A fine-tuned API that reads support tickets and returns the intent — WISMO, returns, exchanges, and six more — in under 10 milliseconds. Built for developers, priced for startups.
No training required, no labeling, no ML expertise. Send a ticket, get back a label your automation rules can act on.
POST the raw customer message to /classify. No preprocessing, no tokenizing. The raw text is enough.
The API returns an intent label, a confidence score, and the full probability distribution across all nine classes. Results arrive in under 10ms.
Route to the right agent queue, trigger a macro reply, set ticket priority, or write the intent as a tag in Gorgias. The label is your decision variable.
Designed from real e-commerce ticket data. Not generic NLP labels — the exact categories your support team already uses.
General-purpose LLMs handle intent classification passably. A model fine-tuned on 46,000 e-commerce examples handles it precisely — at a fraction of the cost.
No seat licenses. No annual commitments. Scales from your first prototype to production traffic.
Send a POST request with the ticket text to /classify and get back the intent label and confidence score in under 10ms. EcomIntent covers the 9 intents that make up the vast majority of e-commerce support volume: WISMO, returns, exchanges, cancellations, damaged items, billing disputes, product questions, account issues, and other.
Yes — significantly. EcomIntent costs approximately $0.001 per 1,000 calls. GPT-4o mini costs $0.015 per 1,000 calls zero-shot, or $0.045 with few-shot examples. That's 15–45× more expensive for a task a purpose-built fine-tuned model handles more accurately anyway. EcomIntent also responds in under 10ms vs. GPT-4o mini's 300–600ms.
The EcomIntent Gorgias plugin (available in the Gorgias App Store) receives new ticket webhooks, classifies the ticket text, and applies an intent tag — ei-wismo, ei-return, ei-exchange, etc. — directly in Gorgias. You then build Gorgias automation rules on top of those tags to route tickets, trigger macros, or set priority. Setup takes under 5 minutes.
On the Bitext e-commerce benchmark (held-out test split), EcomIntent achieves 99.9% accuracy. On real-world production tickets — which are messier, shorter, and more varied than benchmark data — estimated accuracy is 87–93%. We're transparent about this distinction in the full model card. We recommend testing against a sample of your own ticket history before committing.
If the model's top confidence score falls below the threshold you set (default 0.70), EcomIntent returns OTHER instead of a low-confidence guess. This means ambiguous tickets — greetings, vague messages, multi-intent edge cases — are flagged for human review rather than mislabeled. You can tune the threshold per your use case.
Yes. EcomIntent is a standard REST API — one POST request, one JSON response. No model hosting, no GPU setup, no labeling, no fine-tuning required. If you can make an HTTP request, you can use EcomIntent. The Gorgias plugin requires no code at all — just install, enter your API key, and it starts tagging tickets automatically.
Yes. EcomIntent is a generic REST API that works with any helpdesk platform that can make HTTP requests or receive webhooks. Direct integrations exist for Gorgias (App Store plugin) and Shopify Inbox. For Zendesk, Freshdesk, or custom systems, you call the API from your own webhook handler and use the returned intent label in your routing logic.