Cash Forecasting - Jobs & Desired Outcomes¶
Generated from .jobs.json — do not edit directly.
Confirmed Outcomes¶
Corroborated by 2+ sources
Job: Explain and demonstrate forecast reliability to stakeholders¶
| Desired Outcome | Status | Solved By |
|---|---|---|
| Minimize the effort required to demonstrate forecast accuracy | ⚠️ Partial | Reporting, Variance Analysis |
| Increase the transparency of model selection and performance | ⚠️ Partial | Forecasting |
| Reduce skepticism from finance stakeholders about ML-based forecasts | ❌ Not solved | — |
| Increase ability to answer "can I trust it?" with data | ⚠️ Partial | Variance Analysis |
| Increase ability to investigate root causes of forecast misses | ⚠️ Partial | Variance Analysis |
| Minimize the manual effort to compare forecast versions week over week | ❌ Not solved | — |
| Increase visibility into whether categorization changes improved accuracy | ❌ Not solved | — |
Gaps: No model performance dashboard, no accuracy trends over time, no plain-English explanations of model predictions.
Sources: - ON (2025-11-11) - "Can I trust it? Machine learning can be such a big thing. So I think this will be really cool to have that tap." - Lucia - ON (2025-11-17) - "Variance is not there. No comparison. This is what we are not doing either." - Yulia; confirming they can't currently explain or validate forecast accuracy - ON (2024-11-19) - "You have to be able to feel like you're contributing and understanding what the model is doing. Otherwise I wouldn't present this to my management." - Lucia - ON (2026-01-22) - "Week by week, you see the average map for each forecast. And then gives you if it reduced... it means that it's improving." - Federico; building validation methodology - ON (2026-03-10) - "Palm is definitively the superior model since it provides a stable, single-digit risk environment on the short term." - Giannis/Federico; ON's FC Validation deck: 7-week head-to-head, Palm ~8% WMAP vs Kyriba ~16-20%. WMAPE heatmap methodology. Presented to leadership. Palm chosen as 'definitive choice to drive the new cash planning ecosystem.'
Job: Configure forecasting models based on data quality and account characteristics¶
| Desired Outcome | Status | Solved By |
|---|---|---|
| Minimize forecast errors caused by poor data sources (e.g., unreliable ARP data) | ⚠️ Partial | Forecast Settings |
| Increase ability to disable/enable ML per account based on suitability | ✅ Solved | Forecast Settings |
| Reduce time spent correcting forecasts that could have been configured better | ⚠️ Partial | Forecast Settings |
| Increase forecast accuracy by combining ML with reliable ERP data (e.g., AP/AR) where available | ❌ Not solved | — |
Gaps: No guidance on which accounts/categories are good candidates for ML vs manual. No data quality indicators. No ability to combine ML predictions with ERP AP/AR data.
Sources: - ON (2025-11-11) - "For this account, please deactivate machine learning. Or for this account, please make sure that machine learning definitely includes ARP, whereas for this one our ARP is shit." - ON (2025-10-07) - ML picking up wrong pattern (weekly vs biweekly payroll); wants improved pattern detection - ON (2025-06-26) - Forecast source preferences: Open AR/AP should use combination (ML + ERP data)
Job: Produce a short-term cash forecast I can trust and act on¶
| Desired Outcome | Status | Solved By |
|---|---|---|
| Minimize the time spent updating daily forecast (target: 15 minutes or less) | ✅ Solved | Forecasting |
| Minimize the time spent gathering data from multiple ERP tables | ⚠️ Partial | Data Layer |
| Increase the percentage of forecast based on booked/actual data vs guesses, especially for certain categories (AP/AR) and short time horizons | ⚠️ Partial | Forecasting |
| Increase confidence in inputs by understanding their source and historical accuracy | ⚠️ Partial | Forecast Settings, One-Off Items |
| Reduce the frequency of surprise payments not captured in forecast | ⚠️ Partial | One-Off Items |
| Increase transparency on data sources (% from ERP vs manual vs ML) | ⚠️ Partial | Forecast Settings |
| Increase visibility into large upcoming payments that may need confirmation | ⚠️ Partial | One-Off Items |
| Minimize time spent gathering and collating forecast data | ⚠️ Partial | Forecasting |
| Minimize the variance between forecasted and actual balances at quarter end | ⚠️ Partial | Variance Analysis |
Gaps: AP/AR integration not yet available. No "push button" weekly forecast generation.
Sources: - ON (2024-11-19) - "The difficult part is getting the right data from the ERP... the accounting system is not designed as a treasury management system" - Lucia - ON (2024-11-19) - T+7 benchmark: "You should be able to have up to 90% reliability" - Sonder (2024-11-19) - "We try to do it in 15 minutes or less, right? You don't want to spend a lot of time on data entry." - David Watt - Levi's (2025-12-11) - "My treasurer just wants a button... push a button. It spits out everything for the week." - Dette - ON (2025-08-07) - 95% accuracy target at balance level; "The more is okay. Less is bad." - ON (2026-03-10) - "On the shorter term (2w) Palm establishes a highly reliable baseline at just 4.5%, whereas Kyriba record a 12.5% error rate." - Giannis/Federico; FC Validation deck: Palm outperforms short-term at 4.5% WMAP vs Kyriba 12.5% at 2 weeks ahead. Volatility control — Palm stays single-digit through Week 4.
Job: Optimize use of operational cash¶
| Desired Outcome | Status | Solved By |
|---|---|---|
| Minimize buffer balances held on operational accounts | ⚠️ Partial | Forecasting |
| Minimize idle cash sitting in low-yield operating accounts | ⚠️ Partial | Cash Visibility |
| Increase investment income by deploying excess cash to higher-yield investments | ⚠️ Partial | Forecasting, Investments Visibility |
| Reduce buffer requirements through forecast accuracy | ⚠️ Partial | Forecasting |
| Reduce uncertainty about daily funding needs | ⚠️ Partial | Forecasting |
| Avoid breaking investments early due to unexpected cash shortfalls from forecast misses | ❌ Not solved | — |
Gaps: Investment maturity planning not integrated with forecast. No "available for investment" calculation.
Sources: - Personio (2024-10-03) - Holding buffers because forecast isn't reliable enough; want daily transactional forecast for just-in-time funding - Personio (2024-11-26) - "Cash forecasting from a Treasury perspective is purely around efficiency... having full visibility of flows in and out at a transactional level allows you to be fully efficient with your cash." - Sonder (2024-11-19) - "If you have too much you're losing out on, you know, interest income." - David Watt; run on minimal cash, deploy excess to investments
Job: Forecast cash flows entity-by-entity across decentralized group structure¶
| Desired Outcome | Status | Solved By |
|---|---|---|
| Minimize the time spent chasing entities for forecast submissions | ❌ Not solved | — |
| Minimize the effort required to produce entity-level forecasts | ✅ Solved | Forecasting |
| Reduce errors from manual Excel manipulation (format, decimal conventions, typos) | ✅ Solved | Batch Uploads |
| Increase the granularity of cash visibility at entity level (not just consolidated) | ✅ Solved | Cash Visibility |
| Increase visibility into forecast changes vs original figures | ⚠️ Partial | Variance Analysis |
| Get fresh forecasts without waiting for quarterly/monthly budget cycles from regional teams | ✅ Solved | Forecasting |
Gaps: No workflow for collecting forecasts from regional teams. No forecast-vs-forecast comparison yet.
Sources: - Euroports (2025-10-27) - Multiple entities across Belgium (5), Spain (2), France - want country-level consolidation - Volvo Cars (2025-03-27) - "When it's on the Excel, anything can go wrong with it" / 40 entities report monthly Excel forecasts
Job: Improve forecast accuracy by handling outliers¶
| Desired Outcome | Status | Solved By |
|---|---|---|
| Minimize the impact of one-off large payments on future forecasts | ✅ Solved | One-Off Items, Forecast Settings |
| Reduce manual effort to identify and flag unusual transactions | ⚠️ Partial | Forecasting |
| Increase confidence in forecasts by excluding known anomalies | ✅ Solved | Forecast Settings, Forecasting |
Gaps: No proactive anomaly alerts to users. Detection is automatic but user review of flagged outliers is manual.
Sources: - ON (2025-11-17) - "If there is a one off payment which is sometimes couple of millions... to untick it from the learning... because we checked them, they are really one off and they will just skew the whole future forecast." - Yulia - Sonder (2024-11-19) - "Algorithms have a hard time with sudden changes like that, right? So just being able to ignore this outlier, mark it as an outlier." - David Watt - Personio (2026-02-18) - Tom noticed outlier flags in system. Wants to proactively mark round-value treasury transfers as outliers to exclude from forecast learning. - Tom
Job: Upload forecast data from files without extensive reformatting¶
| Desired Outcome | Status | Solved By |
|---|---|---|
| Minimize the manual changes required to file formats before upload | ✅ Solved | Batch Uploads |
| Reduce the time spent aligning number formats and column structures | ✅ Solved | Batch Uploads |
| Increase the success rate of first-time file uploads | ✅ Solved | Batch Uploads |
Gaps: None — batch uploads feature is shipped.
Sources: - ON (2025-10-02) - Testing batch uploads with tax/talent payment files; issues with European number formatting, column limits in mapping UI - Volvo Cars (2025-03-27) - "When it's on the Excel, anything can go wrong with it" / 40 entities submit monthly Excel forecasts; format alignment is a persistent pain point
Job: Detect missing or anomalous items in entity forecasts proactively¶
| Desired Outcome | Status | Solved By |
|---|---|---|
| Minimize the risk of missing cyclical payments (payroll, tax) in entity forecasts | ❌ Not solved | — |
| Reduce late-day cash surprises caused by unforeseen large transactions | ❌ Not solved | — |
| Increase proactive detection of anomalous forecast inputs before they impact positioning | ❌ Not solved | — |
Gaps: No anomaly detection or proactive alerting for missing forecast items.
Sources: - Volvo Cars (2025-03-27) - "We have very high volumes... and sometimes things slip through" - Lee McEneff; need proactive detection of missing items in entity forecasts - IHG (2025-04-07) - "Sometimes the guys in India will just miss, they'll just forget to add something... even a payroll run has been missed off" - Matthew Hook; India service center manually submits forecasts; payroll runs and other cyclical items get forgotten - ON (2026-02-04) - "$8M IC payment from Canada failed and no one noticed" - Amanda; large IC payment failure went undetected. Need proactive alerting when expected items don't appear.
Job: Track influential customers for AR forecasting accuracy¶
| Desired Outcome | Status | Solved By |
|---|---|---|
| Minimize effort to identify high-impact customers (top 10) | ❌ Not solved | — |
| Improve forecast accuracy by applying customer-specific payment patterns | ❌ Not solved | — |
| Reduce noise from low-impact customer variability | ❌ Not solved | — |
Gaps: No customer-level AR tracking or segmentation in forecast.
Sources: - Levi's (2025-12-11) - "If you have an influential group, i.e. 10, then you go down to that level. Not only are you able to adjust these different factors, but then you can actually apply machine learning for the forecast. This is how they pay." - Dette - Instacart (2025-07-01) - "One client pays $20 million a month... it comes in spikes. That needs to be handled separately with manual input from the relationship team." - David Watt; large client exception — $20M/month spikes vs $200k/day baseline, needing separate category with ML off
Job: Understand and challenge forecast model assumptions at category level¶
| Desired Outcome | Status | Solved By |
|---|---|---|
| Minimize the time required to understand what assumptions the forecast model has applied to a category | ❌ Not solved | — |
| Increase the ability to override model assumptions when external context contradicts the trend | ❌ Not solved | — |
| Reduce the risk of double-counting a trend already captured by the model when adding manual adjustments | ❌ Not solved | — |
Gaps: No visibility into model assumptions. Users can't see what trends are applied to know whether to override.
Sources: - Personio (2026-02-18) - "Tom distinguishes between 'nudging the forecast' (correcting model assumptions) and 'scenario analysis' (layering hypothetical events on top) — two related but distinct workflows." - Tom; wants visibility into WHY a forecast value is what it is — what trends are being applied, so he can confirm or override. Currently invisible. - Instacart (2025-07-01) - "I want to see what rules generated each forecast number... variance should be split into actuals variance vs assumption change variance" - David Watt; wants to see attribution: what rules/assumptions the system applied, and whether variance came from actuals differing or from assumptions being changed
Emerging Signals¶
Single source - needs corroboration
Potential Job: Apply judgment to stakeholder inputs before including in forecast¶
Potential Outcomes: - Minimize the frequency of blindly accepting optimistic projections - Increase ability to sanity-check inputs against historical patterns - Reduce forecast error caused by accepting unrealistic inputs
Source: Sonder (2024-11-19) - "If you don't have an opinion about is that the right thing? Or how likely is that really to happen? It can go pretty wrong if you don't apply your own judgment. Just take a bunch of other people's guesses? Then there's no kind of logic to it." - David Watt
Potential Job: Incorporate external signals into revenue forecasting for unpredictable industries¶
Potential Outcomes: - Increase the ability to use non-financial signals (social engagement, surveys) to inform forecasts - Reduce reliance on historical data when it's not predictive - Minimize the forecast error from assuming past patterns will repeat
Source: Live Events (2025-04-01) - Using Instagram engagement, post-event surveys ("who would you want to see next year"), and weather forecasts to inform ticket sales and on-site revenue predictions
Potential Job: Understand AR/AP movements driving working capital changes¶
Potential Outcomes: - Minimize the time required to identify which customers or payment terms are driving AR changes - Increase the ability to drill down into working capital by customer segment (e.g., freight forwarding vs terminal)
Source: Euroports (2025-10-27) - Need better drill-down into working capital movements; 60-70% of receivables go through factoring, losing customer-level data
Potential Job: Distribute monthly collections forecast across working days based on historical patterns¶
Potential Outcomes: - Minimize the inaccuracy of assuming equal daily collections within a month - Increase the ability to see intra-month cash flow patterns (e.g., higher collections in first two weeks) - Reduce the gap between the FP&A monthly number and treasury's daily visibility needs
Source: Personio (2025-10-21) - Collections forecast is just one monthly number, need daily distribution; January has significant uptick due to yearly customers
Potential Job: Ensure consistent forecasting approaches across regional teams¶
Potential Outcomes: - Increase the consistency of forecasting approaches across regional teams - Minimize the effort to preserve forecast history for accuracy tracking
Source: Sonder (2024-10-03) - Need to track forecast versions separately from actuals; different approaches across teams
Potential Job: Incorporate confirmed AP/AR items into forecast as known future cash flows¶
Potential Outcomes: - Minimize manual effort uploading AP/AR files for forecast inputs - Increase forecast accuracy by incorporating confirmed receivables and payables with known dates - Reduce dependency on batch file exports from ERP for forecast data
Source: ON (2025-10-02) - "The data goes to BigQuery... Palm could be here instead of Kyriba." - Amanda; AP/AR data in BigQuery/ERP contains confirmed future cash flows (open invoices with known dates) that should feed the forecast directly, improving accuracy beyond ML pattern-based predictions.
Notes¶
- "Investigate variances between forecast and actuals to create accountability" is tracked in variance-analysis as the primary domain for that job.
- "Plan liquidity by understanding when deposit maturities return cash" moved to investments-debt — overlaps with "Make informed investment decisions with confidence about cash availability."
- "Apply conservative assumptions to forecasts" is tracked in scenario-modelling as the confirmed job "Apply percentage-based assumptions to forecasts while preserving underlying patterns."
- The following emerging signals were folded into parent jobs: "Generate weekly forecast with push button simplicity" → outcomes folded into "Produce a short-term cash forecast" (Source: Levi's 2025-12-11); "Achieve reliable balance forecasts for liquidity allocation" → outcomes folded into "Produce a short-term cash forecast" (Source: ON 2025-08-07); "Reconcile forecast variances to improve model accuracy" → outcomes folded into variance-analysis (Source: ON 2025-08-07); "Track forecast accuracy improvement over time" → outcomes folded into "Explain and demonstrate forecast reliability" (Source: ON 2026-01-22)
- "Maintain reliable forecast data despite upstream system limitations" moved to data domain.
- cf-e-002 promoted to confirmed cf-008 (2026-03-10). IHG and ON evidence corroborates Volvo Cars — three independent companies describe proactive detection of missing forecast items.
- cf-e-008 promoted to confirmed cf-009 (2026-03-10). Instacart describes same influential-customer AR pattern as Levi's.
- cf-e-009 promoted to confirmed cf-010 (2026-03-10). Instacart wants same model assumption visibility as Personio.
- cf-e-006 split (2026-03-10). 'Reduce misalignment between Treasury, FP&A, and Accounting categorizations' moved to bridging domain as br-e-002. Cash-forecasting retains regional consistency half.
- con-e-001 moved from connectivity to cash-forecasting as cf-e-010 (2026-03-10). Reframed from 'Integrate AP/AR data' (implementation) to 'Incorporate confirmed AP/AR items into forecast' (job).