Skip to content

Apply percentage-based assumptions to forecasts while preserving underlying patterns

Investment Thesis

Every customer and prospect does scenario modelling in Excel because no TMS handles it well. Simple percentage-based adjustments that preserve underlying weekly/daily patterns are more valuable than complex scenario trees. First to solve this elegantly wins.

Notes on Scope & Direction

Five Building Blocks of Assumptions — 2026-03-09

Emma Sjöström

Across all feedback sessions, five distinct assumption types keep surfacing. These are the building blocks — the LEGO bricks — that compose into any scenario a treasury team needs. Each assumption a user creates combines some or all of these primitives.

  1. Percentage — Scale a category's forecasted values up or down by X%. Example: "-5% on collections" or "+3% payroll raise." The foundation — validated by every customer. Preserves the underlying weekly/daily patterns in the ML forecast while shifting the level. A multiplicative adjustment applied to each forecasted data point in the selected scope.

  2. Amount — Add or remove a fixed absolute value on a specific date. Example: "50M acquisition outflow on June 15" or "2M bonus payout in March." Covers one-off events that don't exist in historical data — acquisitions, loan drawdowns, bonus payouts, capital expenditure. An additive adjustment applied to a single point in time. (Personio: "50 million out of the door in six months time"; Instacart: bonuses as explicit one-time events distinct from trends)

  3. Timing — Shift inflows or outflows forward or backward by N days. Example: "AR from Customer X is paying 15 days late" or "IC payments to the German entity delayed by 30 days." Changes when cash moves, not how much. Critical for modeling payment term renegotiations, collection slowdowns, and supplier delays. Redistributes forecasted values across periods for a specific counterparty or sub-segment within a category — the total remains the same but the cash lands in different weeks. (ON: "timing shift is actually quite good and quite useful in lots of different cases")

  4. Date range — Define when an assumption is active and whether it represents a permanent change or a temporary effect. A new normal is open-ended ("from July 1 onwards" — e.g., annual payroll raise). A temporary disruption has a start and end ("Q2 only while we renegotiate supplier terms"). A one-off is a single date ("March 15 bonus payout"). A time filter applied to any of the other building blocks, scoping their effect to specific periods. The permanent vs temporary distinction matters for how the assumption interacts with rolling forecast updates. (Instacart: effective date picker validated; ON: adjustable window within the 13-week horizon)

  5. Volume — What portion of a category's total volume the assumption should apply to. Example: "apply to 60% of AP" or "only IC payments to EMEA entities." Not every assumption hits 100% of a category — a collection slowdown might only affect one customer segment, a payment delay might only hit certain counterparties. A scaling factor (0-100%) or filter that narrows the scope within a category before the percentage/amount/timing adjustment is applied. (ON: entity-level bottom-up approach; Personio: multi-entity selection for subsets like tax groupings)

The current skateboard only supports percentage. How might we explore the other four?

Desired Outcomes

Current Focus

These are the outcomes we're actively investing in.

# Desired Outcome Evidence
1 Minimize the risk of overestimating collections or underestimating outflows Personio
2 Increase the ability to model scenarios quickly (e.g., -5% on collections) Personio
3 Reduce manual effort in creating multiple forecast scenarios Personio, Levi's, ON
4 Minimize the loss of weekly/daily patterns when adjusting forecast totals Personio

Already Addressed

No existing features directly address this job yet.

Not Yet Addressed

Known outcomes we're not focusing on yet.

Desired Outcome Evidence Planned Job
Minimize the effort to model multi-faceted business events as coherent scenarios ON, Personio Composable assumptions
Increase reusability of individual assumptions across different scenario combinations ON, Personio Composable assumptions
Reduce the complexity of understanding combined impacts from multiple simultaneous changes ON, Personio Composable assumptions
Minimize the manual effort to model 'what if payments are delayed by N days' scenarios ON, Instacart Timing shifts
Increase the ability to quantify the cash impact of intercompany payment term changes ON, Instacart Timing shifts
Reduce the time to model the impact of collection slowdowns on entity-level cash ON, Instacart Timing shifts
Minimize the effort to calculate worst-case cash position from historical min/max by category group Personio, ON Worst/best case
Increase the ability to layer multiple severity levels by toggling category groups on/off Personio, ON Worst/best case
Reduce dependency on manual spreadsheet analysis for stress testing Personio, ON Worst/best case
Minimize difficulty of remembering past forecast assumptions Instacart
Increase ability to attribute variance to assumption changes vs actuals variance Instacart
Reduce unexplained variance by tracking all assumption changes systematically Instacart
Minimize manual translation between accrual-based budget and cash forecast Instacart
Minimize manual input by using reusable percentage-based adjustments for market disruptors Levi's
Increase transparency of adjustment reasons for future analysis Levi's
Minimize the ad-hoc nature of investment decisions throughout the month Personio

Current Approach

Stage: Skateboard (Validate)

Building an "Assumptions Studio" — a simple interface for applying percentage-based adjustments to ML forecasts at category level. Key design principle: preserve underlying weekly/daily patterns when adjusting totals. Users should be able to save named scenarios and compare them side-by-side.

What's Validated

  • Apply assumptions to forecast — "Being able to save a version of the forecast" (David Watt/Instacart, 2025-07-01)
  • Effective date picker — Validated concept (Instacart, 2025-07-01)
  • Visual preview of impact — "I'd always want to see the last couple months of actuals" (Instacart, 2025-07-01)
  • Percentage-based adjustments — "Apply -5% on collections as a conservative assumption" (Personio, 2025-10-21)
  • Prototype UI consistency — "UI looks good... in line with the rest of the system" (Tom/Personio, 2026-02-18)
  • Category + entity + currency as filter dimensions — "the parameters work here... that kind of covers it" (Tom/Personio, 2026-02-18)
  • Percentage approach sufficient for treasury — "replicates 90% of the scenario analysis I've done in the past" (Tom/Personio, 2026-02-18)
  • LEGO-style composable assumptions — "it creates trust... when you change one brick of the Lego, it's easier to judge if the change makes sense" (Lucia/ON, 2026-03-04)
  • Bottom-up entity-level approach — "I prefer bottom up approach because it is easier to validate if it is working or not" (Yulia/ON, 2026-03-04)
  • Time-bounded assumptions (adjustable window) — "you can play with the 13 weeks, shorten it to two, three weeks" (ON, 2026-03-04)

Next Milestone

Validate skateboard prototype with ON and Personio — can they replace their Excel scenario workflows?

Feedback Log

Date Company Validated Summary File
2025-07-01 Instacart (Expert) Yes Strong validation of concept, build MVP with version tracking View
2026-02-18 Personio Yes Prototype UI validated, covers "90% of scenario analysis". Key asks: one-off absolute items, layered assumptions, save/share workflow, forecast explainability View
2026-02-18 ON Yes 8-9/10 excitement, part of 2026 goals. Key asks: working capital scenarios, IC categories, fixed scenario tracking vs rolling actuals, new entity scenarios View
2026-03-04 ON Partially Prototype tested live. Standout idea: dynamic minimum cash buffer line derived from forecasted outflows (not hardcoded) for CFO guidance and regional negotiations. Also need: timing shifts, category-level impact breakdown, goal-seeking View