Opportunities

An opportunity is the unit Compass exists to produce: an experiment worth trying. Sometimes that's an AI build; just as often it's a simpler change — a process fix, a different tool, a step removed. It reads like "auto-classify inbound tickets by queue and urgency," not "use AI in support." The difference is that an opportunity is anchored to real work, so it's something you can actually go test.

Grounded in the map

An opportunity can attach to a Workflow page — down to a particular step — and link to any other pages it touches: the pain point it addresses, the systems it uses, the value or priority it serves, the people it involves. That grounding is the whole point: an idea floating in a backlog doc ages out and loses its context, while an opportunity tied to the map carries everything around it. When you go to run it, the brief is already there.

You don't have to spot every one yourself. On any Workflow page, Suggest opportunities reads the workflow — its body plus the connected people, systems, and pain points — and proposes experiments. Proposals arrive marked proposed and wait in the Inbox until you accept or dismiss them; nothing AI-suggested becomes real without your say.

The pipeline

The Opportunities tab is a board: New, Backlog, Experimenting, Settled, Rejected. Drag a card to move it. A table view sits behind a toggle for sorting and filtering across the three optional 1–5 scores — Value, Feasibility, Risk — plus notes and tags.

The lanes mean what they say:

• New — surfaced, undecided.
• Backlog — yes eventually; not now.
• Experimenting — the experiment is running.
• Settled — it concluded, and the outcome is on the record.
• Rejected — dismissed without running it. Kept, not deleted: where you chose not to experiment is a result too.

Say what you expect — before the evidence

Moving an opportunity into Experimenting asks one question: what do you expect this to change? That sentence is written down before the evidence exists, as an entry in Ledger — the workspace's decision memory. You can skip it, but the skip shows: an experiment that never registered an expectation wears Unregistered when it concludes.

When the experiment ends, you settle the entry: did the expectation hold? The card carries the answer from then on — Confirmed or Missed — and the lesson lives in Ledger. Nothing in the Settled lane can claim success by drag alone; the verdict comes from the settlement.

Over time this makes the board something a backlog never is: a record of what was tried, what was expected, and what actually happened.

From opportunity to built

When the experiment calls for an AI build, Draft in Workbench — right on the opportunity — generates a runnable first-draft flow in Workbench to start from, which you then prove out in Caliper. What that measurement exposes points back at the next opportunity. The map keeps turning.