AI classification for tickets, products and documents
Route tickets, classify products, classify documents by type.
CATEGORIZE · CHOOSE REPLIES · ASSIGN SHAREPOINT & FILE PERMISSIONS · LABEL PHOTOS
From spreadsheets and images
to your own private AI model.
Upload your team’s past decisions from Excel, CSV or Google Sheets, or photos already sorted into your categories, and StayCharted AI Model Trainer (AMT) trains a model that makes them the same way. No code, and tested on examples it never saw before you use it.
Using AI agents? See how StayCharted fits
Also available on AWS MarketplaceMicrosoft Marketplace
- Faster: handle routine work in batches.
- Consistent: apply the categories your team taught it.
- Private: your data trains only your workspace’s models.
Free plan · No code · Your data trains only your models · 93.7% on a public banking benchmark (Dedicated AI Model)
● HOW IT ALL FITS
Your team’s judgment, for every app and agent.
1. Learns from your past work
Support tickets→ Which queue
Leads and messages→ Which reply, who sends it
Invoice lines→ GL account and cost center
Photos and notes→ Pass, or which defect
2. APPLIES IT TO NEW WORK FROM
- AI agents (Claude, ChatGPT)
- Your apps (API)
- Power Automate
- Google Sheets
- Files and photos
StayChartedA private model of your team’s past decisionsAnswers in seconds, with a confidence.✓ Confident: done
Straight into your system, or back to the agent to act on.
Unsure: Review Queue
A person confirms or corrects it, with similar past examples beside it.
Every correction teaches the next version.
Private to your workspaceTested on examples it never sawFiles, Google Sheets, API, Claude and ChatGPT
Train an AI once on your decisions instead of prompting it every time.
Asking a chat assistant to categorize can produce different answers when the prompt or model changes. A trained model learns from your team’s past decisions and applies a published version to new inputs, with a confidence score, without rewriting a prompt for each batch. Review and retrain it when your categories or data change.
On a public banking benchmark with 77 intents, StayCharted’s Dedicated AI Model answered 93.7% correctly on held-out rows. Claude, given a definition of each category, answered 84.7%. Dedicated AI Models require Business; results vary by task. Read the benchmark and method →
The routine goes to the model, and the judgment stays with your team.
Reply selection for leads and customer messages
Choose the reply your team would send, and who sends it, from the replies you already use. It does not make up answers.
AI for SharePoint and file permissions
Access for new files and sites, groups for new starters, owners for new requests.
Multi-field coding
Several fields at once, like GL account and cost center.
Image classification for product and inspection photos
Inspection results and damage types, reading the photo and the note with it.
What your team already decided
| Your examples | What your team decided |
|---|---|
| Support tickets | Which queue |
| Inbound leads and website messages | Which reply, and who sends it |
| Files and SharePoint sites | Who has access, and at what level |
| Invoice lines | GL account and cost center |
| Product listings | Catalog category |
| Inspection photos and notes | Pass, or which defect |
| New starters | Which groups they join |
If your team has done it a few hundred times and can tell when an answer is right, it can probably be a model.
FROM EXAMPLES TO A WORKFLOW
How it works
Upload what you've already categorized.
Upload work your team has already done: a spreadsheet with the answer each row was given, photos in folders (with a note column if you have one), or your SharePoint permission export as it is.
Check your examples.
Before training, StayCharted finds duplicate rows, categories that mean the same thing, categories with too few examples, and personal information such as email addresses and phone numbers. Fix each in a click.
Train and test.
See how often the model is right on rows it never saw during training, which categories it mixes up, and how many rows a person would need to check.
Put it to work.
Fill a CSV or Excel file, fill the empty rows in Google Sheets, call it through the API, or ask it from Claude or ChatGPT.
ONGOING
A person checks what the model isn't sure about.
Answers below the confidence you choose go to a Review Queue. Your team confirms or corrects them, and the next version learns from the corrections.
Or call it from your own code, with one record in and one answer out.
POST https://amt.staycharted.com/api/v1/models/{id}/classify
Authorization: Bearer amt_live_…
{ "input": "I was charged twice this month" }
→ { "label": "Billing",
"confidence": 0.94,
"needsReview": false,
"modelName": "Support ticket routing" }Grouping unlabeled rows, the Review Queue and API access are available on Essentials and Business. Compare plans →
FILE & SHAREPOINT PERMISSIONS
Set up new permissions the way you always have.
Train on your past access decisions. StayCharted suggests permissions for new sites, files and starters, with confidence scores to help you decide what needs review.
Learns your history, not rules.
No permission matrix to write or keep up to date.
Nothing changes until you apply it.
StayCharted only suggests. You approve any suggestion that would widen access.
You stay in control.
You apply what you approve, in SharePoint or with your own scripts.
Choose a classification task
Explore all solutions ↗Support ticket classification
Customer request → Support category
Explore solutionClassify product catalog records
M8 stainless bolt → Fasteners
Explore solutionStandardize category labels
Different descriptions → One category
Explore solutionClassify priority bands
Quote needed by Friday → High priority
Explore solutionMap to internal categories
Source codes → Your internal categories
Explore solutionDocument classification
Supplier agreement → Contract
Explore solutionHow no-code text classification works →Compare trained models with an AI assistant →
VISION · PHOTOS AND TEXT
Categorize product and inspection photos the way your team already does.
Train from photos your team has already categorized: folders of pictures, or a spreadsheet of picture links. Your model can read the text that comes with each photo too, such as an inspector’s note, a customer’s message or a product title. It then categorizes new photos with a confidence for each, and sends the unsure ones to a person.
- Inspections: a photo of the part and the inspector’s note → Pass, Scratch, Dent or Needs re-inspection.
- Damage claims: the customer’s photo and message → Crushed corner, Water damage or No visible damage.
- Product listings: the product photo and title → the right category in your catalog.
Your labeled photos, with their notesYour AI Image Classifier Model
Inspection outcomeDamage typeProduct categoryNeeds review
Photos alone work too. A row with a note but no photo is answered from the note. Needs the Vision add-on; on Business, a Dedicated AI Image Model can learn your photos and notes as well.
Try it with sample parcel photos →
Explore image classification →Start here → Product and inspection photos
SENSITIVE DATA CHECKS · EVERY PLAN
Teach the task and mask sensitive details.
Check text examples for recognizable personal information and secrets before training. Mask detected values, exclude affected rows, or keep them. Models trained on masked text also mask new inputs. Detection does not catch everything; names, addresses and free-text sensitive details need manual review.
See what the checks cover →START WITH ONE RECURRING TASK
Bring examples. See what your model can do.
Your task is a good fit when you have past examples, clear categories and a way to check the answer.
A Free workspace that stays free.
Start with Classifier Models, one AI Classifier Model and file downloads. Paid plans add capacity and capabilities, with no per-seat charges.
Compare text and image models →Is a trained classifier right for your task? →Frequently asked questions
What happens to sensitive information in my training data?
StayCharted AMT checks text before training so you can mask detected values, exclude affected rows, or keep them. Models trained on masked text also mask new inputs. The check does not detect everything; names, street addresses, and free-text medical details still need manual removal.
Can AMT learn from images?
Yes. The Vision add-on enables AI Image Classifier Models on Essentials and Business. Train from categorized photos, then categorize new images through files or your own apps. Business with Vision also supports Dedicated AI Image Models, which train the AI itself on your examples. A picture model can also read a text column with each photo, such as an inspector’s note. Vision recognizes what a picture shows; it is not OCR for reading text inside receipts or forms.
What is AI Model Trainer (AMT)?
AI Model Trainer turns examples of your team’s past work into a private AI model. Upload text or photos with the answer your team gave each one, such as a category, the groups that can open a file, or an account code, and it trains a model that gives the same kind of answer for new ones, with a confidence for each. Use it on a file, in Google Sheets, through the API or from Claude and ChatGPT. Answers it’s unsure about go to a person.
What data do I need to get started?
Start with a spreadsheet or business-system export containing examples and their correct outcomes, such as tickets paired with categories or supplier descriptions paired with approved categories. For images, use a ZIP with one folder per category or a spreadsheet of public picture links and category labels. For SharePoint permissions, your permission export as it is.
Do I need coding or AI experience?
AMT provides no-code AI model training: supply labeled examples, train custom AI models, and review their results without writing training code or needing machine learning expertise. Connecting the Prediction API to your application may require a developer.
Which model should I start with?
Start with the kind of examples you have. For clearly worded text categories, try a Classifier Model. When wording varies, consider an AI Classifier Model. For pictures, start with an AI Image Classifier Model through Vision. If your task needs finer distinctions, compare a Dedicated AI Model using separate test examples. Choose based on the errors that matter in your workflow.
How sure is each answer?
Every answer comes with a confidence from 0 to 100%. Confidence is a guide, not a guarantee. Choose a cutoff to flag uncertain answers for review. On Essentials and Business, answers below it wait in the Review Queue for a person; your integration decides how to use answers in another system. Before relying on a model, inspect its held-out accuracy overall and per category, and the share of answers that need checking at your cutoff.
How is this different from Jev or a chat AI?
Jev and chat AI can use instructions to categorize text without training on your examples. StayCharted can train on your team’s reviewed decisions and report held-out accuracy, with a Review Queue on Essentials and Business.
Is my data private?
Yes. Models you build are private to your workspace. The shared, ready-made AI behind AI Classifier Models and AI Image Classifier Models does not learn from your uploaded data. Dedicated AI Models are trained on your examples for your workspace.
Can I start for free?
Yes. Free includes Classifier Models, one AI Classifier Model and file downloads. Essentials adds more AI Classifier Models and API access. Dedicated AI Models are available on Business and above.
See the results for yourself
All benchmarks →BENCHMARK
Part numbers from a photo: 100 look-alike parts, measured
StayCharted identified 87.8% of 1,000 unseen photos across 100 part numbers; 99.7% accuracy on the 32.3% above confidence 0.5. Compare both image models.
Read the results →BENCHMARKHow accurate is AI HS code classification? We measured it on 6,788 CBP rulings
We trained two models on 5,621 CBP rulings and tested them on 1,167 later ones. Accuracy at 10, 6, 4 and 2 digits, and how often the surest answers are right.
Read the results →BENCHMARKCar photos for claims, service and OEMs: damage, damage type and part, measured
Compare StayCharted’s image models on car damage, 21 exterior parts and eight damage types. See measured accuracy, rare-category errors and review projections.
Read the results →STAYCHARTED AI MODEL TRAINER
Your team’s past work, doing the next round.
Start with examples you already have, see how often the model is right on ones it never saw, and let people check only the answers it’s unsure about.
Start freeYOUR AI PLATFORMS
Use the AI your company already runs.
Connect your own AI platform and test it with your team’s past decisions, without training. On Business, you can also train a Dedicated AI Model in your own Amazon Bedrock, Microsoft Foundry or Google Cloud account.
Compare the model options →