> ## Documentation Index
> Fetch the complete documentation index at: https://docs.francis.app/llms.txt
> Use this file to discover all available pages before exploring further.

# Forecast OPEX

> How to forecast operating expenses in Francis: driver-based, statistical, and hardcoded approaches across different cost lines.

OPEX covers a wide range of cost lines (software, marketing, facilities, professional services), each with different predictability and drivers. Salaries and payroll are covered on the [Headcount](/masterclasses/budgeting-forecasting/pnl-forecasting/headcount) page; this page covers the rest.

## What approach to use

Statistical is the right default for most OPEX lines. Recurring costs without a clear driver are well-served by a rolling average or a YoY reference.

Use driver-based only when a meaningful, reliable driver exists for the cost line. Equipment and rent are good examples: where cost scales with headcount, model it as # FTEs × cost per FTE. For most OPEX items, statistical is simpler and often more accurate. Avoid forcing a driver where the relationship is weak.

Use hardcoded for committed costs where the amount and timing are already fixed: a signed contract, an annual software renewal, a booked consulting engagement. A signed contract isn't a forecast. Enter the number.

## Where to forecast OPEX

* **Directly on the P\&L:** right for most cases. A rolling average or hardcoded value lives on the cost row itself. Simple to set up, easy to maintain.
* **Supporting sheet:** right when a cost area has multiple inputs or needs granular breakdown (software split by vendors, marketing split by channel). The sheet handles the detail; the P\&L row references the total.

## Forecasting approaches

<Tabs>
  <Tab title="Driver-based">
    Use driver-based when a meaningful, reliable driver exists for the cost line. Add a formula that links the cost to its driver and add the assumption to a separate assumptions sheet.

    Equipment and rent are good examples where cost scales with headcount:

    ```typescript theme={null}
    = "Headcount"[0] * "Cost per FTE"[0]
    ```

    Marketing spend as a percentage of revenue is another common pattern:

    ```typescript theme={null}
    = "Revenue"[0] * "Marketing %"[0]
    ```

    Avoid forcing a driver where the relationship is weak. A loose correlation produces a less accurate forecast than a rolling average, with more complexity to maintain.
  </Tab>

  <Tab title="Statistical">
    The right fit for most OPEX lines. When costs are stable and historical patterns are a reliable guide, a rolling average smooths short-term volatility:

    ```typescript theme={null}
    = avg_last("Software subscriptions", 3)
    ```

    Where seasonal patterns matter, reference the same month last year with a growth adjustment instead:

    ```typescript theme={null}
    = "Marketing"[-12] * (1 + 10%)
    ```

    For lines where trend and seasonality interact, `predict()` handles both automatically.

    Statistical becomes unreliable when a cost line has changed structurally: a renegotiated contract, a shift in spend policy, or a significant scale change.
  </Tab>

  <Tab title="Hardcoded">
    Use hardcoded values for committed costs where the amount and timing are already known: a signed office lease, an annual software contract, a booked consulting engagement. There's no forecasting uncertainty to model; enter the value directly.

    For recurring costs without a fixed amount, hardcoded values require a manual update every cycle and become a maintenance problem over time.
  </Tab>
</Tabs>