Xumulus Logo

Ecommerce for large catalogs

Past about ten thousand SKUs, the ecommerce platform stops being the centre of the architecture. Product data lives in an ERP or a PIM, search lives in a separate index, pricing comes from the ERP, and the platform takes the order. The asset is the index where all of that converges. This is how to build around it, whatever platform you are on.

The short answer

A large catalog is not a bigger version of a small one. It is a different problem. The platform admin cannot be where product data lives, the product model cannot hold the attributes that matter, native search cannot answer the questions customers ask, and every channel needs a different shape of the same data. Teams solve each of these separately, with an extension, a plugin, and a sync job, and end up with five copies of the truth that disagree.

The alternative is to treat the searchable catalog as infrastructure: one index, built deliberately from the systems that author product data, that every surface reads from. The storefront, site search, feeds, structured markup, and whatever answers customer questions next all become readers of the same thing. The platform becomes one of them, and its job narrows to checkout.

This is platform-agnostic on purpose. It applies on Shopify, BigCommerce, Adobe Commerce, or a custom stack, and it is the approach behind Neuron, the index-first product we are building.

Discovery bottlenecks

What actually goes wrong

These arrive in a predictable order, and the first three are usually already happening by the time anyone asks for help.

01
The admin is a lossy copy
Nobody edits ten thousand products by hand. Imports become the interface, and the truth moves upstream to a system the platform only mirrors.
02
Attributes have nowhere to live
Fitment, units of measure, compatibility, contract pricing, and warehouse stock do not fit a model built for sizes and colours.
03
Native search hits its ceiling
Part numbers, unit variants, synonyms, and anything that depends on what the customer owns are out of reach. A vendor gets added, and a second copy of the catalog with it.
04
Category pages fall over
Facet counts across fifty thousand products are expensive, and the cache holds until a sale, a price update, or a crawl invalidates it all at once.
05
Every channel is another copy
Google Shopping, marketplaces, and feed tools each transform the catalog their own way. Five channels, five versions of the truth.
06
Pricing goes back to the ERP
Contract and tiered pricing and real multi-warehouse stock are questions only the ERP can answer. The platform becomes a middleman for the number that matters most.
Key takeaway
None of these is a bug. They are what happens when a catalog outgrows the system it was loaded into.
The stack

The three layers to own

Own these and the platform becomes a component you can swap. Rent them and the platform owns you.

01
Data

The index

One canonical, enriched representation of the catalog, built from the ERP, the PIM, and supplier feeds.

  • Derived facts: fitment resolved, units normalised, relationships explicit
  • Every surface reads from it, so a fix propagates once
  • Portable across platforms, which is what makes replatforming cheap
02
Retrieval

Search and answers

Site search, category pages, feeds, and whatever answers customer questions next, all served from the index.

  • Relevance and merchandising logic in version control, not a vendor dashboard
  • No per-request meter that grows with success
  • Ready for agents and answer engines, because the facts are structured
03
Storefront

The front end

Reads product, category, search, pricing, and availability from the index, and nothing product-shaped from the platform.

  • Cheaper than it has ever been to build, and getting cheaper
  • Where brand and conversion live, so it should be yours
  • Platform reduced to cart, checkout, payments, and orders

Tell us about your catalog

Send us your SKU count, where product data is authored today, and which channels read it. We will tell you what an index-first build would look like for you, including when the honest answer is that your current setup is fine.

Search is the foundation of AI commerce