When using Einstein Reply Recommendations, it’s helpful to understand how we identify replies and recommend them to service reps at runtime.
Required Editions
| View supported editions. |
Identifying Reply Candidates
In Einstein Reply Recommendations, Einstein learns from your past chat transcripts and clusters similar, common service rep utterances. Einstein identifies potential reply templates from these clusters of similar messages. Select and edit useful candidate templates before publishing them to Quick Text for your service reps to use in chats with customers.
Recommending Replies at Runtime
At the heart of Einstein Reply Recommendations is real-time natural language processing (NLP), which Einstein uses to identify data patterns in live chats. When you train a model, Einstein analyzes past chats and learns the patterns of service rep utterances and the conversation that preceded them. Einstein uses this combined understanding of each utterance and the context of utterances in sequence to predict which of your published reply templates is the most similar to what service reps said in past chats. Einstein then recommends that reply to service reps in real time.
Things To Keep in Mind
Your service reps are on the front lines helping customers every day. They know the ins and outs of your products, common problems and quick solutions. So it's key to involve service reps in identifying ideal reply templates. If templated replies don’t mimic actual, past replies, they’re less likely to be recommended.
Try to avoid lengthy or overly specific templates that are rarely seen in chats. Make replies broad so that service reps can apply them to different customers or issues. Use merge fields for dates, names or products to keep them broadly applicable. To save service reps time by automating common responses, design replies to reflect situations that your service reps encounter regularly.