States and cities write the categories. Let residents, vendors and businesses check their answer against your own decisions.

• Sales and use tax product categories. Whether a product is taxable depends on how the state defines it: prepared food, candy, clothing, digital goods. Streamlined Sales Tax states publish their treatment of each definition in a taxability matrix, and each seller still has to decide which definition their product falls under.
• Procurement commodity codes. Many states and cities classify purchases with NIGP codes and match vendors to bids by the codes vendors pick at registration. Texas, for example, tells vendors to choose the class and item codes closest to what they actually provide. A vendor who picks the wrong ones may never hear about the bids they could win.
• Business license and registration categories. A new business describes what it does and picks a category, which decides fees, inspections and which office it deals with.
• 311 and service request types. A resident describes a problem in their own words and picks from a long list. A wrong pick means a request that waits in the wrong department's queue.
• Permit types. A homeowner or contractor describes the work, and the permit type decides the review path.

The training data is what staff already did: requests rerouted to the right type, registrations corrected, taxability questions answered, purchases coded. Each is a description and the category staff decided.

StayCharted AI Model Trainer learns from it. It checks the examples first (the same description under two categories, categories with too few examples), then tests the model on cases it never saw. The report shows how often it's right for each category, which ones it confuses, and what share of answers are right above each confidence. Staff publish the model when they're satisfied with the report.

Below the cutoff, the government decides what a person sees: "We're not sure. Choose from these categories or ask us," or a link to the formal process.

The public form is an integration your web team builds, not a hosted public StayCharted page. Keep API credentials on the server and add input limits and rate controls. API and mobile answers below the model’s cutoff enter the Review Queue on Essentials and Business. Access codes also require the Owner’s permission; all use counts against the owning workspace’s allowances. Agree capacity for a public service with us.

Reviewing a case improves the next model version; it does not automatically notify the visitor or send them a revised answer. Your website must provide any follow-up process.

Many categories are set once and applied in many places. A state's tax definitions apply to sellers in every city. A state's commodity codes are used by its agencies and often by cities, counties and school districts that buy through state contracts. A county's permit types may be shared by its towns.

The agency that owns the standard trains and publishes one model. Every agency, city and county that uses the standard asks that same model, through the API or an access code of its own, and every unsure answer comes back to the owning agency's Review Queue, marked with where it came from. Each office gets the same answer for the same description, and the owner sees where the standard is causing confusion.

This uses the owning agency’s workspace and API credentials, with access controlled by that agency. It is not cross-organization model sharing. An office that needs its own Review Queue would train its own model in its own workspace.

Consistent answers, from staff practice. The same description gets the same category, whoever asks and wherever they are.

Hard cases in one queue. Answers below the cutoff go to the Review Queue, marked with the channel and office they came from. Staff see which products, activities or problems the public can't place.

A model that improves as staff decide. Each decision in the Review Queue becomes a training example. Publishing the next version updates every channel at once, and the published version keeps answering while the next one trains.

Review and activity records. Review items retain their confidence and source channel. Activity records training, publication and other changes; agree the retention and audit requirements for your program before deployment.

Three of our public benchmarks are close to these jobs:

These benchmarks do not measure a state or local government’s own task; NHTSA is public federal data. Your categories may overlap more, or less. The report on your own decisions is what decides.

• Not a ruling or a determination. A suggested tax category, license type or commodity code is guidance from past staff decisions. It doesn't change who is responsible for a filing, and it doesn't carry the liability relief a state may give for relying on its published matrix. The government says so on its page, in its own words.
• Not a replacement for staff. It sends the hard cases to them, and their decisions train it.
• Not a reader of the law. It learns patterns from decisions. When a description leaves out facts that decide the category, a person should decide. Low confidence can flag uncertainty, but incomplete descriptions can also receive confident wrong answers.

• A database schema of your own, holding only your organization's data, with each workspace (a department, say) walled off.
• Encrypted at rest and in transit, hosted on AWS in the United States.
• Used only for your models. Your data never trains anyone else's.
• Your own cloud, if preferred. On the Business plan, models can be trained in your own Amazon Bedrock or Microsoft Foundry account.
• Bought on AWS Marketplace, on your AWS invoice, or directly.

Security and your data →

StayCharted is not FedRAMP or GovRAMP (formerly StateRAMP) authorized. Marketplace availability and training in your cloud do not provide those authorizations. Data still passes through StayCharted for the cloud-training workflow. Before a pilot, the agency should approve the data, privacy notice, retention and security arrangements; public submissions can be stored as predictions or review items.

1. Weeks 1 to 2: pick one standard and one office that owns it. Export a few thousand recent decisions: the description and the category. Remove names, addresses and tax IDs.
2. Weeks 2 to 3: train, read the report per category with staff, and set the cutoff.
3. Weeks 3 to 6: staff use it first on incoming cases, and decide the unsure ones in the Review Queue.
4. Weeks 6 to 8: open it to one outside group: a vendor registration page, a 311 form, or one city using the state's model. Compare the report on the newest decisions with the first.

Start free → · Talk to us about a state or city pilot →