AI để tuân thủ thuế bán hàng

Before 2018, the Quill decision meant a state generally could require a seller to collect sales tax only if the seller had a physical presence there. In South Dakota v. Wayfair (2018), the Supreme Court overruled that standard. It upheld South Dakota's law requiring collection by remote sellers with more than $100,000 in sales or 200 transactions in the state. Nearly every state with a sales tax adopted an economic nexus rule afterward. Thresholds and the sales that count toward them vary, and many states later dropped the transaction count. Most states also passed marketplace facilitator laws, which move the collection duty for sales made through platforms like Amazon or Etsy to the platform. Compliance involves three hard problems. The first is taxability. Whether an item is taxable depends on the state and on legal definitions. Clothing is exempt in some states. Groceries are often taxed at a reduced rate or exempt, while candy and prepared food may not be. Under the Streamlined Sales and Use Tax Agreement definition, candy that contains flour is not classified as candy, so it can be taxed like food instead. The second is jurisdiction and rate. Thousands of state, county, city and special-district rates exist, and they change regularly. The third is filing: registrations, returns on different schedules, and managing exemption certificates for resale and nonprofit buyers. Vendors such as Avalara, Vertex, Sovos, Stripe Tax and TaxJar automate parts of this. AI is most visible in product classification and in sorting messy transaction data. Rate calculation and filing rely mostly on maintained rules and content databases. One misconception is that automation transfers liability. It does not. The seller remains responsible for collecting and remitting the correct tax. Another is that small sellers are exempt everywhere; they are exempt only below each state's threshold.