Most US retail spending still happens in physical stores, yet much of that inventory is invisible to AI. The web is indexed. The shelf isn't. Local inventory data is the missing layer, and it may be the biggest catalog nobody has built.
The number that frames it
According to the US Census, e-commerce was 17.1% of US retail sales in Q2 2026.
Read that the other way. Roughly 83% of retail sales still happen somewhere else, and much of that is physical places: shops, shelves, counters, back rooms. That's where the products are. It's also where the data is thinnest.
The online share keeps growing, but the point stands. If you want an agent to help someone buy what they need, you can't stop at what's on a website.
Shelves change by the hour
Even when a store has a website, it rarely tells the whole story.
What's on a shelf changes all day. A shopper picks up the last one. A delivery lands. A price gets updated in the register but not on the page. A product sold out an hour ago and still shows as available, or the reverse.
No crawler sees any of that. A web index is good at pages that stay put. A shelf doesn't stay put.
So a catalog of physical inventory can't be a snapshot you take once. It needs to say where each fact came from and when it was last seen. A price from this morning and a price from last month are not the same kind of fact.
Why it matters more for agents
When a person searches, they can tolerate a miss. They call the store, walk in, shrug and try somewhere else.
An agent doesn't have that slack. If it tells a shopper "it's available nearby" and the shelf is empty, the shopper pays for the mistake. If it can't see local stock at all, it defaults to whatever it can see, which means the same few online sellers every time.
That's a quiet loss for everyone else. Stores that carry the item, close to the buyer, simply never come up.
Urgent needs make this sharper. A phone charger, a pair of shoes for tomorrow, a part for a repair. Those are the moments when "close and in stock" beats "cheapest somewhere."
How JENNi builds local inventory data
We build it as one shared graph, the SKU Graph. It joins three things:
- Canonical products, anchored on barcodes (UPC/GTIN), so one item has one identity no matter where it's sold.
- Store offers, which are the stores that carry each product, with price, stock, source and when we last saw it.
- Confidence, meaning how sure we are right now.
Freshness is part of the data, not an afterthought. Every offer carries when we last saw it, so its age is never hidden.
Orders close the loop. When an order is delivered, the graph learns the offer was good. When an order fails, it learns that too. Over time, the answers get better because real purchases keep correcting them.
Search results show in-stock items only. And we don't claim to cover every store or every product. The graph grows as sources and orders do.
What makes a shelf index different
A web index can be rebuilt by crawling again. A shelf index can't, because the thing it describes isn't a page.
Three habits make it work. First, tie every offer to a barcode, so the product is the same product everywhere. Second, record the source of each fact, so we know whether a price came from a store feed or from something else. Third, keep the time of the last sighting next to the fact itself. With those in place, an agent or an app can decide how much weight to give an answer, instead of treating every answer as equally true.
From index to order
An index is only useful if it leads somewhere. On shop.jennipro.com, shoppers pick a product, not a store, and every offer reads "Delivered by JENNi." Agents can reach the same catalog through our MCP server at https://shop.jennipro.com/mcp.
We ship anywhere in the US. When a store with the item is within about 30 miles, local delivery is offered, and an order can arrive the same day when the quote says "Get it today."
The shelf is next
The web took years to index. The shelf is the next big surface, and it matters most to the people who make and sell things in the real world.
Want your shelves visible to AI agents? Get in touch at jennipro.com.