Item mapping: How to make sense of grocery data

Item mapping is the process of connecting or matching corresponding products and data fields across different software systems or databases so they are recognized as the same entity.

Before AI can understand demand, it has to understand the item

Every forecast, every order recommendation, every inventory estimate starts from the same question: What is this item? In grocery data, that question is surprisingly hard to answer.

One product accumulates codes over years. Vendor changes, system migrations, a merchandising team retyping a description. UPCs, PLUs, retailer item IDs, order-guide codes, abbreviated descriptions, pack sizes, and department-specific records often point to what is, in the real world, one product.

Some retailers maintain their own links between codes, but those links are often incomplete or inaccurate and cannot always be taken at face value. That creates a basic problem for any system trying to understand inventory and demand: Which records actually belong to the same item?

Guessing wrong costs in both directions

Group two codes that aren’t the same product, and their sales and demand histories get combined. The system is now learning from a history that accurately represents neither item, distorting the resulting forecast and order recommendation.

Fail to group two codes that are the same product, and demand that should be calculated together is split instead. Each record appears to have less history than the product actually does, making it harder to forecast accurately.

In either case, the data no longer reflects what is actually happening on the shelf. Item identity sits upstream of the demand history and recommendations that depend on it.

For Afresh, an item is meant to represent one unique retail product. Getting that identity right lets Afresh bring together demand from the records that belong together and connect recommendations back to the correct item in a retailer’s systems.

How Afresh turns messy grocery data into item identity

Afresh starts by making sense of the information available about the item itself.

1. Normalize the raw data

Retail item descriptions are often abbreviated, punctuated inconsistently, or cluttered with tags that don’t identify the product. Afresh standardizes the text and strips out information that does not help establish product identity, including operational tags like a markdown flag. Two codes for the same product often differ only in formatting. Left alone, those differences make one product look like two.

ORG BLUEBERR 8OZ MKDN becomes org blueberr 8oz

2. Pull out the attributes that matter

The description mixes what the product is with how it is packaged and sold. Attributes like size and organic status separate products that otherwise look almost identical, so Afresh pulls them out as structured facts rather than leaving them buried in a string.

Afresh extracts useful grocery attributes into their own fields, including sell size, organic status, and whether an item is sold by each. That turns an opaque string into structured information:

org blueberr 8oz becomes:

Product: blueberr

Organic: yes

Size: 8 oz

3. Work out what the product is

What is left is still a merchandiser’s shorthand. Afresh uses AI to interpret it and identify the actual product behind the abbreviation. This allows a truncated or misspelled description to resolve to the same product concept as a correctly spelled one, without someone maintaining a list of every way a retailer might write “blueberry.”

The goal is not simply to clean up a record, but to understand what real-world grocery item that record represents.

blueberr becomes blueberries

4. Resolve it to a consistent Afresh item

The interpreted product and its attributes combine into a single identity: an eight-ounce package of organic blueberries.

Retail records that independently resolve to the same interpreted product and attributes receive the same Afresh item ID, even when nothing in the retailer’s data directly connects them. A conventional eight-ounce blueberry package resolves somewhere else, and so does an organic twelve-ounce package, because the attributes differ.

What happens when a code gets recycled?

Shelves don’t hold still, and neither does the data describing them. Retailers reissue codes. Descriptions get rewritten. Sizes change.

If a code starts representing a new product but remains mapped to the old one, the histories of two different products get mixed together, weakening the forecasts and recommendations built on top of them.

Afresh ingests and reevaluates data and identity clusters every day. When a code starts describing a different product, Afresh catches the distinction and creates a new mapping so the new product’s history doesn’t inherit the old one’s.

Better identity creates a better data foundation for better decisions

Better item identity → cleaner item history → stronger model inputs → better recommendations → better operational decisions

When records that belong to the same item are brought together correctly, Afresh gets a more complete view of that item’s history. When distinct items stay distinct, their demand signals remain clean. That stronger item-level foundation feeds everything that comes next.

A forecasting model cannot recover demand history that has been split across identities it thinks are unrelated. A recommendation engine cannot fully compensate for two different products whose histories have been mixed together. Before either can make a good decision, the underlying data has to describe the business correctly.

That is why Afresh invests in the data foundation beneath the recommendation itself. Forecasting and inventory models tend to get most of the attention in grocery AI. But those models only work with the reality represented in their inputs.

Before AI can decide how much of something a retailer should order, produce, or buy, it has to answer a more fundamental question: What item are we actually talking about?

Item mapping is just one of the ways Afresh helps grocery retailers build an AI-ready foundation that supports better decisions at scale.

Explore the Afresh platform →