Where each deal is actually heading.
Your pipeline report repeats what someone typed into a stage field. Deal Compass reads the evidence — the calls, the threads, the calendar — and renders a judgment you can check: every reading carries a receipt, and nothing sends without your approval.
Reads your existing stack through connectors. Nothing is migrated.
Stage fields are claims. Deals are behavior.
A deal marked Commit can be quietly dying: the champion's language shifts from "when" to "if," the economic buyer drops off the thread, the next meeting gets pushed and never rebooked. None of that changes the stage field — so none of it shows up in the forecast until the quarter is already lost.
Deal Compass puts a second needle on every deal. One mark for what the record claims. One for what the behavior shows. The gap between them is the risk no stage report surfaces — and it's the first thing you see.
Evidence in, arithmetic shown, judgment out.
Signals with receipts
Compass reads the sources you already have — email threads, meeting transcripts, calendar, your CRM history — and extracts behavioral signals. A signal without a receipt is discarded, never shown.
Deterministic arithmetic
Each signal is scored, weighted by class, and decayed over time — recent behavior counts more. The result is a drift index from −100 to +100. Same evidence, same answer, every time. No black box in the scoring path.
A heading, and one next move
The index lands in one of five heading bands. If the heading disagrees with the recorded stage, the deal is flagged divergent — and Compass proposes one recommended action, drafted and waiting for your approval.
Five headings, fixed bands, no curve-fitting. A deal reads Dead, not marked only when the evidence says everyone has stopped behaving like it's alive — while the record still says otherwise.
Built to be checked, not believed.
Receipts or nothing
Every signal persists a pointer into the source record — a transcript span, a thread, a calendar event. You can open every line behind every heading.
No fabricated numbers
Precedent statistics compute only from outcomes your firm has actually recorded. Until at least five matched outcomes exist, Compass says the pattern library is still building.
Coverage honesty
If a source isn't connected or has gone stale, that gap is shown on the reading — not hidden. Confidence is capped when a source class is missing, so a thin reading never masquerades as a strong one.
You approve every send
Compass proposes one next move per deal and drafts it. It sits awaiting your approval — approve and it queues through your approved sender, or decline with a reason. Either way the outcome is scored back in.
It remembers what your deals did last time.
Every ratified or declined action gets its outcome scored back in. Over time, Compass can tell you not just that a champion is hedging — but that this account hedged the same way in Q1, and that deal slipped six weeks.
The memory is firm-owned: built from your outcomes, on your data, useful to nobody else. And it only speaks when it has enough to say.
Meridian hedged the same way in Q1 — champion moved to "if" language at Commit. That deal slipped before closing.
$100–250K enterprise deals showing champion-hedge + buyer-absence at Commit.
Point it at one list. See the gap.
Connect the sources you already use, generate a Compass for a single account list, and open the worst deal first. If the reading doesn't hold up to its own receipts, close the tab.
We reply from a human inbox. No sequence, no drip.