Configure forecasting models based on data quality and account characteristics¶
Investment Thesis¶
Customers with 100+ accounts need granular control over which models run where. The current one-size-fits-all approach doesn't scale — ML works brilliantly for some categories (regular vendor payments) but poorly for others (intercompany, tax). Giving users the right configuration controls builds trust in the overall system and reduces noise from bad predictions.
Related Domains¶
- Cash Forecasting — forecast accuracy, model configuration, variance analysis
Desired Outcomes¶
Current Focus¶
These are the outcomes we're actively investing in.
| # | Desired Outcome | Evidence |
|---|---|---|
| 1 | Minimize forecast errors caused by poor data sources (e.g., unreliable ARP data) | ON |
| 2 | Reduce time spent correcting forecasts that could have been configured better | ON |
Already Addressed¶
Existing features that help get this job done.
| Desired Outcome | Status | Feature |
|---|---|---|
| Increase ability to disable/enable ML per account based on suitability | ✅ Solved | Forecast Settings |
| Minimize the effort required to demonstrate forecast accuracy | ⚠️ Partial | Forecast Settings |
| Increase the transparency of model selection and performance | ⚠️ Partial | Forecast Settings |
Not Yet Addressed¶
Known outcomes we're not focusing on yet.
| Desired Outcome | Evidence |
|---|---|
| Provide data quality indicators per account/category to guide configuration decisions | ON |
| Recommend optimal settings based on historical accuracy | ON |
| Support combination forecasts (ML + ERP data for same category) | ON |
Current Approach¶
Stage: Skateboard (Validate)
Expanding forecast configuration from simple ML on/off toggle to richer per-category/per-account settings. Users should be able to set forecast method preferences (ML, manual, fixed recurring, combination) at the category level, with account-level overrides where needed.
What's Validated¶
- Per-category control for forecast method (ML vs manual) — "For this account, please deactivate machine learning" (ON, 2025-11-11)
- Fixed recurring amount option — Works for tax, salaries, fees (ON, 2025-06-26)
- Simplicity advantage vs enterprise tools — Praised vs Kyriba/SAP (ON, 2025-06-26)
- Global settings approach — Validated as starting point (ON, 2025-06-26)
Next Milestone¶
Validate expanded settings UI with ON — can they configure their 150+ accounts efficiently?
Feedback Log¶
| Date | Company | Validated | Summary | File |
|---|---|---|---|---|
| 2025-06-26 | ON | Yes | Early validation, simplicity praised | View |
| 2025-07-01 | Instacart (Expert) | Yes | Combination forecasts validated | View |
| 2025-10-07 | ON | Partially | ML pattern detection needs improvement | View |