Revenue Forecasting (FORECAST Framework)
Design a revenue-forecasting pipeline that produces a defensible, calibrated number - not a rep-roll-up that's been over-promised twice. FORECAST blends bottoms-up pipeline math with a tops-down model, runs scenarios, and closes the loop with calibration so the forecast improves quarter over quarter.
Core Principle
A forecast is only as good as its calibration loop. Most forecasts re-anchor every quarter and never learn. FORECAST treats forecasting as an ensemble of models with explicit error tracking, so the system gets more accurate over time.
The FORECAST Framework
| Letter |
Stage |
The Question |
| F |
Foundations |
What's the ARR / bookings definition, period boundary, and currency convention? |
| O |
Outlook (Bottoms-Up) |
What does pipeline-weighted by stage and rep commit produce? |
| R |
Run-Rate Model |
What does the time-series / cohort model produce independent of pipeline? |
| E |
Ensemble Blend |
How are bottoms-up and tops-down blended, and what's the confidence band? |
| C |
Calibration |
What's the historical forecast error by segment, stage, and rep? |
| A |
Adjust |
What manual adjustments are in, and which are evidence-based vs hope-based? |
| S |
Scenarios |
What are the base / upside / downside cases and their drivers? |
| T |
Track |
How is forecast vs actual tracked, and how does it feed back into the model? |
Bottoms-Up Forecast
| Element |
Spec |
| Stage Conversion |
Historical conversion % from each stage to closed-won, refreshed quarterly |
| Time-in-Stage Decay |
Probability decay for opportunities aging past expected stage duration |
| Rep Commit Categories |
Commit / Best Case / Pipeline / Omitted with named definitions |
| Coverage Multiples |
3x for new logo, 1.2-1.5x for renewal, segment-specific |
| Hygiene Rules |
Stale opps demoted, no-next-step opps flagged, close-date discipline |
Tops-Down Run-Rate Model
| Method |
Use For |
| Cohort run-rate |
Established motions with stable retention |
| Channel attribution roll-up |
Multi-channel motions; identifies channel-level slow-down |
| Seasonality-adjusted trend |
Markets with clear quarterly / monthly seasonality |
| Leading-indicator regression |
Mature businesses with stable lead → revenue mapping |
Ensemble Blending
Don't pick one model - blend them, weighted by historical accuracy:
| Component |
Weight Rationale |
| Bottoms-up rep commit |
Weight up when historical commit accuracy > 90% |
| Bottoms-up stage-weighted |
Weight up for new motions or new reps |
| Tops-down run-rate |
Weight up for mature, stable segments |
| AI / ML model |
Weight up only if it beats the others on out-of-sample tests |
Always produce point estimate + confidence band - never a single number with no error bar.
Scenarios
| Scenario |
Construction |
| Base |
Ensemble central estimate |
| Upside |
Top quartile of pipeline conversion + favorable mix |
| Downside |
Bottom quartile conversion + concentration-risk realization |
| Stress |
Material churn / lost-deal / macro event sensitivity |
Each scenario must name the 2-3 drivers that move it, not just shift a number.
Calibration Loop
This is where most forecasting programs fail.
| Step |
Action |
| Track forecast vs actual |
By period, segment, stage, rep |
| Decompose error |
Conversion error vs timing error vs mix error |
| Update model weights |
Reweight ensemble based on out-of-sample accuracy |
| Revise stage conversion |
At least quarterly; sooner if material drift |
| Coach rep commit accuracy |
Visible scorecards |
Output
Save to outputs/revenue-forecasting-[period]-[YYYY-MM-DD].md
| Artifact |
Description |
| Definitions Sheet |
ARR / bookings / period / currency conventions |
| Bottoms-Up Spec |
Stage conversion, decay, commit categories, hygiene rules |
| Tops-Down Model |
Run-rate / regression / cohort approach with assumptions |
| Ensemble Spec |
Component weights with historical-accuracy rationale |
| Scenario Pack |
Base / Upside / Downside / Stress with named drivers |
| Calibration Report |
Forecast vs actual error decomposition, trend |
| Adjustments Log |
Every manual adjustment with rationale and owner |
| Forecast Dashboard |
Single source of truth across finance, sales, RevOps |
Process
- Lock the definitions - same ARR, period, and currency rules across teams
- Build the bottoms-up with disciplined hygiene and decay rules
- Build at least one tops-down model as a check
- Blend into an ensemble with weights from historical accuracy
- Run scenarios with named drivers, not just percent shifts
- Close the calibration loop every period - forecast credibility lives or dies here
Tips
- Single-number forecasts hide risk - always publish a confidence band
- Decay stale opportunities ruthlessly - they're the #1 source of forecast misses
- Manual adjustments need evidence - log them or they become hope
- Calibrate per segment / per rep - aggregate accuracy hides poor accuracy underneath
- The forecast is a product - ship versioned releases, not slack messages
Pairs With
- revenue-analytics - Drivers and leading indicators feed the forecast
- renewal-orchestration - Risk score informs renewal-stage probability
- customer-analytics - Cohort retention curves feed run-rate models
- budget-allocator - Forecast scenarios drive reallocation decisions
1---2name: revenue-forecasting3description: Revenue forecasting pipeline - bottoms-up pipeline forecast, tops-down model, ensemble blending, scenario analysis, and forecast calibration loop. Use when: revenue forecast, sales forecast, pipeline forecast, bookings forecast, NRR forecast, ARR forecast, forecast calibration, scenario planning, ensemble forecasting, board forecast.4---56# Revenue Forecasting (FORECAST Framework)78Design a revenue-forecasting pipeline that produces a defensible, calibrated number - not a rep-roll-up that's been over-promised twice. FORECAST blends bottoms-up pipeline math with a tops-down model, runs scenarios, and closes the loop with calibration so the forecast improves quarter over quarter.910## Core Principle1112**A forecast is only as good as its calibration loop.** Most forecasts re-anchor every quarter and never learn. FORECAST treats forecasting as an *ensemble* of models with explicit error tracking, so the system gets more accurate over time.1314## The FORECAST Framework1516| Letter | Stage | The Question |17|--------|-------|--------------|18| **F** | Foundations | What's the ARR / bookings definition, period boundary, and currency convention? |19| **O** | Outlook (Bottoms-Up) | What does pipeline-weighted by stage and rep commit produce? |20| **R** | Run-Rate Model | What does the time-series / cohort model produce independent of pipeline? |21| **E** | Ensemble Blend | How are bottoms-up and tops-down blended, and what's the confidence band? |22| **C** | Calibration | What's the historical forecast error by segment, stage, and rep? |23| **A** | Adjust | What manual adjustments are in, and which are evidence-based vs hope-based? |24| **S** | Scenarios | What are the base / upside / downside cases and their drivers? |25| **T** | Track | How is forecast vs actual tracked, and how does it feed back into the model? |2627## Bottoms-Up Forecast2829| Element | Spec |30|---------|------|31| **Stage Conversion** | Historical conversion % from each stage to closed-won, refreshed quarterly |32| **Time-in-Stage Decay** | Probability decay for opportunities aging past expected stage duration |33| **Rep Commit Categories** | Commit / Best Case / Pipeline / Omitted with named definitions |34| **Coverage Multiples** | 3x for new logo, 1.2-1.5x for renewal, segment-specific |35| **Hygiene Rules** | Stale opps demoted, no-next-step opps flagged, close-date discipline |3637## Tops-Down Run-Rate Model3839| Method | Use For |40|--------|---------|41| **Cohort run-rate** | Established motions with stable retention |42| **Channel attribution roll-up** | Multi-channel motions; identifies channel-level slow-down |43| **Seasonality-adjusted trend** | Markets with clear quarterly / monthly seasonality |44| **Leading-indicator regression** | Mature businesses with stable lead → revenue mapping |4546## Ensemble Blending4748Don't pick one model - blend them, weighted by historical accuracy:4950| Component | Weight Rationale |51|-----------|------------------|52| **Bottoms-up rep commit** | Weight up when historical commit accuracy > 90% |53| **Bottoms-up stage-weighted** | Weight up for new motions or new reps |54| **Tops-down run-rate** | Weight up for mature, stable segments |55| **AI / ML model** | Weight up only if it beats the others on out-of-sample tests |5657Always produce **point estimate + confidence band** - never a single number with no error bar.5859## Scenarios6061| Scenario | Construction |62|----------|--------------|63| **Base** | Ensemble central estimate |64| **Upside** | Top quartile of pipeline conversion + favorable mix |65| **Downside** | Bottom quartile conversion + concentration-risk realization |66| **Stress** | Material churn / lost-deal / macro event sensitivity |6768Each scenario must name the **2-3 drivers** that move it, not just shift a number.6970## Calibration Loop7172This is where most forecasting programs fail.7374| Step | Action |75|------|--------|76| **Track forecast vs actual** | By period, segment, stage, rep |77| **Decompose error** | Conversion error vs timing error vs mix error |78| **Update model weights** | Reweight ensemble based on out-of-sample accuracy |79| **Revise stage conversion** | At least quarterly; sooner if material drift |80| **Coach rep commit accuracy** | Visible scorecards |8182## Output8384Save to `outputs/revenue-forecasting-[period]-[YYYY-MM-DD].md`8586| Artifact | Description |87|----------|-------------|88| **Definitions Sheet** | ARR / bookings / period / currency conventions |89| **Bottoms-Up Spec** | Stage conversion, decay, commit categories, hygiene rules |90| **Tops-Down Model** | Run-rate / regression / cohort approach with assumptions |91| **Ensemble Spec** | Component weights with historical-accuracy rationale |92| **Scenario Pack** | Base / Upside / Downside / Stress with named drivers |93| **Calibration Report** | Forecast vs actual error decomposition, trend |94| **Adjustments Log** | Every manual adjustment with rationale and owner |95| **Forecast Dashboard** | Single source of truth across finance, sales, RevOps |9697## Process98991. **Lock the definitions** - same ARR, period, and currency rules across teams1002. **Build the bottoms-up** with disciplined hygiene and decay rules1013. **Build at least one tops-down** model as a check1024. **Blend into an ensemble** with weights from historical accuracy1035. **Run scenarios** with named drivers, not just percent shifts1046. **Close the calibration loop** every period - forecast credibility lives or dies here105106## Tips1071081. **Single-number forecasts hide risk** - always publish a confidence band1092. **Decay stale opportunities ruthlessly** - they're the #1 source of forecast misses1103. **Manual adjustments need evidence** - log them or they become hope1114. **Calibrate per segment / per rep** - aggregate accuracy hides poor accuracy underneath1125. **The forecast is a product** - ship versioned releases, not slack messages113114## Pairs With115116- **revenue-analytics** - Drivers and leading indicators feed the forecast117- **renewal-orchestration** - Risk score informs renewal-stage probability118- **customer-analytics** - Cohort retention curves feed run-rate models119- **budget-allocator** - Forecast scenarios drive reallocation decisions
varunk130