Written by: Nicole Ogonowska, IT Growth Manager, Digital Colliers
Whatnot moved around $8B in goods in 2025 and is tracking toward $1B+ in revenue in 2026. That is not a content story. Plenty of brands have run live shopping shows. The story is that Whatnot built the data spine underneath the show, and most traditional retailers did not. If you cannot join a live session to a customer LTV record, you are running live commerce blind, and the economics catch up with you fast.
What a joined live-session-to-customer model actually needs
The data model is not exotic. It just has to exist and be queryable in near real time. At minimum you need:
- A session ID for every stream, with host, start and end time, product catalogue shown, and the exact timestamp each SKU went live.
- A viewer identity graph that ties anonymous stream viewers to logged-in customer IDs, including guest checkouts stitched back later via email or phone.
- Event-level engagement: joined, watched duration, chat messages, add-to-cart, purchase, all bound to the session ID and the SKU timestamp.
- Post-purchase signal joined back to the session: returns, refunds, second orders, subscription retention, LTV cohorts.
That is the spine. Once it exists, you can ask the questions that matter. Which hosts drive repeat buyers, not just first orders. Which SKUs get bought on stream and returned at double the site rate. Which stream segments correlate with a 90-day LTV bump. Whatnot lives inside this data by default because they built the platform. If you are a retailer plugging a live tool into Shopify plus a CRM plus a returns app, nobody owns the join.
Why most brands have not built it
The honest answer is that live commerce sits between three teams that do not share a schema. Marketing owns the stream tool. eCommerce owns the storefront and orders. Data owns the warehouse and LTV models. The session ID from the live tool rarely lands in the warehouse in a shape anyone can join to orders. Guest checkouts fragment the identity graph further. Returns data lands 30 to 60 days later in a different system entirely.
So the reports you get are vanity reports. Peak concurrent viewers. GMV during the stream. Chat volume. None of that tells you whether the show made money once returns and paid acquisition are subtracted. And the ones that do try to build it often stall in the same place enterprise AI projects stall, with roughly 95% of them never reaching production because the data plumbing was underestimated.
The cost of running live commerce without the spine
This is where the left-behind risk gets concrete. Online return rates already run around 19 to 20% of gross sales, and apparel runs higher. Live commerce skews toward apparel, beauty and collectibles, so your stream returns are almost certainly worse than your site average. If you cannot attribute returns back to the session and the host, you keep booking hosts whose GMV looks great and whose net contribution is negative.
Stack that on the wider DTC picture. Customer acquisition cost across DTC has risen roughly 40% since 2023, Meta CPMs kept climbing through 2024 and 2025, and UK eCommerce is growing at about 3% year over year. Single-digit growth is the new baseline. In that environment, roughly 30% of SKUs at a typical multi-channel brand already lose money per order once returns and ad spend are counted. Live commerce without a data spine adds another unmeasured channel to that pile.
What the operators shipping this in 2026 tend to do
The pattern is not glamorous. The teams pulling ahead treat live commerce as an event stream, not a campaign. A few things they get right:
- Session ID is a first-class column in the warehouse, populated in real time from the streaming tool's webhook.
- Identity resolution runs on a schedule, not just at checkout, so guest viewers get stitched to customer records within a day.
- Return data is joined back to session and SKU-timestamp inside 72 hours, not at end of quarter.
- Host performance is measured on 90-day contribution margin, not stream GMV.
None of that requires a new AI model. It requires somebody to own the join. If your team is still exporting stream reports to a spreadsheet and reconciling by hand, you are not competing with Whatnot on content. You are competing on data, and you are already behind.