AI order intake#

StackCube uses AI to prepare order candidates, not to ship orders automatically. The system extracts structure from messy B2B orders so reviewers can make faster and safer decisions.

StackCube incoming channel history — Incoming email and KakaoTalk messages are kept with source text, channel, sender, status, and the generated order link.

What AI extracts#

AI can identify:

Customer and contact
Item names and aliases
Quantities and units
Requested delivery date
Shipping notes
Attachment context
Possible duplicate orders

Candidate creation#

An order candidate is a draft created from raw source data. It is not an approved order.

The candidate includes extracted fields, matched item suggestions, source evidence, and review flags.

Matching logic#

AI extraction is combined with deterministic catalog and pricing rules. This matters because item and price decisions should be explainable during review.

For example, a candidate may combine an AI-read buyer item name with a catalog alias and a customer price tier rule.

Data handling#

Order data is processed to provide StackCube functionality. It should not be treated as public training data. See security for the operational data-handling model.

Tuning#

The fastest way to improve AI intake is to correct real candidates, then add catalog aliases, customer price tiers, or tier rules that explain those corrections.