AI that identifies what sold — and admits when it doesn't know
Barcode decoding and vision identification on every logged sale, purchase orders parsed into real inventory, and a live coach reading the show as it happens. Strict thresholds mean unknowns get flagged, never quietly costed wrong.
- Barcode and UPC decoding from stream frames
- Exact-identifier matching before any name guessing
- Purchase orders parsed into SKUs and lots
- Live coaching suggestions during the show
3-Day Free Trial · 30 Logging Credits Included · No Card Required
How identification actually works
The order of operations is what keeps your COGS trustworthy.
- Step 01
Read the identifier
If a barcode, UPC or cert number is visible in the captured frame, it's decoded and looked up directly — an exact match beats every kind of guessing.
- Step 02
Ground the candidates
With no identifier, the model is given a shortlist of your own SKUs and lots and must choose one of them or return nothing. It can't invent a product that isn't in your catalogue.
- Step 03
Enforce the threshold
The pick still has to clear a strict similarity bar and agree on category. Anything below the bar is flagged for review rather than assigned a cost.
Wrong COGS is worse than no COGS
A matcher that always answers will confidently price a $40 card at $0.71 and quietly corrupt your margin history. Streamalytix is built to refuse instead — a flagged sale takes you ten seconds to fix, a wrong one takes months to notice.
- Barcode and UPC decoding from the captured frame
- Containment-aware name similarity with token tolerance
- Hard category-mismatch rejection
- Corrections remembered by stable identifiers only
AI on the boring work and the live decisions
Two places AI earns its keep: turning a supplier invoice into inventory in seconds, and telling you mid-show that one category is dragging your average down.
- Purchase order and invoice parsing into SKUs and lots
- Freight allocated proportionally across parsed lines
- Live coach comparing tonight to your own history
- Sourcing suggestions from category sell-through and margin
How the AI is kept honest
Identifier priority
Exact UPC, barcode or cert lookups always take precedence over similarity scoring.
Grounded choices
The model picks from a candidate list of your own inventory or returns null. No free-form invention.
Category guards
A candidate from an incompatible category is rejected outright, no matter how close the name looks.
Frame capture
Photos and frames are analysed per sale, so each item is identified from its own evidence.
Alias discipline
Aliases are added only when you correct a sale or an exact identifier confirms the match.
Live coach
Suggestions during the show based on pace, margin and category performance versus your history.
Straight answers
If something isn't covered here, the team replies within a business day.
Three jobs: reads barcodes and product details from stream frames to identify what sold, parses purchase orders and invoices into inventory, and suggests adjustments live based on how the show is performing.
Let the AI do the identification, not the guessing
Upload a purchase order and run a show during the free trial to see the whole pipeline end to end.
