Performance Intelligence runs almost everything you see through a background job queue. When new data lands — a review is completed, an objective is created, a feedback entry is submitted — Compass adds a job to the queue and processes it asynchronously. This article explains what's running, how to read the queue alerts, and what to do if something looks stuck.
Why background processing
Most AI analysis is too slow to do live. A review extraction takes seconds of model time; a trajectory synthesis can take longer. Doing this in the background means:
Users aren't blocked when they submit reviews or feedback
Admins can open PI any time and see what's analysed so far
The system can fairly schedule across all clients
What you give up: a small delay between data landing and analysis appearing. Most jobs run within minutes of being queued — though review analysis is queued by an hourly sweep, so allow up to an hour or so for a newly completed review to be picked up.
The queue status banner
When you open a module's Introduction page while background work is active, you'll see a banner near the top of the page like:
3 extraction jobs processing
12 quality analysis jobs queued
The banner shows:
The count of jobs
The job type for that module (extraction on Reviews, trajectory analysis on Employee Trajectory, quality analysis on Objectives and Fast Feedback)
Whether they're queued (waiting) or processing (running now)
A small auto-refreshes indicator
The banner refreshes automatically every 30 seconds and disappears when the queue is empty, updating the underlying cards.
The PI landing page (Admin > Performance Intelligence) shows a combined counter beneath the workspace buttons — "5 AI analysis jobs running. Some insights may update shortly." — so you can see at a glance whether Compass is busy without opening the workspace.
Job types
There are eight job types you might see in the queue:
Job type | What it does | Where it surfaces |
Review extraction | Reads each completed review and produces extraction signals (performance level, sentiment, themes, etc.) | Reviews tab |
Trajectory analysis | Synthesises cross-cycle trajectory per employee (archetype, readiness, gaps) | Employee Trajectory tab |
Objective quality | Scores each objective on SMART dimensions | Objectives > Quality & SMART |
Objective embeddings | Generates a vector embedding per objective for similarity matching | Behind the scenes — Alignment view |
Objective entity extraction | Pulls clients, matters, themes, keywords from each objective | Behind the scenes |
Fast feedback quality | Scores each feedback entry for writing quality | Fast Feedback > Quality |
Fast feedback NLP | Sentiment classification and topic extraction per feedback entry | Fast Feedback > Sentiment, Research |
Collaboration points NLP | Sentiment of the "reason" text on collaboration point allocations | Reviews > Collab Points |
How jobs get created
Jobs are created automatically, in two ways:
Event-triggered: when a feedback entry is completed, an objective is saved, or collaboration points are allocated, the relevant jobs are queued immediately.
Hourly discovery sweep: every hour, an automated background process sweeps for anything that needs analysis and creates the jobs. This is what queues review extraction and trajectory work — it picks up newly completed reviews, and re-analyses reviews that were edited or had their PDF regenerated since the last extraction. The same sweep catches any event-triggered items that slipped through, and skips job types whose AI feature is disabled.
Admins can also run a discovery sweep on demand from the Background Jobs tab under Admin > Settings > Audit Logs — useful if you don't want to wait for the next hourly pass.
AI insight briefings on Employee Trajectory
The Employee Trajectory views (Overview, Development, Risk, Succession) each carry a Compass AI insights card that does something slightly different from the queue work. Instead of analysing individual records, it sends an aggregate of the trajectory signals to Compass and asks for a short narrative briefing on that view — top patterns, biggest risks, what's working and what isn't.
The briefing is generated the first time you open the view, then cached at the workspace level (not per employee). It isn't refreshed automatically when new trajectory data lands — click the small Regenerate icon on the card to produce a fresh briefing.
What to do if something looks stuck
If you see a job stuck in "processing" for a long time, or a number that doesn't go down:
Wait 30 seconds and refresh. The banner polls automatically but a hard refresh sometimes helps.
Check the source data. A common cause is the underlying review or objective being missing or in an unexpected state.
Check whether Performance Intelligence is enabled. PI is a licensed AI feature; if it's been switched off, no new jobs are created — the landing page will say Performance Intelligence is not enabled. Ask your account manager.
If it persists, raise it with support. Long-stuck jobs usually indicate an upstream issue that needs investigation.
One case that looks stuck but isn't: trajectory analysis deliberately waits until all extraction jobs have finished before it runs, so a trajectory queue that isn't moving while extraction jobs are processing is expected.
Why a view might be empty
If a PI view looks empty or shows a notice card explaining "no analysis yet", the cause is usually one of:
Reviews haven’t been completed: extraction only runs on completed reviews with PDFs.
No primary cycles for trajectory: trajectory needs at least two primary-classified reviews per employee. Cycle classification is itself an extraction signal, so new firms can take a cycle or two before trajectory becomes meaningful.
No analysis run yet: quality and NLP analyses are queued the moment data lands but take a few minutes; the page will populate as jobs complete.
Filters too narrow: a view filtered to a department with no eligible data will look empty even if the rest of the workspace is healthy.
Tips
Don't pre-emptively re-run jobs. Compass is good at picking up new and changed data automatically — running things manually mostly adds noise to the queue.
The queue is a useful health signal. A queue that's consistently growing means something upstream is jamming. A queue that's regularly hitting zero means you're in steady state.
Background jobs are how PI scales. Even when the queue looks busy, you can keep using the workspace — views show what's been processed so far, with a banner explaining what's still in flight.