Catalog completeness isn’t one score. It’s one answer per channel.
A completeness percentage counts filled fields. It doesn’t say whether a marketplace will accept the listing, or whether AI and shopping agents have enough to work from.
On this page
The completeness conversation in many organizations goes one of two ways. Leadership demands 100% before launch, and teams stall while they fill obscure attributes nobody filters by. Or nobody tracks completeness at all.
In that second case the bill arrives months later. Channels reject listings. AI enrichment returns drafts the team can’t use. Products don’t appear in AI search. An audit turns up compliance gaps.
A product doesn’t have a single completeness score. It has one score for each place its data has to go. A jacket can be:
- 92% complete for your English DTC webshop. Hero image, description, size range and price are all filled in.
- 34% complete for Amazon DE. The translation is missing, the Amazon bullet points are empty, the Browse Node isn’t mapped, the EAN isn’t assigned.
- 78% complete for an AI shopping feed. The structured attributes are there. Sustainability fields and variant availability aren’t.
- 15% complete for the data a textile passport is expected to need. Materials composition, origin, recyclability and durability are all empty.
Those are four different problems, and they take four different actions. One number flattens them into something nobody can prioritize against.
Completeness isn’t a metric. It is operational infrastructure. It decides what other software can do with your catalog: whether AI generates content worth shipping, whether a marketplace accepts the listing, and whether a purchasing agent puts your product in a comparison. It also decides whether a publishing pipeline runs end to end or stops for a person, and whether the product experience a shopper meets is intelligent or broken.
What incomplete data breaks.
The cost isn’t a lower data quality score. It’s specific operational failures that compound across every system that reads your product data.
Scroll sideways to see the whole table.
| System | What breaks | What it costs |
|---|---|---|
| AI content generation | With no source attributes to work from, drafts come back generic or plainly wrong. | A high rejection rate on AI output. The team stops trusting the tooling. Manual enrichment stays the bottleneck. |
| Marketplace listings | Required fields are missing, so listings are rejected. Submissions can fail on the first attempt. | Weeks of hand-fixing per batch. Products miss seasonal windows. Revenue slips or is lost. |
| AI search and discovery | Products without typed attributes can’t be compared. Missing specs can keep them out of AI Overviews and out of agent comparisons. | Your products may not show up in the discovery channel. Competitors with complete data can show up instead. |
| Recommendation engines | Without relationship attributes such as compatible_with, alternative_to and use_case, the engine has only purchase correlation to guess from. | Recommendations miss. Cross-sell is poor. Bundling suggestions stay generic, and personalization has little to work with. |
| Autonomous workflows | Completeness gates don’t pass and publishing pipelines stop. An enrichment agent can’t tell a missing value from an attribute nobody modeled. | Every run waits for a person to triage it. |
| AI shopping agents | Without structured variant attributes and availability, an agent can’t resolve a query like “blue, size M, ships to DE.” | Your products sit outside AI-mediated purchases. No transactions come from the agent channel. |
| Compliance validation | You can’t check General Product Safety Regulation (GPSR) or Digital Product Passport (DPP) requirements in software unless the attributes are modeled and filled per product. | Regulatory exposure. Audits turn into fire drills. The machine-readable declarations can’t be produced. |
| Customer experience | Filters come up empty. Comparison tables have blank cells. The chatbot can’t answer a question about specifications. | The experience reads as poor. Bounce rate rises. Support tickets arrive for things a shopper should be able to look up. |
Each of these failures gets blamed on something else. “The AI tool is bad.” “Amazon changed requirements.” “The recommendation engine needs tuning.” “The chatbot needs training.”
The root cause underneath is the same. The product data those systems operate on is incomplete.
Completeness is per channel, per locale and per use case.
Many completeness features count required fields per channel and language. What they often leave out is weighting by business impact, and whether a filled value passed validation.
Note
Ask the question per channel
The question isn’t “what percentage of fields are filled.” It’s “does this product have everything this channel, this AI system, this regulatory body or this agent protocol requires?” The answer comes one per channel, one per locale and one per protocol, rather than as a single number.
Missing fields stop software, not just people.
When people did this work by hand, a missing field meant somebody typed it in. Annoying, but manageable on a small catalog. When software does the work, a missing field stops the software.
AI writes what the record gives it.
To describe a “Running Shoe,” the model needs brand, model, cushioning_type, drop_mm, weight_g, upper_material, sole_type, intended_surface and target_runner_type. With those structured attributes, the draft comes back specific and ready for the channel.
Without them you get “A great shoe for runners who want comfort and performance.” Generic prose that adds nothing, and it can invent specifications that aren’t in the record.
Completeness shapes enrichment quality. Products with rich structured attributes are likely to get more of their drafts accepted; with thin records, reviewers end up rejecting or rewriting more. Same model, different input context.
Pipelines use the score as a gate.
A publishing pipeline asks one question. Is this product ready for this channel? The answer comes from scoring the product against that channel’s requirement template.
Without a per-channel score, the pipeline can’t decide on its own. Someone reviews every product for every channel, and the work then grows with headcount instead of with the software.
Agents compare what they can verify.
A purchasing agent comparing your product against competitors reads specific attributes: battery_life_hours, noise_reduction_db, weight_g, warranty_months. If two of those are missing from your record and a competitor has all four, the agent has two choices. Guess your values, or leave your product out of the comparison.
Guessing risks the user’s trust, so an agent is likely to leave your product out. Completeness is the entry ticket to AI-mediated commerce.
Completeness isn’t a project with an end date.
The traditional model is a data cleanup project. Fill the missing fields, declare the catalog complete, move on. It fails for one reason: the requirements don’t sit still.
- Marketplaces add new required attributes regularly, and category requirements change with little notice
- New channels arrive with completeness requirements of their own: TikTok Shop, Google AI Mode, ChatGPT product feeds
- Regulations create new attribute mandates: GPSR safety fields, DPP sustainability attributes, energy labels
- AI systems improve, and then want richer context, because more structured attributes make better generated content
- New products enter the catalog daily from suppliers who provide inconsistent levels of detail
- Existing products change through reformulations, new variants and updated specifications, and each change needs enriching again
A one-time cleanup starts decaying the day it is finished. Keeping a catalog complete is an operational discipline that systems maintain, rather than a project humans complete.
What it takes to keep it current.
More editors don’t hold the line on a large catalog. These are the parts that do.
Scroll sideways to see the whole table.
| Part | What it does | Why doing it by hand fails |
|---|---|---|
| Per-channel requirement templates | Says exactly which attributes, at which quality level, each channel, locale and protocol needs. | Without templates, “complete” means nothing in particular. Teams argue about what is needed, and nothing can be scored automatically. |
| Continuous scoring | Scores every product against every template that applies, as attributes change rather than once a quarter. | A manual audit is a snapshot. Products drift between audits, and the gaps turn up at publish time. |
| Publish gates | A product below the threshold can’t be exported to the channel. The gate sits in the software, not on someone’s checklist. | Without gates, incomplete products reach the channel. You find out when the rejection comes back, and the fix cycle runs into weeks. |
| Gap detection | Lists the missing attributes per channel so the right team can pick them up. | Without it, gaps sit in a spreadsheet nobody owns. Nobody can rank them without knowing which channel each one blocks. |
| AI-assisted enrichment | Where a source exists (supplier docs, images, existing descriptions), AI drafts the missing attributes and they are validated. | By hand, the work grows with the catalog: 10,000 products × 5 missing attributes each is 50,000 decisions for a person to make. |
| Validation on enrichment output | Every AI-generated attribute is checked against the type schema, the business rules and the source before it lands in the catalog. | Without that check, AI fills the gaps with invented values. A record that is “complete” but wrong is worse than one that is short. |
| Completeness trend monitoring | Follows the score over time and raises an alert when it drops, whether a requirement was added or a supplier’s data got worse. | Without monitoring, completeness slips quietly. You notice when something downstream starts failing. |
The goal is to treat completeness as an operational system: per-channel, weighted, kept current, and wired into the checks and agents that act on it. CataZenta doesn’t weight completeness or use it to gate publishing. Two related pieces exist: Amazon and Shopify pushes are checked against the channel’s field rules before sending, and an AI Catalog Quality score tracks what agents have found, not completeness.
Before and after, on the same jobs.
What per-channel completeness would change, job by job.
Scroll sideways to see the whole table.
| Job | As a metric | As infrastructure |
|---|---|---|
| Marketplace submission | Say you send 200 products. 73 come back rejected for missing fields. The team spends 2 weeks fixing them in Seller Central, then does the same thing next quarter. | Before anything is sent, each product would be checked against Amazon’s required fields and rules. A product that fails would be held back with its gaps listed, AI could draft the missing values for review, and only products that pass would go out. |
| Adding a new channel | “How do we know what TikTok Shop needs?” Then 3 weeks of research, a manual audit of 5,000 products, and a spreadsheet to track it. | Requirements would be defined once for the new channel. The catalog would be checked against them, with the gaps listed per product. AI could draft values where the data supports them; people would handle the rest. |
| Product launch (150 SKUs) | Enrichment takes 6 weeks. Nobody can see what is blocking any given product. The work runs one product after another, by hand. | A per-channel view would show: 80 products ready for the webshop; 45 waiting on German translation for the locale team; 25 waiting on media for creative. All three can move at once. |
| AI content generation | Say you run AI across the whole catalog, and much of the output comes back rejected. There is no way to tell in advance which products will come back usable. | AI enrichment would run first on products with rich structured attributes, where drafts are most likely to be accepted. Thin records would need their missing attributes filled first. |
| Regulatory audit | “Are we GPSR compliant?” Nobody knows. Finding out takes a consultant assessing the catalog by hand. | GPSR requirements would be checked across the whole catalog, showing which products pass and which have gaps, itemized per product. |
Attributes aren’t worth the same.
A missing hero image and a missing internal note field are not the same problem. A filled-or-empty count registers both as “incomplete.” Their impact on listing acceptance, conversion and AI usability can differ widely, so the scoring has to be weighted.
Scroll sideways to see the whole table.
| Attribute tier | What it affects | Weight (example) | What it does to the score |
|---|---|---|---|
| Critical (hero image, title, primary description, EAN) | Without them the listing is rejected. The product can’t be sold. | 40% of total score | Miss any one of these and the product scores below the publish threshold, however complete the rest of the record is |
| High (variant structure, key specifications, pricing, availability) | The product lists but performs poorly. Missing specs can keep it out of AI answers and hurt discovery. | 30% of total score | A gap here holds the product in the “enrichment needed” state |
| Medium (secondary images, enhanced descriptions, cross-sell links) | Lifts conversion and the quality of AI output. Nothing is blocked, but revenue is affected. | 20% of total score | The product can publish without them, and it is flagged for Wave 2 enrichment |
| Low (internal metadata, long-tail locales, optional sustainability claims) | Matters for reporting and for niche audiences. No immediate effect on revenue. | 10% of total score | Picked up as the catalog improves; never blocks a publish |
The specific weights depend on your channels and your strategy. The principle underneath doesn’t change: score business impact, not field count.
A product with 4 of 50 fields filled, where those 4 are hero image, title, description and price, may be more “operationally complete” for an initial launch than a product with 45 of 50 fields filled that is missing the hero image.
The score can trigger the work.
Designed well, completeness isn’t only a reporting metric. It becomes the event the next step listens for.
- A product crosses 80% for Amazon DE, and a review task opens for the marketplace team
- A requirement changes, a product drops below threshold, and an enrichment task lists the exact gaps
- AI enrichment fills the gaps and the product reaches threshold. It routes to the approval queue, not to publish
- An approved product clears the threshold on each channel, and it can go to the channels it qualifies for
- You add a requirement template, every product is re-scored, the gaps come back in bulk, and enrichment agents can start on the high-priority items
- A supplier batch imports with low completeness, and it is flagged for enrichment priority rather than an immediate publish attempt
Completeness can stop being a number someone reads in a dashboard and become an event the system acts on. It only works where a requirement template has been defined for the channel.
Start in three waves.
Completing everything before launch delays revenue. Define three publish waves per product assortment instead.
Scroll sideways to see the whole table.
| Wave | Completeness target | What it covers | How it runs |
|---|---|---|---|
| Wave 1: Launch (week 1-2) | 75-80% weighted | Critical attributes only: hero image, title, description, price, basic variant structure | Products start earning while enrichment carries on. The gate: critical fields validated. |
| Wave 2: Enrich (week 3-6) | 90-95% weighted | Secondary images, enhanced descriptions, cross-sell links, all primary locales, key specifications | AI drafts values for many of the gaps. A person reviews the high-impact attributes. Tuning per channel follows. |
| Wave 3: Optimize (ongoing) | 98-100% weighted | Long-tail locales, sustainability attributes, A+ content, regulatory fields, agent-ready specifications | Enrichment keeps running. Monitoring picks up new gaps as requirements change, and they are worked off a few at a time. |
Wave 3 doesn’t end. It is the steady state. New requirements emerge, new products enter, existing products change, and the checks keep running against them. That is what completeness as infrastructure means in practice.
What to report upward.
At the executive level it comes down to one question. What percentage of the catalog is operationally ready for each revenue-generating channel and each AI commerce protocol?
Track these.
- Weighted completeness for your top 100 revenue SKUs on the primary channel. That is the current baseline
- Completeness velocity. Is the score improving week over week, or degrading?
- Gap concentration. Are the missing attributes clustered, so you can fix them in bulk, or scattered, so each product needs its own attention?
- AI enrichment eligibility. What share of the gaps can AI resolve, and what share needs a person to source the data?
- Publish-readiness by channel. How many products could go out today if you flipped the switch?