Introduction
Welcome to our Trackers & Feedback Loop guide. This document provides general usage instructions and answers to common questions about how our AI system utilizes feedback received from you to improve future detections.
Instructions
Upon reviewing an interaction, on the “Scorecard” panel you will find all the detected trackers.
These trackers were detected by our AI model, which continuously refines itself based on the feedback received from our users.
To help us improve our models and better customize them for you, you can head to the left “Actions” panel, and tag each tracker with either:
One-click Thumbs Up - for correct detections.
One-click Thumbs Down - for wrong detections.
Double-click Thumbs Up - for detections that are out-of-context(*).
Out-of-context detections are trackers that could be a correct detection, given another context. For example, if a customer says “their company is a scam” we don’t want it to trigger our system, but if he says “your company is a scam” we might want it to be detected.
You can also tag phrases that our AI system failed to detect. To do so:
Open the “Transcription” panel, and go to the desired text location.
Mark the desired text. Please make sure you mark the whole phrase and not just part of it. A rule of thumb is to include at least 3 words, and no more than 8.
Click on the “Tag” icon, and select the right tracker for the marked phrase.
FAQs
How does the AI Trackers train itself?
In order to train our models we run a Feedback Loop which continuously collects labeled data based on our users’ feedback and retrains the models according to predefined policies.
It's an on-going process, designed to keep our AI models up to date with new trends and regulations.
How long does it take to improve the tracker detection models?
Usually it takes 4-6 weeks to stabilize the system, but it varies according to the amount of feedback we get from our customers.
I have tagged multiple trackers, but the system still doesn’t detect them. What should I do?
Training the models and deploying them may take up to 5 days. During that time, the system won’t detect newly defined trackers.
have fully reviewed 10 interactions, but the system still detects false trackers. What should I do?
Training the models and deploying them may take up to 5 days. During that time, changes won’t take place.
I want to tag a long phrase that appeared in the transcription but wasn’t detected. What should I include in the tag?
It is important that the essential part of the phrase will be included. A rule of thumb is to tag phrases of at least 3 words, and no more than 8 words.
Will the new models affect the detections in old interactions?
Once trained, new models affect only new interactions. Therefore, detections in old interactions won’t be changed.