Real, timestamped work-activity events -- built on a deliberate principle of tracking work, not people, and explicitly avoiding invented "wellbeing" or productivity scores that no dataset can honestly support.
Completed, in-progress, and overdue task counts per employee are computed from real Task rows, alongside hours actually logged in the last 7 days -- no invented "deep work hours" or psychological score, since no dataset here could honestly support one.
A transparent rule -- 5+ in-progress tasks is HIGH, 3+ is MEDIUM, otherwise LOW -- decides workload level, not an opaque AI judgment nobody can inspect.
A workload-imbalance suggestion only appears when the busiest and idlest employees are both real and the gap is genuinely wide (5+ vs. 1 or fewer in-progress tasks); an empty suggestions list is an honest "nothing to flag," not a placeholder waiting for content.
Enterprise deployment, workflow design, and data migration — talk to the team that builds it.