A data-quality read of the full catalog — and where it leads.
This is a real EKOM catalog analysis, with the brand's identity removed. The client is a supplements and wellness brand — protein powders, capsule supplements and topicals, sold direct through its own storefront, alongside a large partner-linked range in the same product feed. EKOM ran a full-catalog quality read with no schema supplied and surfaced nineteen findings, eleven of which are real, actionable catalog issues. Eight were heuristic and are excluded from every count here; two of the eleven did not survive verification as first written and are accounted for on page eight.
This is a brand that competes on what is in the product and what is not. That makes the product record something more than marketing copy: it is the claim itself, restated on every channel that reads it. A shopper checking whether a product suits them does not open a lab report. They read the sentence under the product name.
Where the eleven verified findings concentrate, by theme.
| Theme | Count | Who feels it |
|---|---|---|
| Product copy contradicting the product | 1 | Shoppers |
| Partner destination wrong or invalid | 2 | Shoppers + revenue |
| Duplicated and repurposed listings | 2 | Shoppers + search |
| Draft artifacts on live titles | 1 | Shoppers |
| First-party and partner classification crossed | 2 | Ops + reporting |
| Products with no or unstructured copy | 1 | Shoppers + channels |
| Conflicting values within one product line | 1 | Shoppers + ops |
| Incoherent pricing state | 1 | Shoppers |
The profile pass found this before it found any defect, and it explains the shape of everything that follows.
This storefront carries two different businesses in one product feed. The first is the brand's own line: manufactured, stocked, priced and shipped. The second is a large partner and cross-promotion program: partner products, promo codes, outbound links — no inventory, no fulfillment, and a price of zero. Nothing declares this split. EKOM inferred it from how vendor, price and the tag field behave across the catalog, and then held every product to whichever set of rules its own side implies.
That inference is what makes the rest of this analysis useful rather than merely accurate. Most of what surfaced sits on the partner side — which matters less for what it costs today than for who should fix it, and for the fact that partner listings are created fastest and reviewed least. It also means the two most serious findings are not the same kind of problem: one is a first-party copy contradiction, the other is a partner routing failure, and they belong to different owners inside the business.
A tag reserved for stocked first-party inventory sits on at least nine partner rows priced at zero — so any inventory-scoped report or outbound feed built on that tag inherits rows that are not inventory. Read from the other end, five of the brand's own manufactured products are classified under the partner vendor value, so any report scoping to the first-party vendor to enumerate the brand's own line silently omits them. Neither is visible to a shopper. Both are visible to every system that trusts those fields.
Two findings rated critical. One is on the flagship line; the other is on the partner side, where most of the rest sit.
The provenance is visible in the data rather than inferred about anyone's process. The brand runs a separate dairy-protein line, and that line publishes the same sentence as its own product headline — where it is accurate. The plant-based line was built from it, and the copy came along unedited. That is the difference between a typo and a mechanism: the sentence is not wrong because someone mistyped it, it is wrong because it was correct somewhere else and got carried.
It is also the finding whose cost is least contained. Product copy does not stay on the storefront it was written for — the description field is exactly what retail-partner feeds, marketplaces, comparison engines and AI assistants ingest and repeat. Correcting it at the source is two fields on four products. Chasing it after it has propagated is a different job.
Most of the eleven sit here, where listings are created fastest and reviewed least.
Both share a structural cause rather than a content one. The tag field is being used to carry a destination, and nothing in the observed data suggests the value is checked — that it is a link, that it resolves, or that it resolves to the product it is attached to. A field that holds a URL by convention will hold whatever is pasted into it.
Duplicating an existing product is the fastest way to add a listing on this platform, and it is how much of the partner program appears to have been built. The pattern works. What it leaves behind is a set of identity fields still describing the product that was copied — at least seven storefront URLs naming a different item than the one on the page, and five live titles still carrying the draft artifact. Whatever search value those URLs accrue is accruing against the wrong product, and a shopper who checks the address bar before buying sees a mismatch. One partner product is live three times over with the same destination, the same code and the same tags.
Eleven findings sounds like eleven problems. Each one is downstream of a process that is wrong in one specific, repeatable way — and each of those leaves a signature in the data that points to it rather than asserting it.
| Upstream cause | The tell — what points to it |
|---|---|
| Copy inherited from a sibling product and never edited | The plant-based line carries the dairy line's exact headline sentence, and the dairy line publishes that same sentence as its own, where it belongs. |
| Duplicate-and-repurpose, with the identity fields left behind | The URL handle and the title still name the product that was copied, not the product that exists. Each affected handle still carries the platform's duplication marker. |
| A field carrying a destination, with nothing validating it | The same URL, down to the variant id, on five listings — correct on one and wrong on four — and a raw image path in that same position on a sixth. |
| One business's conventions applied to the other | A stocked-inventory tag on rows priced at zero; a partner vendor value on products the brand manufactures itself. |
| Two values on one record with nothing deciding between them | A serving-count tag that disagrees across a single flavor line; a compare-at price identical to the price it is meant to compare against. |
Four recently launched products have no body copy at all — not thin copy, none. Four more carry prose but no structured highlights, which is the shape most channels ingest and most shoppers scan; the pattern to match already exists in the catalog on a product that is populated correctly. For a brand whose case for a product is the ingredient rationale, the first four pages are asking a shopper to buy a supplement on the name alone, and giving a retail feed and an AI assistant nothing to repeat.
Nineteen findings were raised and eleven are reported. This page accounts for the difference, and for two that did not survive verification as they were first written. Both were corrected before anything was written down rather than repeated.
Eight findings came from the structural profiling layer rather than the analytical one — pattern observations about the shape of the data, not specific checkable defects on named products. They are not in the eleven, and not in any total on these pages. Reporting all nineteen would have inflated the reportable count by roughly three-quarters — eight more on a base of eleven.
The fifth product was cited alongside the other four, and the written rationale claimed it carried the wrong copy too. It does not. The change the analysis itself proposed for that product recorded the before and the after as identical, correct copy — a no-op — and the raw storefront feed agrees with the change rather than with the prose. Every product in the line was then swept rather than trusting the cited list, and the page re-checked live. Exactly four carry the claim, and it is reported here as four.
It named ten recent products with blank structured highlights, reasoning that they had body copy to draw on — though only eight were bound to the finding with identifiers, and those eight are what EKOM checked. Those highlights are a field EKOM derives, by splitting bullets out of the body copy — so a defect there could have been EKOM's own extraction rather than the brand's data, and that is what was checked first. The extraction dropped nothing. But the stated reason was wrong: four of those eight have prose with no bullet structure as described, and four have no body copy at all — a worse problem than the one reported.
We would rather hand a brand eleven findings we have stood behind than nineteen with two that cost an afternoon and our credibility. The confident-sounding finding is the one that earns the extra check.
Nothing here was confirmed with the merchant. These are leads pending human confirmation, and a share of any cold read turns out to be intentional — a tag used for something other than its name suggests, a classification with a reason behind it. That is expected, and it is why every finding is shown with the actual value behind it wherever a product-level value exists, and why the analysis says so plainly where a correction needs a judgment only the brand can make.
Of the eleven, five are prepared correction sets whose right value is derivable from the catalog itself. Six are questions whose answer exists inside the business — on a package, in a partner's system, or in someone's judgment — and this read does not guess at any of them. That division is the whole point: it is what makes the five safe to apply and the six safe to ask.
This pass read and diagnosed. The same structural understanding powers the work that follows — turning a diagnosed catalog into one that reads right everywhere it's seen.
This is EKOM's first read of this brand's catalog. The pipeline profiled the catalog with no schema and no manual setup — identifying the vertical and every field's role before any analysis ran — then read the full live catalog rather than a sample. Of the nineteen findings it produced, eight came from a structural layer describing how the catalog was shaped for analysis and were excluded; two of the remaining eleven were corrected against source before anything was written down, one over-cited by a single product and one whose stated cause blamed a field EKOM derives itself. The findings here are what remained, with the brand's identity removed.