Desired Outcome
Composite nowcast running on production data. Status classification (on_track / warning / critical) matches reality within one business day. Drift detected before a human notices.
Contributing Factors
| Factor | Required For | Owner |
|---|
| CRM pipeline data populated | Pipeline signal | Engineering (done) |
| Activity logging active | Activity velocity signal | Engineering (done) |
| Agent comms flowing | Agent velocity signal | Engineering (done) |
| Commissioning L-levels tracked | Commissioning signal | Dream team (done) |
| Forecast baselines set | Variance computation | Dream team (gap) |
| Prediction schemas populated | Calibration signal | Both (gap) |
Obstacles
| Obstacle | Type | Impact |
|---|
| No forecast baselines exist | Data gap | Cannot compute variance without a target |
| Sparse signals early (cold start) | Algorithm | Confidence score unreliable below 3 signals |
| Signal scales differ (dollars vs counts vs levels) | Normalization | Apples-to-oranges without 0-1 normalization |
Investigations
| # | Question | Method | Owner |
|---|
| 1 | What pipeline coverage ratio signals healthy? | Industry benchmarks + internal targets | Dream team |
| 2 | What activity velocity per deal is sustainable? | Historical data from CRM seed | Engineering |
| 3 | How fast do signals decay in relevance? | Test 7, 14, 28-day half-lives | Engineering |
Success Measures
| Measure | Binary Test |
|---|
| Composite score produced | calculateNowcast() returns valid NowcastResult |
| Classification accurate | Status matches manual assessment for 5 consecutive days |
| Drift detected early | Warning fires before weekly review catches the same issue |
| Confidence meaningful | Score below 0.5 when fewer than 3 signals present |
Questions
Which outcome in the pipeline nowcast map is the leading indicator — the one that reliably predicts all others?
- Which outcome is hardest to measure but most important to track — and what proxy metric comes closest?
- If the primary outcome were achieved but secondary outcomes lagged, what would that signal about the system?
- At what threshold does each outcome shift from "in progress" to "proven" — and who decides that threshold?