How Scorecards Work
A scorecard is an attribute of type Scorecard with 1 to 50 items. Each item scores one attribute:
The score is the weighted average of the items that earned points. botBrains multiplies each item’s points by its weight, adds the results, and divides by the sum of those items’ weights. Items with weight 0, values without points, and empty values drop out, so they don’t pull the score down. Unanswered items are the exception: they hold the score back. If no weighted item earned points, the conversation has no score.
This conversation scores (100 × 2 + 50 × 1 + 100 × 1) / 4 = 87.5. If the refund policy item were No at 0 points, the conversation would score 0 because the item is critical.
Points for Numbers
For Integer and Decimal attributes, you give points to anchor values and botBrains interpolates linearly between them. A value outside the anchors gets the points of the nearest anchor. For example, with CSAT Rating anchors 1 = 0 points and 5 = 100 points, a rating of 4 earns 75 points.Unanswered Items
A manual item is a scorecard item whose attribute uses the fill mode Manual. While a manual item has no value, or a critical item has no value because the AI couldn’t determine one, the scorecard has no score and the details panel shows Awaiting review instead. Score filters and monitor thresholds leave the conversation out until then. An active monitor whose scorecard has a manual item works as a review queue, where people answer the missing items. What counts as an answer. A person sets the value or clears it, for example with N/A in a review queue. The AI answers when it fills the value or decides that no option applies. Critical AI items. A critical item the AI fills holds the score back only when the AI couldn’t determine a value for a conversation in the attribute’s selection. A person then has to answer it. Critical failures. A critical item at 0 points sets the score to 0 right away, even while other items wait for an answer. Built-in attributes. Built-in attributes never hold the score back. A conversation without a rating skips that item. Existing scores. A manual item holds back the score of every conversation. Adding one to a scorecard that’s already in use, or switching an item’s attribute to Manual, removes the score from all conversations until someone answers.Creating a Scorecard
Click New scorecard on the conversation attributes page, or open the scorecard editor directly. Add items, pick an attribute for each, and set its points, weight, and critical switch. New items start with weight 1, and Yes / No items start with Yes = 100 and No = 0. Each item needs at least one value with points, and at least one item needs a weight of 1 or more.
A new scorecard scores a conversation the next time one of the changes below happens to it. To score existing conversations, click Recompute on the scorecard. You can’t delete an attribute while a scorecard uses it, or a scorecard while a monitor uses it.
When Scores Update
botBrains recalculates scores after each of these changes:CX Score
The CX Score is a ready-made scorecard that estimates the customer experience. Install it from the conversation attributes page by clicking New attribute and choosing CX Score under Templates. It adds six AI signals with equal weight, such as customer effort and AI agent quality, which the AI fills after conversations with at least two customer messages. You can rename the CX Score, but you can’t change its items or points. Installing it doesn’t score past conversations; backfill the signals to score them.Using Scores
Details panel. The Scorecards section of a conversation’s details panel shows each score as a pill: red below 50, amber below 75, and green from 75. A dash means the conversation has no score. Click the pill to see each item’s value, points, and weight, with critical items marked.
Conversation filters. Filter the conversation list by a score range with the scorecard’s Min and Max fields. The URL keeps the filter, so you can share the filtered view. To build such links yourself, see Filter by Attributes.
Monitors. Flag conversations that score below a threshold or fail a critical item. See Flag Criteria.