Skip to content

ON Feedback - 2025-10-07

TL;DR

Strong validation of need for manual inputs to baseline forecast. Rodrigo wants ability to input known recurring payments (taxes, bonuses, social security) that ML can't predict from history alone. Also wants improved pattern detection (biweekly vs weekly payroll) and account grouping for holistic views.

Overall Sentiment
Direction validated Yes
Stage shown Internal dashboard (not product UI)
Ready for next stage Yes - build manual input capability
Blocker for them? No, but would significantly improve trust

Source

  • Transcript: 2025-10-07-forecast-performance-deep-dive.md
  • Date: 2025-10-07
  • Participants: Rodrigo Cabrera (Treasury), Emma, Christian, Art
  • Context: Forecast performance deep dive - reviewing ML accuracy via internal dashboard

This Iteration: Validation

What they validated about the forecast settings concept:

  • Manual input for known payments - "If we see something that is required like that every couple of months, we can input it in the system in advance"
  • ML + manual hybrid approach - "With that, we did with the forecast that we have... the machine learning to be better" (wants both)
  • Pattern detection matters - Biweekly payroll being predicted as weekly is "really wrong"
  • Account/entity grouping - Need to see forecasts at entity/currency level, not just individual accounts

Important Nuances

  • Not asking to replace ML, but to supplement it with known items ML can't predict yet
  • First-time events (yearly bonus, new tax) won't be picked up by ML until year 2, 3, 4
  • Wants the system to learn and adjust over time ("if it's 10.2 the next month, they get adjusted")

Future Iterations: Suggestions

High Value

  • Input known recurring payments - "Expecting 10 million every three months to taxes or salaries" so model can consider it
  • Frequency pattern settings - Turn off ML for specific account/category pairs and set explicit frequency (e.g., biweekly)
  • Account grouping / entity views - See forecast at entity or currency level, not just individual accounts

Medium Value

  • Cash pool roll-up - Participant account forecasts should affect master account balance view
  • Balance alerts - "Hey, you will run out of DKK in two weeks"
  • Outlier detection - Surface large unexpected forecasted payments for investigation

Longer Term

  • Transfer history between accounts - When switching banks, use old account's history for new account
  • Multiple data source blending - HR tracker + Tax tracker + ML for comprehensive forecast

Questions Asked

  • Would improved biweekly pattern help? → "Yeah, 100 percent"
  • Would you want ML to predict this or input from elsewhere? → Wants ML that can learn, but with manual override capability
  • How often do account structures change? → "Hopefully not... I don't expect any change"

Raw Feedback Quotes

"If we see something that is required like that every couple of months, we can input it in the system in advance. Like, Hey, we're expecting I don't know, 10 million every three months to taxes or salaries or whatever so the the model can consider it for the future forecast." - Rodrigo

"Of course, we can give an exact amount, right? We will not have the... all the way to the same, but then the tool will already know, the model will already know. Hey, more or less we're expecting 10 million out. I'll put 10 million this month if it's 10.2 the next month, they get adjusted." - Rodrigo

"If I know well there's everyone's getting a 3% raise at the end of July. Like how do I go in and bump it up by? Hey take June times 103%" - Rodrigo (on salary adjustments)

"I don't want to see it on a daily basis. How it's changing minor changes. But if there's any big outgoing forecasts, or incoming forecast to easily identify it." - Rodrigo

"I can see here, okay, I get worried that I have a 50 million payment in two weeks. But if I have a 60 million incoming payment in next week, then there's not much that I can do." - Rodrigo (on why entity-level view matters)

Action Items

  • [ ] Improve biweekly payroll pattern detection
  • [ ] Design manual input flow for known recurring payments
  • [ ] Consider account grouping / entity-level forecast views