explicitly rather than leaning on an account default — a failure that surfaces when a store switches on multi-currency selling through Shopify Markets and the feed stops matching the landing page. link must use http or https.

That 150-character ceiling is why feed title templates are engineered to a budget rather than written freehand: brand, product name, key attribute, size and colour all have to fit, and a supplier dump that concatenates everything gets truncated or rejected. The same discipline applies to google_product_category versus product_type: the first is Google’s taxonomy, the second is yours, and only the second is safe for campaign segmentation alongside custom_label_0.

One dated deadline belongs in a client’s calendar now: Google will enforce a minimum image size of 500×500 pixels for image_link and additional_image_link, across all product categories, from 31 January 2027, having begun surfacing warnings on undersized images inside Merchant Center during 2026 ahead of enforcement. Any retailer still serving legacy 300-pixel thumbnails must re-render its library or lose Shopping eligibility. That sits alongside the longstanding prohibition on promotional overlays, watermarks and borders, so assets with a burnt-in sale badge cannot be reused in the feed.

Variants are the other reliable source of confusion. item_group_id ties a colourway or size run together, availability_date handles pre-orders, sale_price with sale_price_effective_date handles promotions without rewriting price, and unit_pricing_measure matters for anything sold by weight or volume, where the Price Marking Order 2004 also has views. additional_image_link and shipping are optional in the specification and near-mandatory in practice.

One piece of plumbing has already moved. Merchant API became generally available in August 2025 as Google’s primary programmatic interface, superseding Content API for Shopping, which shut down on 18 August 2026 — so any middleware or ERP bridge still written against it is already broken rather than heading for a deadline. Stores that maintain products only by manual file upload, scheduled fetch or a Google Sheet were never affected by that switch. That is definable work, unlike a proposal offering a “Merchant Center Next migration”: Google upgraded all retailers to Merchant Center Next by September 2024 and has since dropped the suffix.

The September 2026 agentic commerce updates — AI performance insights in Merchant Center, the Universal Commerce Protocol checkout tools and the new feed guidance — are explained in the Google agentic commerce update guide.

Google’s guidance on faceted navigation is unusually prescriptive. Where faceted URLs do not need indexing, the preferred remedies are to disallow them in robots.txt while still allowing individual item pages, or to use URL fragment identifiers such as #products=fish&colour=green instead of query parameters such as ?products=fish&colour=green.

The alternatives are explicitly weaker. Google calls rel=canonical a less effective option that may only reduce crawling of non-canonical faceted URLs over time, and notes rel=nofollow has to be implemented on every relevant link to do anything. Plenty of UK ecommerce sites rely on canonicals alone, and Google’s own documentation says that is the weaker choice — more persuasive in a development sprint than a consultant’s preference.

Where faceted URLs must be indexable — and for some retailers a small set genuinely should be, because “waterproof walking boots size 9” is a real demand pattern — Google asks for three things. Standard & separators rather than commas, semicolons or brackets. Consistent filter order, so ?colour=blue&size=m and ?size=m&colour=blue do not become two crawlable URLs for one result set. And an HTTP 404 status code when a filter combination returns no results — not a redirect, not a soft 404, not an empty page with a 200.

That last instruction conflicts with the default behaviour of many platforms, which render an empty results page and return 200 — one of the few defects demonstrable on a live site within minutes, using nothing but a browser and the response headers.

Two adjacent wastes get missed by audits that look only at filters. Sorting parameters such as ?orderby= and ?sort= produce reordered duplicates of one item set and consume crawl budget. And infinite scroll on app-heavy themes is a crawling concern rather than a design preference: content loaded only on user interaction may not be reachable by Googlebot Smartphone unless paginated URLs also exist, which can leave a large catalogue’s long tail undiscoverable. Parameter-order duplication is invisible in an XML sitemap audit and obvious in server logs, which is why log file analysis belongs in the audit rather than the optional extras. A noindex robots meta tag, a correctly scoped hreflang set and a clean canonical policy finish what robots.txt starts.

Interaction to Next Paint is the Core Web Vital a transactional page is most exposed to, because a product page is very little except interactions, and unlike most speed advice it comes with hard thresholds: good is at or below 200 milliseconds, needs improvement is above 200ms and at or below 500ms, poor is above 500ms. That gives a product page a pass mark rather than an instruction to be faster.

The reason a site that comfortably passed First Input Delay can now fail is structural. INP observes all interactions made with a page during a visit and measures from the moment a user initiates one until the next frame is painted; FID measured only the input delay of the first interaction. INP became a Core Web Vital in March 2024 and FID was fully deprecated in September 2024, so every variant picker, filter widget, accordion, quantity stepper and add-to-cart handler is now in scope.

On an app-heavy store the long interactions are rarely in the theme. They are in the reviews widget, the wishlist app, the size-guide modal, the upsell recommender, the live-chat loader, the consent banner and the Google Tag Manager container that fires four vendors on click. Diagnosing that means attributing long tasks to the script that owns them, not buying a faster host.

There are two legitimate ways to measure and they routinely disagree. Field data comes from Real User Monitoring or the Chrome User Experience Report, reflecting real devices on real connections. Lab data comes from PageSpeed Insights or Lighthouse, or from simulating interactions in a controlled run. The web-vitals JavaScript library handles the edge cases hand-rolled measurement gets wrong.

The other vitals still matter and are easier to move. Largest Contentful Paint on a product page is usually the hero image, which means image format, dimensions and fetch priority rather than heroics. Cumulative Layout Shift is usually unreserved space for a badge, a banner or a lazily injected review star row. Time to First Byte exposes hosting and template rendering. Total Blocking Time and First Contentful Paint are useful lab proxies while field data refreshes.

On the paid side the mechanics are equally nameable. A Meta catalogue in Meta Commerce Manager underpins Advantage+ catalogue ads and Dynamic Product Ads, and it can be fed from the same product feed that serves Google Merchant Center — though Meta’s required fields differ from Google’s product data specification, so reusing one source needs mapping rules rather than a toggle.

The Meta Conversions API sends events server-to-server instead of relying solely on the browser-based Meta Pixel — the documented response to signal loss from Apple’s App Tracking Transparency and browser cookie restrictions, and unlike most tracking work it produces a number you can check. Event Match Quality rates, out of 10, how effectively customer information sent from a server matches event instances to a Meta account, based on which parameters arrive, their quality, and the share of events matched. The Dataset Quality API scores datasets on the same scale with named bands: under 4 is poor, 4 to 5.9 is OK, 6 to 7.9 is good, 8 or above great. A before-and-after on those bands is a deliverable. “Improved tracking” is not.

Why Ads Manager and the shop ledger never agree

Meta’s default attribution setting is 7-day click plus 1-day view. That single fact explains most of the gap between platform-reported revenue and what Shopify or WooCommerce actually banked, before any tracking fault is considered. Attribution platforms report modelled and claimed conversions, not observed causal effect, which is why platform-reported return on ad spend routinely exceeds blended return calculated as ledger revenue over total spend. The attribution settings exposed in Meta Ads Manager are worth pinning explicitly rather than inheriting, and the pixel and Conversions API events in Events Manager are worth checking for duplicates before any number is trusted.

The practical arrangement is to run blended figures from the ledger as the decision number, use in-platform figures for comparing campaigns against each other only, and treat lift tests as the only real answer to causal questions.

Consent sits underneath all of it. UK GDPR, the Data Protection Act 2018 and the Privacy and Electronic Communications Regulations 2003 require consent for non-essential cookies including advertising and analytics tags, and Google Consent Mode v2 is required for advertisers serving UK and EEA users who want full Google Ads and Google Analytics 4 functionality. A banner that blocks tags without signalling consent state silently degrades remarketing audiences and conversion reporting at once, which then gets misread as a media problem. Enhanced conversions, a server-side Google Tag Manager container and a Conversions API Gateway are the usual remedies, each with a real implementation cost worth quoting. Standard Shopping campaigns still have a place alongside Performance Max where a retailer needs control over which products get pushed.

Selling across a border changes the arithmetic of a basket, and the thresholds are specific enough to plan around. Take the UK inbound rules first, because they determine what a checkout must display and therefore what the feed’s price should contain.

1. For goods sold to UK customers, the £135 limit applies to the total consignment imported, not to individual items within it — basket-level calculation is what catches out sellers whose average order value straddles the line.
2. For consignments at or under that limit, the seller charges and accounts for UK VAT at the point of sale, unless the customer is a business supplying a UK VAT registration number, in which case the reverse charge applies and the buyer accounts for it.
3. Above it, ordinary UK VAT and customs rules apply at importation. A seller whose bundling or cross-sell pushes a consignment over the threshold may become liable for import VAT and customs duty and must adjust VAT already charged, so an upsell widget can create an unbudgeted duty liability.

Outbound to the European Union, the Import One Stop Shop covers goods imported from outside the EU in consignments valued under €150, applied per consignment. Above the €150 consignment threshold IOSS cannot be used and normal import VAT and customs procedures apply. A seller based in Great Britain cannot register for IOSS directly and must appoint an EU-established intermediary to register the business and handle monthly filings — a real cost to budget before any EU expansion campaign.

Then the change that many UK brands have not modelled. From 1 July 2026 the EU abolished the customs duty exemption for low-value imports, and Council Regulation (EU) 2026/382 introduced a €3 customs duty on goods in consignments worth €150 or less, applying regardless of VAT scheme — IOSS, the Special Arrangements or standard VAT alike. Read the charging rule carefully, because it is the part that gets mis-briefed: the €3 is levied per item rather than once per parcel, so a box of five identical shirts carries one charge while a box holding a shirt and a watch carries two. It is a transitional measure, set to run until the permanent regime under the EU customs reform package takes over in July 2028, when standard tariff rates are due to apply to imports regardless of value. A per-item charge on cheap multi-item baskets changes their unit economics outright, and the right response is usually a change to product mix, bundling and minimum order value rather than to ad budget.

IOSS continues to operate for VAT collection on business-to-consumer imports up to the same €150 level and coexists with the new duty obligations, so VAT and duty are two separate workstreams. It is easy to assume an IOSS registration covers duty. It does not, and surprise charges at the door generate refund requests and bad reviews that no amount of creative fixes.

The adjacent machinery is where the operational work lands: an EORI number, correct commodity codes, the Customs Declaration Service, Postponed VAT Accounting, VAT Notice 741A from HM Revenue & Customs, the choice between Incoterms DDP and Incoterms DAP, the Union OSS scheme and the Non-Union OSS scheme under the EU VAT e-commerce package, the Union Customs Code with its Implementing Act and Delegated Act, the EU General Product Safety Regulation, and the EU Digital Services Act for anyone running a marketplace rather than a shop.

My published figures are these. SEO packages start at £299 a month on the pricing page; monthly retainers are £650, £1,200 or from £4,000; one-off projects are listed on the packages page, from £650. For US clients the retainer equivalents are $800, $1,500 and from $5,000. There is no percentage of revenue.

What the tiers buy differs in scope rather than in effort per pound. At the lower monthly figure the scope is narrow and sequential: feed hygiene and Merchant Center diagnostics cleared, then structured data on product templates, then one class of crawl waste removed. At the middle figure technical work runs alongside paid management and measurement, including the Conversions API and Consent Mode work described above. At the top figure the scope can carry a multi-market build, an EU compliance workstream and continuous testing at once. The one-off project is a defined audit or implementation with a written end state, often the right first purchase precisely because it ends.

The only result I can point to is small and I will not dress it up. An independent UK artist spent £25 on Meta Ads and saw more than 500 Spotify saves and ten new followers over five days, verified in a five-star Fiverr review. That is one small result, on a tiny budget, over a very short window, in a category unrelated to retail — not a benchmark, not a forecast, and not evidence about your catalogue.

Who should not hire anybody yet

• Anyone whose average order value, gross margin per order and returns rate are not known. Without those three nobody can tell whether a campaign is profitable, and the agency becomes a spender rather than an operator.
• Anyone whose checkout is non-compliant on price disclosure. Fix the drip-pricing exposure before paying anybody to send more traffic into it.
• Anyone mid-migration, whether replatforming from Magento Open Source to Shopify or switching a BigCommerce product-URL setting. Migrate, stabilise, then optimise.
• Anyone whose stock feed is unreliable. Marketing an out-of-stock catalogue generates refunds and price-and-availability disapprovals at the same time.
• Anyone with fewer than about twenty saleable lines and no repeat-purchase mechanic. The leverage is in range and retention first, and a mailing list in Klaviyo costs less than a retainer.

A defensible engagement begins with what is measurable, names the mechanic it intends to change, and states how you will know whether it worked. Everything else in this industry is vocabulary.

Send the shop address and, if you can, read-only access to Merchant Center and analytics. You get a written answer naming which of the four leak points is costing you most right now, what the evidence for that is, and what I would fix first. It costs nothing and there is nothing to sign. If the answer is that the store is sound and the constraint is traffic volume rather than conversion, that is what the answer will say.