01
One model does not always fit every customer
This is a common problem for vertical SaaS products. Your application may serve hundreds of companies performing essentially the same workflow. But those companies often organize their information differently. Consider a field service platform. One customer may categorize work as: Repair / Installation / Maintenance Another might use: Emergency / Standard / Inspection / Warranty
A generic model would have to somehow understand every customer's internal terminology.
02
Give each customer its own model
With StayCharted AMT, each customer can train from its own labeled examples. Conceptually:
- Your SaaS application selects the appropriate customer model: Customer A, Customer B or Customer C.
Your application knows which customer is making the request. It sends the prediction to the appropriate model.
03
One product integration
Your engineering team doesn't need to build a separate AI architecture for every customer. The application integration can remain consistent:
- Record + customer
- Your application
- StayCharted AMT API
- Appropriate customer model: A, B or C
- Prediction
What changes is the customer's model—not your core product code.
04
A better way to offer customization
SaaS products often deal with customer differences through:
- Large rule engines
- Custom configuration
- Professional services
- Customer-specific code
- Hard-coded lookup tables
Some of those will always be necessary. But when the customization involves interpreting text or examples, a customer-trained model may be a much simpler option.
05
Make AI a capability inside your product
Your users do not necessarily need to know anything about machine learning. From their perspective, the workflow could simply be:
- Upload examples.
- Tell the application which column contains the answer.
- Train.
- Review results.
- Turn the model on.
StayCharted AMT handles what happens behind that experience.