Skip to content

Forecast Settings

Status: Shipped

Domain: Cash Forecasting Linear Projects: Turn ML on/off by category+account v0

What It Does

Palm gives users control over how forecasts are generated for different categories and accounts. Rather than a black-box ML system, treasury teams can configure which categories should use machine learning predictions versus other forecast sources, and fine-tune settings to match their business reality.

This control is essential for building trust in forecasts. Users can see exactly how each category is being forecasted and make informed decisions about when ML adds value versus when they want more direct control.

Capabilities

Capability Status Notes
ML on/off by category Shipped Enable/disable ML for specific categories
ML on/off by account Shipped Enable/disable ML for specific accounts
Category-level settings Shipped Configure behavior per payment type
Forecast source visibility Shipped See what's driving each forecast

Jobs Fulfilled

From cash-forecasting/jobs.md

1. Explain and demonstrate forecast reliability to stakeholders

Desired Outcomes Addressed: - [x] Minimize the effort required to demonstrate forecast accuracy - [x] Increase the transparency of model selection and performance - [x] Reduce skepticism from finance stakeholders about ML-based forecasts - [ ] Increase ability to investigate root causes of forecast misses (partial - settings help but not full investigation)

How Palm Addresses This: - Transparent configuration shows exactly what ML is and isn't doing - Users can disable ML where it doesn't make sense - Explainable forecast sources build stakeholder trust

2. Investigate variances between forecast and actuals to create accountability

Desired Outcomes Addressed: - [x] Increase the ability to make informed model configuration decisions - [ ] Minimize the time to identify root causes of forecast variances (not direct) - [ ] Minimize the time required to identify which payment types are causing variances (not direct)

How Palm Addresses This: - Based on variance analysis, users can adjust which categories use ML - Poor-performing categories can be switched to alternative sources - Configuration decisions are informed by actual results

Pain Points Addressed

Pain Point Addressed? Notes
Stakeholder skepticism about ML Yes Transparent control and configuration
No smart forecasting in TMS Yes ML available with user control
Need to understand what model is doing Yes Settings show exactly what's enabled

What's NOT Included (Yet)

  • ML performance metrics by category
  • Automatic recommendations for ML settings
  • A/B testing different configurations
  • Historical settings audit trail

How It Works (Technical)

TODO: Fill in via codebase analysis

Component Technology Notes
Settings storage
Forecast source router
API endpoints

Key files/services: - TBD

Last updated: 2026-02-17