Skip to main content

AI tiles

The wallboard uses AI in two places: building a tile from a sentence, and the shift summary tile. Both need AI to be enabled for your environment (see Data sources and setup). Without one, Ask AI reports that AI is not configured and summary tiles show an error; everything else works.

Build a tile from a sentence​

In the editor, choose Ask AI and describe the tile, for example:

  • "abandon rate per queue today as bars"
  • "service level for the sales queue against an 85 % target"
  • "agents by state"

The AI fills in the same tile description the tile form edits: metric, window, breakdown, filters, chart and thresholds. It never produces code or queries.

  • It is given the real queue names (and, with a dialer, the campaign names), so "the sales queue" becomes a filter on that queue.
  • Its answer is checked against the metric catalog and the tile rules. If the check fails, the problems go back to the model for one repair round. A tile that still fails is not added.
  • The new tile is placed in the first free spot on the board, shrinking to fit if needed. When the board is full, it says so.
  • You see a preview with sample data, can Adjust the request, and can still edit the result in the tile form before saving.

The AI shift summary​

The ai_summary tile (in the Day summary template) shows a few sentences about how today is going.

  • What the model sees: only today's KPIs for the board's scope: offered, handled, service level, abandon rate, ASA, AHT, the busiest hour and the worst queue. It can quote these numbers but has nothing else to draw on.
  • How often: one call per board and language about every 10 minutes, shared by every screen showing the board. A failed call is retried after 2 minutes.
  • Quiet days: with no contacts yet today, the tile shows a fixed sentence and the model is not called.
  • Language: the summary is written in the board's TV language.