Banking Intent Classifier

Runs inside your own AWS account: your text is processed on a SageMaker endpoint you control, inside your VPC, and is never sent to a third-party API.

This SageMaker model package provides a REST API to classify customer messages into 77 fine-grained banking intents (card issues, transfers, top-ups, account access, fees, and more).

Use it to route support tickets, power chatbots and IVR systems, or analyze contact-center transcripts. Each prediction returns the intent label and a confidence score.

The API accepts input as JSON or CSV and supports real-time endpoints and batch transform.

We welcome your feedback at [email protected]

Model and training data The model is a ModernBERT-base (answerdotai/ModernBERT-base) text classifier fine-tuned with PyTorch and Hugging Face Transformers on the public banking77 dataset. It runs in network isolation on SageMaker, so no data leaves your account.

Known limitations - English only. Other languages are not supported. - The model was fine-tuned on inputs up to 512 tokens. Longer text is accepted, but accuracy is only characterized to that length; split long documents into passages before sending. - The label set is fixed to the 77 banking77 intents. Messages outside retail banking are still mapped to one of them, so use the confidence score to route low-confidence traffic to a human. - Single-intent: a message covering two intents returns only the strongest.

Measured performance - Accuracy 92.2 percent, macro-F1 0.923, measured on 2,616 held-out examples from banking77. The split is seeded (70/20/10, seed 42) and was not used in training. - Real-time endpoint on ml.m5.large: 252 ms median per request (267 ms p95). - Throughput on the same instance: about 24 texts per second at a batch size of 32.