This is where the article content will be displayed. The rich text field from the CMS collection will be bound here to show the full article body with formatted text, images, and other rich content.
General settings
When you create a new grid, start by defining its basic parameters:
Title: give your grid a clear name, ideally reflecting its purpose — for example "Support call quality" or "Commercial email compliance"
Agents and teams: choose who the evaluation targets. You can target specific agents, entire teams, or opt for anonymized analysis if you want to evaluate overall quality without individual attribution
Filters: refine the data feeding the grid. You can filter by channel, period, conversation type, language, or any other available metadata — useful to only include truly relevant data in the grid
Sections and criteria
A grid is organized into sections, which group criteria thematically. For example, a "Commercial posture" section could contain criteria on active listening, value proposition, or objection handling.
Each criterion includes three elements:
A title / question: formulated as a clear question — "Did the agent propose an appropriate solution?"
An evaluation prompt: the instruction given to the AI so it knows exactly what to evaluate. Be precise and contextualized (1000 characters maximum)
A scoring system, choosing from three options:
Yes / No — for binary criteria, this is the default option
Numeric scale (1-10) — for a graduated evaluation
Performance level — to freely define your own output labels, for example: Excellent, Good, Partial, Poor, Not applicable
Score weighting
Each criterion is assigned a number of points at creation. The rule is simple: the sum of all criteria must equal 100 — you cannot save a grid until this condition is met. This forces you to think from the outset about the relative importance of each criterion.
Weighting works with all score types:
Yes / No: you define how many points a Yes is worth and how many a No is worth — for example 10 points for Yes, 0 for No
Numeric scale: you assign a maximum number of points to the criterion, and the AI's score is then converted proportionally — if the AI evaluates 8/10 on a criterion worth 15 points, the criterion will contribute 12 points
Performance level: each label you've defined is assigned a point value — for example Excellent = 20pts, Good = 15pts, Partial = 8pts, Poor = 0pts
At the end of an analysis, Gravite adds up the points obtained across all criteria to produce an overall score out of 100, directly readable and comparable over time.
