Set up conversation evaluations
Add the questions the AI answers from each call and thread, and choose how each answer is recorded.
Last reviewed November 7, 2026
Where evaluations live
Each campaign has its own list, on the Conversation intelligence step under Advanced in campaign setup. A new recovery campaign starts with nine:
| Evaluation | Answer format |
|---|---|
| Conversation summary | Text summary, up to 400 characters |
| Purchase intent | Classified: ready_to_buy, considering, not_interested, unclear |
| Main objection | Classified: price, shipping_cost, delivery_time, product_fit, trust, timing, technical_problem, none_given, other |
| Discount offered | Yes / No |
| Checkout link sent | Yes / No |
| Asked for a call back | Yes / No |
| Asked for a person | Yes / No |
| Asked not to be contacted | Yes / No |
| Conversation quality | Number, 1 to 10 |
Add an evaluation
- Open the campaign's setup, expand Advanced and choose Conversation intelligence.
- Select New evaluation.
- Give it a Name, up to 120 characters. Reports label the answers with it.
- Pick the Answer format, then fill in what that format asks for (see below).
- Write the Question for the AI, up to 2,000 characters. Ask one thing and say what counts, for example: "Did the shopper say shipping cost stopped the purchase? Answer yes only if they said so."
- Select Add evaluation, save the step and publish the campaign.
A campaign can have up to 30 evaluations. Use the row menu to Edit or Remove one.
Answer formats
| Format | You set | Counted on Interaction Evaluations |
|---|---|---|
| Yes / No | Nothing more | Yes, as a yes and no split |
| Classified | 2 to 20 labels; press Enter after each. The AI picks one per conversation. | Yes, per label |
| Number | Lowest and highest score, 1 to 10 unless you change them | Yes, per score, with the average |
| Text summary | Longest answer in characters, 400 by default and at most 4,000 | No; read them one by one |
Text summary also has a Shortest answer field. It is saved, but nothing enforces it. When you want something you can count, pick Yes / No, Classified or Number.
Publishing starts the scoring
New and changed evaluations score conversations from the moment you publish. Earlier conversations keep the answers they had, and none are scored after the fact, so add your questions before the campaign goes live.
Editing an evaluation keeps it as the same question, so old and new answers stay together under one name. If you change what a question means, add a new evaluation instead. A removed evaluation stops being asked once you publish and drops off the Interaction Evaluations tab.