- Import moves data; processing turns it into a usable catalog with categories, attributes and variants.
- Read a SKU together with its supplier. Matching names do not prove that two entries are the same product.
- Assign a source to each field so repeat imports preserve your own photos and edited descriptions.
- An unavailable feed is an exchange problem, not evidence of zero stock.
- Start with one source, a trial import and checks of the product pages and filters on your site.
A supplier sends a new price list. One column mixes the product name and size, another says “blk.”, photos are supplied as links, and the category is called “New for autumn”. A manager opens the spreadsheet, fixes names, sorts products into sections and copies their attributes into the store. Another file arrives the following week. Some of the work starts all over again.
This is where automated product processing helps. A system reads the supplier’s data, applies your store’s rules, checks the results and prepares product pages for publication. Importing products into the website becomes a repeatable operation. People handle exceptions and work on the assortment instead of copying the same cells each time.
Artbrain can develop supplier integrations and automate catalog updates for your existing online store. We start with a sample price list and a review of the site to determine what can be configured, which import needs custom development and what drives its cost.
But an “Upload Excel” button does not solve the whole problem. A successful upload can still create three categories for one type of footwear, duplicate products and sizes that never appear in a filter. Judge the result on the storefront: can a customer find the product, choose the right variant and understand what they will receive?
What automated catalog preparation includes
Import moves rows from a source into a system. Processing establishes what those rows mean. An entry might be a new model, another size of an existing model or an offer from a different supplier. The next steps are assigning a category, putting attributes into the right fields, preparing images and separating entries ready for publication from those needing review.
Preparing an online store’s catalog this way can begin with a manual file upload. Automation does not have to mean that everything immediately runs on a schedule without a manager. Getting a consistent result on repeat imports is initially more useful than publishing every row quickly without checks.
- Source: which supplier provided the file and when it was updated.
- Preparation: categories, attributes, units, the model and its variants.
- Validation: missing fields, contradictions and entries that cannot be matched confidently.
- Publication and reporting: what was added, what changed and what still needs a manager’s decision.
Automatically adding products to a website needs agreed boundaries. The process can prepare drafts, update only permitted fields or publish validated entries. The owner chooses that mode before launch, rather than after the first unsuccessful upload.
Excel, CSV, XML and feeds: what an owner needs to know
Excel is convenient for people. You can add a second sheet, highlight a row or write an explanation above the table. A program needs explicit columns and rules. If product color is recorded only as a cell’s background color, move it into a separate field. Merged cells, totals and comments between products also need to be accounted for.
CSV is a text table. Reading it correctly depends on the delimiter, character encoding and decimal separator. A SKU such as “00127” must remain text so it does not become “127”. For a developer these are import settings; for the owner they are a matter of preserving product identity.
XML describes data using nested elements. It may have a separate category section, several images and a set of parameters for each entry. It is often convenient for regular exchange, but two XML files can have different structures. A store’s support for XML does not automatically make it compatible with every supplier’s XML.
A feed is a data stream or file a source provides for exchange. It can use XML, CSV or another format. A feed link alone does not tell you what is inside, how frequently it changes or whether it contains the complete assortment. Ask for a sample and the source’s rules, not just its address.
From spreadsheet row to product page: a hypothetical example
Consider this hypothetical price list entry:
- Source: “Supplier A”; SKU: “00127-B-38”.
- Name: “Trainers model 127 blk. size 38”; category: “New for autumn”.
- Model: “127”; color: “blk.”; size: “38”, with size system “EU”.
- Weight: “0.72 kg”, qualified by the source as “pair without packaging”; image: a link to the black variant.
Under agreed rules, this entry can become a product page named “Trainers, model 127”. The “Trainers” category follows the confirmed product type rather than the word “autumn”. Color receives the standard value “Black”, size remains “EU 38”, and the weight of the pair is recorded as “720 g”. The SKU stays unchanged and retains its connection to the source.
If the model has other sizes, they may appear as variants on this page. Grouping them is supported by the model field and an agreed supplier rule. A similar title alone is not sufficient. If the file lacks the product type or the meaning of its weight field, the system must not invent them: the entry needs clarification.
For pricing, agree separately on the source currency and markup rules. Decide who controls discounts and manually set retail prices so a price list update does not cancel them. Unusually sharp price changes should go to review before publication. This is also part of catalog preparation and needs to be defined before the import launches.
Categories and attributes: one dictionary for different suppliers
Suppliers describe their own structure, while customers expect yours. “Women’s shoes”, “Ladies” and “Women” may need mapping to a single section. Meanwhile “Sale” is not a product type, and “Winter” may be a season attribute. Copying every label verbatim quickly turns your navigation into a record of other people’s spreadsheets.
Use a small mapping dictionary: source category, store category and exceptions. An ambiguous value should leave an entry in draft or create a review task. A new supplier category should not automatically appear in the main menu. Managers need to see which products still lack a section and why.
The same applies to filters. “Колір”, “Цвет” and “color” can represent one field, while “blk.”, “black” and “чорний” can share a standard value. But “genuine leather” and “artificial leather” must not be merged. Mapping needs to respect meaning rather than just similar spelling.
Units need rules too: 0.72 kg equals 720 g if both measurements refer to the same object. The weight of one item and the weight of a box are not interchangeable. EU and US sizes cannot be mixed in a single numeric filter without a confirmed manufacturer’s conversion chart. Check several filters on the website after import: a correct value on a product page may still fail to work in the product selection.
SKUs, duplicates and multiple suppliers for one product
A SKU identifies an entry within an agreed system. Two suppliers may use “127” for different items. A connection therefore usually starts with “supplier + their SKU”. Also establish whether that identifier describes a model, a particular size or a complete package.
A product name is an unreliable identifier: the supplier can change word order or add a season. Searching similar names can suggest a candidate for review, but does not prove a match. Merging requires confirmed identifiers, such as a manufacturer’s code together with the brand and variant, or a verified mapping between sources.
One product with two supplier offers is a normal situation. The customer sees a single product page, while the store retains its connection to each source. Creating two identical pages is easier initially, but makes descriptions, photographs and URLs harder to manage later. Merging different products is worse: a person may order one variant and receive another.
A repeat import of the same file should find entries already known to the store. Check this explicitly in the report: uploading twice must not create another copy of the catalog. If a supplier changes a SKU, you need a rule for reconnecting it, rather than an automatic assumption that the old product disappeared and a new one arrived.
Sizes and colors: what belongs on one product page
For shoes, choosing a model and then a size and color feels natural. For tool sets, similar names can conceal different contents that require separate pages. Your assortment and source data determine the variant structure. A universal rule such as “remove the last word from the name” is dangerous here.
Brand and a general description may belong to the model. The variant retains its own identifier, size, color and corresponding images. Check that choosing a color actually displays that color’s photo rather than a random first image of the model. A missing size must not become “one size” unless the supplier confirms it.
Your trial sample should include a model with one variant, one with several sizes and one with several colors. View them as a customer would: are the selector labels clear, are any sizes duplicated, and does the correct variant remain after the file is updated?
Updates that preserve your work
A manager improves a description, takes original photographs and adds an important clarification. The next import restores the supplier’s text and old gallery. The data technically updated, but the store lost its editor’s work. Rules for each field help prevent this.
- Names, categories and descriptions can be populated on creation, then protected against overwriting after manual edits.
- Attributes owned by the source update according to the agreed dictionary.
- Your photographs are kept separately from supplier photographs.
Define what an empty cell means. It may signify “no information”, rather than “delete the previous value”. Likewise, a missing row does not always mean a product has been discontinued: the file might be partial or faulty. Making bulk changes without knowing whether an exchange is complete or partial is risky.
An unavailable or stale feed is a source problem, not proof of zero stock. Keep the time of the last successful exchange and show the error to the responsible person. What customers see during that period, and whether ordering is allowed, needs a separate data freshness rule. Quantities, reservations and sales channels are covered in the separate guide to stock synchronization.
Invalid rows should not quietly reach the storefront
A useful process has a place for entries needing review. A missing SKU, unknown category or conflicting size receives a clear reason for being held. For a manager this is a list of specific problem entries, rather than a “something went wrong” message. A corrected row can be processed again without creating a duplicate.
An error in one product can hold back its publication while leaving valid entries available for processing. A file-wide problem, such as the identifier column disappearing, must stop an unsafe update. Distinguish these two situations. Otherwise an unexplained change in a new price list can damage a catalog that was already correct.
A report should identify the source, time and counts of read, new, updated, unchanged and held entries. Do not confuse supplier rows with store product pages: several sizes can form one page. An “import complete” message without these results does not explain why half the assortment is missing.
Descriptions, translations and images: where AI helps
Once attributes are organized, AI can produce a readable description and assist with translation. Its basis must be verified fields: type, material, purpose and confirmed features. If water resistance or composition is not specified, it cannot be added as a guess. An attractive sentence describing an untrue feature creates a complaint rather than value.
Important properties need checking against the source. Brand terminology and size labels should remain consistent across languages. Generated text can initially stay in draft; manual approval reduces the risk of publishing mistakes. Description workflows are discussed further in the guide to AI for online store content.
Images require their own checks: permission to use them, an accessible link and a match to the model and variant. Identical files need not be stored repeatedly, but visual similarity does not prove they show the same product. Generating invented product details to fill a missing photograph is a poor solution. Customers should see what is actually being sold.
Good attributes help create a useful page and understandable filters. Rewording a description alone does not guarantee Google rankings. Catalog automation removes some manual work; search results also depend on demand, competition and the condition of the site.
Checking the price list and finding missing data. AI can help before descriptions are written by comparing a parser’s output with the original file. It can flag missing SKUs, misread units or values assigned to the wrong column. If a supplier changes the spreadsheet format, this check can reveal discrepancies before the catalog is updated.
For unusual documents, a custom workflow can also use AI to extract data itself: separate a name, size and material combined in one line and suggest mappings to store fields. Check the output against the source and agreed rules. Unrecognized values await review. Send detected errors to a developer to check the settings or program.
Imagine a price list containing only a brand, SKU and price. A configured workflow searches for that SKU in an agreed manufacturer or supplier catalog, identifies the exact product and collects missing documented attributes. The same source can provide images permitted for use and facts for a description. The data is then normalized to the store’s units and field names, translated and prepared as a draft product page.
Before publication, compare the exact model, size and included items. A similar name or a photograph of another variant does not confirm a match. Keep sources for collected attributes and images so a manager can check each value.
If several similar models turn up, sources conflict or a required page is unavailable, keep an incomplete draft. The manager sees candidate matches, their sources and the specific question to resolve. After confirmation, prepare the product for publication under agreed rules; do not fill empty fields with unsupported properties.
Artbrain can develop this workflow for your sources and catalog. Show us your price list and a desired product page to define what to extract, where to find missing data and which results need a manager’s approval.
How to start on an existing website
Replacing the platform is not a prerequisite. First test your store’s built-in import with a real file. If it supports the fields, variants and selective updates you need, setting up mappings and a consistent price list preparation process may be enough.
A custom integration becomes useful when there are several sources with different formats, connections between offers must be retained, or manual edits need protection. It can also help when managers regularly repeat transformations that the built-in tool cannot support. That process determines the scope, not just the number of products.
- Choose one supplier and a small representative sample that includes complex variants.
- Record field rules and save a restorable state of the catalog.
- Run a trial upload into drafts or a test copy, then check product pages and the report.
- Repeat the import, change several file values and check both updates and manually edited fields.
- After acceptance, connect the remaining sources and agree on scheduled runs.
For existing product pages, preserve URLs and connections important to orders. Clearing everything and recreating the catalog can lose more than it saves. When estimating work on your online store, show the current site, a sample price list and what your manager does with it today.
An example from our work: Airstep
In the delivered Airstep project, we connected automated imports from the XML feeds of four suppliers. The published case describes a catalog of more than 6,700 products, with sizes, attributes and AI descriptions. It demonstrates data preparation as part of the store, rather than a separate spreadsheet copying task.
With several sources, each needs mapping to the store’s shared dictionary so customers see one understandable catalog. On your own site, start by checking how suppliers label models, variants and attributes. Even two files in the same format may require different mappings. Details of the delivered project are available in the Airstep development case study.
The cost of automating online store catalog updates
Development cost depends on supplier count, data formats and quality, category and variant rules, platform capabilities and update frequency. We assess existing catalog cleanup, integrations, reporting and AI processing separately if needed. One agreed XML feed and several unusual price lists with different rules require different scopes. We prepare an estimate after checking samples and discuss ongoing support costs separately.
Take several ordinary updates and record time spent preparing, transferring, checking and correcting data. The entire process matters: a fast upload may save nothing if a manager then spends a day finding errors. Separately record the hours that will still be needed for exceptions after automation.
Monthly financial benefit = (manual hours before the change − review hours after the change) × your hourly cost − recurring solution costs. Use your own calculation of team costs for the hourly rate. Recurring costs include support and third-party services if required. Do not substitute promised savings for measurement.
If the result is positive, dividing the one-off implementation cost by that monthly benefit gives a rough payback estimate. Otherwise, narrow the scope or keep the built-in importer. Fewer mistakes are valuable too, but assess them using your recorded incidents rather than adding invented sales growth.
How to order a product import integration and accept the work
Artbrain offers online store catalog automation and custom supplier integration development. It suits store owners and catalog managers whose built-in importer cannot handle different price lists, variants and manual corrections. We can develop Excel, CSV or XML imports, category and attribute mapping, duplicate checks, selective updates, error reports and an agreed schedule. Deliverables and publication rules follow your store; a new website is needed only if confirmed platform limitations require it.
A useful estimate needs examples of all formats, the number of suppliers, approximate row counts and update frequency. Include ordinary and problematic products, your languages, category rules and fields you edit manually. Identify who can resolve uncertain matches and view the report.
Write acceptance criteria before development. They should be visible on your website:
- A repeated file adds no duplicates; different suppliers with the same SKU are not mixed.
- Categories and filters find the right products, with units and size systems preserved.
- A selected color shows its corresponding photograph, and all required languages are populated.
- An edited description and your own gallery survive the next update.
- An invalid row has a reason for being held, and an unavailable source produces a visible notification.
- The report explains the outcome; the agreed recovery method has been checked on a test catalog.
A few more practical questions: who fixes a change in the supplier’s format, who notices a missed update, and how can a manager add a category without a developer? These answers shape daily work after launch. Automation needs to leave an understandable process even when the next file brings a surprise.
To order a product import integration and request an estimate, send Artbrain the store address, price list samples without passwords or private data, and an example of the product page you need. Include complex items and manual edits that must survive updates. We will establish whether built-in importer setup is enough and propose the development scope, acceptance checks and support terms.
FAQ
Can products be imported automatically from Excel?
Yes, if the file has clear columns and stable identifiers. Agree on field mappings, categories, units and variant structure. First test the store’s built-in importer with your file. Unusual transformations or several different sources may require a separate integration.
What determines the cost of online store catalog automation?
Cost depends on supplier count, data formats and quality, category and variant rules, update frequency and store capabilities. Artbrain can develop supplier integrations for an existing website if its platform supports the required exchange. An estimate needs sample price lists, the store address, a target product page and the fields edited manually. A new site is needed only for confirmed limitations; support and third-party services are assessed separately.
How do I avoid duplicates with several suppliers?
Keep each entry connected to its supplier and their SKU. Combining offers requires confirmed identifiers or a verified mapping table. Identical or similar names do not prove a match. At acceptance, upload the file again and check that no new copies are created.
What happens if a feed is unavailable or a product disappears from it?
An unavailable source does not mean zero stock. You need an error notification and the time of the last successful exchange. Handle a missing row under agreed rules for complete or partial files. What customers see and whether orders are allowed while data is stale must be decided separately.
Can AI check a price list and find missing images and attributes?
Yes, a configured workflow can use AI to compare parsed output with the original price list, flag missing or misread fields, or extract data itself. Missing attributes, permitted images and facts for descriptions can be found in agreed catalogs using the brand and SKU. Check the exact model, size and included items, keep sources and send ambiguous matches to a manager. Artbrain can develop this workflow for your data and publication rules; unsupported properties must not be invented.
