Skip to content

AI field mapping: turn messy files into a usable product feed

A beginner-friendly guide to converting CSV files with columns like foto_principal or PVP into channel-ready feeds, with AI suggestions and human review.

J

Julia Villegas

· 9 min read · 268 views
AI field mapping for a product feed with ordered product cards and paper pieces

AI field mapping means using AI to suggest how the columns in your product file match the attributes required by an advertising or commerce channel. If your CSV uses names like foto_principal, PVP or supplier_ref, AI can propose mappings, but you should still review the important fields before publishing the feed.

What AI field mapping means

A product feed is a structured file that tells a channel what products you sell and how each product should appear. Channels such as Meta, Google Shopping, TikTok or Pinterest expect product data in a consistent format: product ID, title, description, price, availability, product URL and main image are common examples.

The problem is that your store, ERP, PIM, supplier file or custom export may not use those same names. Your source file might include columns such as:

  • sku
  • item_name
  • foto_principal
  • PVP
  • web_stock
  • product_url
  • manufacturer_brand

Field mapping is the process of saying: “this source column corresponds to this destination attribute.” AI field mapping adds a layer of assistance. The system looks at column names and, where possible, suggests likely matches. For example, it may suggest that foto_principal should become image_link, that PVP should become price, or that item_name should become title.

In AdMotion, the Feed Manager can connect stores or product files, map fields, clean catalog data with rules, create product sets and publish feeds through a URL generated on each download. The Export Channels module prepares feeds for Meta, Google, TikTok, Pinterest, Snapchat, Criteo, Microsoft Advertising, Thribee and ChatGPT, with an analyzer that checks the feed against the destination channel’s specification.

Why messy CSV files create feed problems

A messy CSV is not necessarily a bad file. It is often just a file built for another purpose: internal reporting, supplier ordering, warehouse management or a legacy ecommerce system. The friction appears when you try to use that file as the source for product ads.

Common issues include:

  • The same concept has different names: price, PVP, retail_price, final_amount.
  • The main image is called image, photo, foto_principal or image_url_1.
  • The product ID is confused with a supplier reference.
  • Availability appears as “yes/no”, “1/0”, “in stock/out of stock” or a numeric stock count.
  • Category fields reflect internal merchandising, not the taxonomy expected by the ad channel.
  • Some columns mix information that the destination channel expects to receive separately.

For beginners, the key point is this: a feed is not just a spreadsheet. It is the raw material behind catalog ads, Shopping campaigns, Performance Max assets and dynamic product ads. If the source data is mapped incorrectly, the downstream campaign can inherit bad images, missing prices, duplicate IDs or unusable product groups.

Example: from foto_principal and PVP to a usable feed

Imagine an online home decor store exporting a CSV called decor_catalog_october.csv. This is a fictitious example, but it reflects the type of file many ecommerce teams receive from stores, suppliers or internal systems.

Source column Example value Suggested destination field Human review
internal_ref LAMP-2381 id Correct if it is unique and stable
item_name Bora ceramic table lamp title Correct
long_text Ceramic table lamp... description Review text quality and formatting
PVP 79.90 price Check currency and decimal format
foto_principal https://.../bora.jpg image_link Correct if the URL is public
web_stock 12 availability Needs a rule: more than 0 = in stock
product_url https://.../bora link Correct
family Lighting product_type Useful for segmentation

This example shows the realistic role of AI. Obvious matches can be suggested quickly, but some decisions still require business context. Is internal_ref truly stable enough to be the product ID? Does PVP include tax or currency? Should numeric stock be transformed into availability? Is family a product type, a campaign segment or simply an internal merchandising label?

That is why AI field mapping works best as assisted mapping, not blind automation. AI proposes; a marketer, feed manager or ecommerce operator reviews.

How to use AI field mapping step by step

Before pushing a file into a channel, treat the feed as a workflow. The goal is not simply to upload a CSV. The goal is to turn product data into a structure that each channel can read.

1. Identify the critical fields

Start with your source file and locate the fields that matter most:

  • Product identifier.
  • Title.
  • Description.
  • Price.
  • Product page URL.
  • Main image URL.
  • Availability or stock.
  • Brand, category or product type, if available.

Do not try to perfect every column in the first pass. First, make sure every product has the data needed to be displayed, linked and updated consistently.

2. Let AI suggest the first mapping

Once the file is connected, assisted mapping can propose which source column should match each destination attribute. This is especially useful when you receive files with non-standard names or when an agency manages multiple clients that use different naming conventions for the same data.

For example:

  • foto_principal → image_link
  • PVP → price
  • item_name → title
  • product_url → link
  • manufacturer_brand → brand

The benefit is not only speed. AI-assisted mapping can also help surface useful fields you may have overlooked, such as color, material, season, margin group or product family.

3. Review ambiguous fields manually

Ambiguous fields cause many feed problems. Do not approve them without checking product examples. Typical examples include:

  • reference: could mean SKU, supplier reference or internal ID.
  • sale_price: may be empty when there is no discount.
  • stock: could be a number, a text value or an internal status.
  • category: may be an internal category rather than a channel taxonomy.
  • image_1: may not always be the main product image.

Human review matters most when the field can affect approval, delivery or user experience. A wrongly mapped price, broken image or unstable product ID can create rejections, misleading ads or reporting confusion later.

4. Normalize values with rules

After mapping comes cleaning. Normalization means converting different source values into a consistent format.

Common rules include:

  • If web_stock is greater than 0, set availability to “in stock”.
  • If web_stock is 0, set availability to “out of stock”.
  • Add currency when the source file only includes a number.
  • Exclude products without a main image.
  • Create a campaign priority field when the source data supports it.

The AdMotion Feed Manager supports rules for cleaning and segmenting catalogs, including rules written in natural language. That is useful when you want to express a business condition without building complex formulas directly in the source file.

5. Create product sets

A product set is a group of products that meets a condition. For example:

  • Products with available stock.
  • Products in a specific category.
  • Products below a certain price point.
  • New seasonal arrivals.
  • Products with a valid image.

Sets help you create better feeds for each channel or campaign. You do not always need to send the full catalog everywhere. One store might create a feed for high-stock products, another for sale items and another for a category-specific campaign.

This also helps when you later work on creatives. With the Templates Editor, you can use catalog fields such as image, title, price and brand to generate one image per product in 1:1, 4:5, 9:16 and 16:9 formats. That creative layer is easier to manage when the feed is already structured.

6. Generate the channel feed and check warnings

Once fields are mapped and values are normalized, create the feed in the format required by the destination channel. The Export Channels analyzer checks errors and warnings against the channel specification. This helps you catch problems before registering the feed URL in the ad platform.

This review is not a guarantee of approval. Each channel has its own policies, validation rules and processing times. Also, the frequency at which your source catalog is imported and the frequency at which the channel downloads or processes the feed URL can be different. Do not assume instant synchronization inside the destination platform.

7. Copy the feed URL into the destination channel

When the feed is ready, it is published through a URL. That URL is then registered in the relevant advertising or commerce channel. The URL is generated at download time, so the channel can access an updated version according to its own fetch and processing schedule.

If you manage several channels, avoid forcing one generic feed to do everything. One catalog can power multiple feeds with different formats, rules and sets: one for Meta, one for Google Shopping, one for TikTok or Pinterest, and a custom feed if your workflow requires it.

How to do it with AdMotion while staying in control

A practical AdMotion workflow would look like this:

  1. Connect your store, file or existing feed in the Feed Manager.
  2. Review the assisted mapping suggestions, such as foto_principal as the image, PVP as the price and item_name as the title.
  3. Manually adjust any ambiguous field before continuing.
  4. Create rules to normalize availability, price, images or categories.
  5. Build sets to separate products by stock, campaign, category or business condition.
  6. Choose the export channel and review errors or warnings with the analyzer.
  7. Copy the feed URL and register it in the corresponding channel.

This avoids two common extremes: mapping everything manually from scratch, or accepting every automated suggestion without checking it. AI speeds up repetitive work; you keep control over what each field means and which products should enter each feed.

If your main challenge is turning inconsistent product files into channel-ready feeds, start with the Feed Manager and test the workflow with your own catalog.

Final check before publishing the feed

Before copying the URL into Meta, Google or another channel, review a sample of real products. Do not rely only on the first row of the CSV.

Check What to review Warning sign
Product ID Unique and stable Changes by supplier or export
Price Consistent format Missing currency or mixed decimals
Image Public URL and main image Broken, secondary or generic image
Stock Converted into availability Internal values left untranslated
Link Goes to the right product page Redirects or missing products
Sets Include the expected products Out-of-stock products in active campaigns
Channel format Matches the destination One generic feed poorly fits several channels

If you find issues, go back to mapping and rules before publishing. It is usually faster to fix structure upstream than to chase errors, warnings or rejected products later in every platform.

What to do this week

  1. Pick one real product file and identify the critical fields: ID, title, price, image, link and stock.
  2. Create a simple mapping table with the original column names and the destination fields you think they should match.
  3. Mark ambiguous fields such as reference, stock, category and sale price, then review them with real product examples.
  4. Define two or three normalization rules: availability from stock, exclusion of products without images and price formatting.
  5. Create one simple set, such as products with stock and a valid image, and use it as the basis for your first channel feed.

AI field mapping is not about handing your catalog to a black box. It is about accelerating the obvious matches, reviewing the decisions that matter and building feeds that each channel can interpret with less friction.

Frequently asked questions

What does field mapping mean in a product feed?

Field mapping means matching the columns in your product file to the attributes required by a channel. For example, you might map foto_principal to the product image, PVP to price and item_name to title. It is a foundational step before a feed can be read correctly.

Can AI map my entire catalog automatically?

AI can suggest likely matches and reduce repetitive work, but important fields should still be reviewed before you apply them. Product IDs, prices, stock, images and categories often need human judgment because the same column name can mean different things depending on the store or supplier.

Does a checked feed guarantee that my products will be approved?

No. Checking errors and warnings helps you prepare a cleaner feed, but each channel has its own policies, validation rules and processing times. The feed update frequency and the frequency with which a channel reads or processes that feed can also be different.

Sources

  1. Feed Manager de AdMotion — AdMotion
  2. Export Channels de AdMotion — AdMotion
Share:
J

Written by

Julia Villegas

Performance Marketing

Julia Villegas works in performance marketing at AdMotion.

Related posts