Answers worked out in-house before an AI engine is asked
Where the company’s own history is enough, Tranxact now answers with models trained in-house on that history and asks an AI engine only when they are not sure: reading social posts and reviews, scoring a candidate against a job order, and suggesting an answer to a support ticket.
| 어디 | What answers first | When an AI engine is still asked |
|---|---|---|
| CRM › Social signals | A classifier trained on the company’s own signals that people read or corrected (intent, sentiment, severity, urgency). | When it is less than 80% sure, when the model has too little history or its accuracy is too low, and always for safety, compliance or critical posts. |
| Talent › Candidates › Summarise | The résumé screen’s own rules give the fit score, top skills, years of experience and education. | Only for the written summary, and only when an engine is configured; otherwise a sentence is built from the facts. |
| Help › Raise a ticket (suggested answer) | An earlier ticket on the same screen with nearly the same words that was answered well. | When no earlier ticket is close enough, or two close ones were answered differently. |
- Each answer says how it was reached: the probability, the words that decided it, and the model’s measured accuracy.
- The models learn only from people’s readings and confident engine readings, never from their own answers.