--- url: 'https://chatfin.ai/blog/driver-based-planning-needs-drivers-you-can-measure/' title: How to Pick Drivers for Planning author: name: Tara Parker url: 'https://chatfin.ai/author/chatfin_admin/' date: '2026-08-07T17:20:00+00:00' modified: '2026-09-28T06:03:41+00:00' type: page summary: 'How to pick drivers for driver based planning: the test a driver has to pass to be useful, and why most plans fail on unmeasurable ones.' published: true --- # How to Pick Drivers for Planning ![Two colleagues reviewing a printed financial report together](https://images.pexels.com/photos/8297478/pexels-photo-8297478.jpeg?auto=compress&cs=tinysrgb&w=1920&h=1100&fit=crop) Explainer · FP&A # Driver based planning needs drivers you can measure By [Ashok Manthena](https://www.linkedin.com/in/ashokman/) The model links headcount to volume and volume to revenue. Neither volume figure is captured anywhere finance can access on a monthly basis. CF ChatFin 8 min read · Published August 7, 2026 Key Takeaways - Driver based planning replaces line item extrapolation with relationships between operational activity and financial outcomes. - It fails when the chosen drivers are not measured consistently, which is more common than the modelling literature suggests. - A small number of well measured drivers outperforms a comprehensive model built on estimates. - The relationship between driver and outcome should be tested against history before it is used to forecast. - Where a driver cannot be measured monthly, it is not usable regardless of how well it explains the business. Driver based planning is a clear improvement over extrapolating last year's numbers. Instead of growing a cost line by a percentage, the model links it to the activity that causes it. The theory is unarguable. The practice depends entirely on whether the drivers can be measured, monthly, reliably, in a form finance can access without asking someone. In many companies the honest answer for the most explanatory drivers is no. Volumes are counted in an operational system nobody has reconciled, or they exist as a report someone produces manually. A model built on drivers that cannot be measured is a more elaborate way of guessing, and it carries more authority than the guess deserves. That authority is the actual risk, because a driver model presented with structure and detail is challenged less than a simple extrapolation would be, even where the inputs are weaker. Anyone presenting one should be able to say which drivers are measured and which are assumed, which is the standard applied in [any figure you have to defend](https://chatfin.ai/blog/the-subtotals-you-will-have-to-defend/) and the reason [being unable to explain a number](https://chatfin.ai/blog/the-controller-who-cannot-explain-the-number/) is a governance problem rather than an inconvenience. Driver Based PlanningFP&AOperational DataModellingForecasting 01 ## What makes a driver usable It has to be measured rather than estimated, in a system of record, on the same basis each period. It has to be available on a cadence that matches the forecast, which usually means monthly and shortly after period end rather than eventually. It has to have a demonstrable relationship with the financial outcome it drives, tested against history rather than assumed from logic. And it has to be stable enough that the relationship holds, which is why volume drivers usually work and satisfaction or engagement measures usually do not. ![A finance professional working through figures at a desk](https://images.pexels.com/photos/8296981/pexels-photo-8296981.jpeg?auto=compress&cs=tinysrgb&w=1600)*A driver has to be measured, timely and tested before it belongs in a model.* 02 ## Four reasons driver models fail None of these is about the modelling approach, which is why replacing the tool rarely helps. **The driver is not measured:** It exists conceptually and nobody records it consistently, so the model runs on estimates.**The data arrives too late:** Available quarterly when the forecast is monthly, which forces an assumption in the intervening periods.**The relationship is assumed:** Never tested against history, so a plausible link is treated as an established one.**Too many drivers:** A model with forty drivers requires forty inputs, most of which nobody maintains, which is the [unpopulated dimension](https://chatfin.ai/blog/the-dimensions-you-never-populate/) problem in another form.03 ## Test the relationship first Before a driver is built into a model, the relationship between it and the financial outcome should be tested against several years of history. The test is simple: does the driver explain the movement in the outcome better than time alone does, and is the relationship stable across periods. Frequently it does not. Headcount and revenue may both be rising without one driving the other, and a model that assumes causation will forecast confidently in the wrong direction when growth stops. Doing that test costs an afternoon per driver and it removes the drivers that would otherwise produce authoritative errors, in the same spirit as [close metrics that change behaviour](https://chatfin.ai/blog/close-metrics-that-change-behaviour/). A plausible relationship is not a tested one. **Test it before it forecasts anything.** 04 ## Choosing drivers that work Three criteria applied to every candidate driver, and most fail at least one. | Criterion | Test | Common failure | | --- | --- | --- | | Measured | Exists in a system of record on a consistent basis. | Produced by a person in a monthly report, which changes when they leave. | | Timely | Available shortly after period close. | Available quarterly, which forces assumptions in two months of three. | | Explanatory | Explains historical movement better than time alone. | Correlated with the outcome without driving it. | Model design should be tested against your own history rather than adopted from a template, since the relationships that hold vary considerably by business model. " A driver you cannot measure monthly is an assumption with a job title. ChatFin 05 ## Building a model you can actually run Five steps that produce a smaller model that works rather than a comprehensive one that does not. #### List candidate drivers with their source Where each is measured, by whom, and how quickly it is available. #### Eliminate anything not measured monthly However explanatory it appears, since it will be estimated in practice. #### Test the survivors against history Several years, checking whether the relationship is stable rather than merely present. #### Build with five to ten drivers Covering the majority of the cost base, with the remainder forecast simply rather than elaborately. #### Reconcile driver data to the ledger Where possible, so an operational figure and a financial one cannot diverge unnoticed, as [operational data outside the ledger](https://chatfin.ai/blog/per-load-profitability-lives-outside-the-ledger/) shows. 06 ## Two failure patterns in practice Both produce a model that is abandoned within a year. Pattern ### The comprehensive model Every cost line linked to a driver, requiring dozens of monthly inputs, most of which are eventually estimated by the FP&A team itself. Pattern ### Drivers nobody owns The operational figures come from a team with no stake in the forecast, so they arrive late or inconsistently and the model degrades quietly, in the way [unowned data](https://chatfin.ai/blog/data-quality-has-no-owner-in-finance/) always does. 07 ## How ChatFin supports driver data Our contribution is making the driver data available and reconciled, which is the constraint rather than the modelling. **Drivers extracted on a schedule:** Operational and financial drivers pulled on the same cadence as the close rather than requested each period.**Reconciled to the ledger:** Where a driver has a financial counterpart, the two are reconciled so divergence is visible.**History assembled for testing:** Several years of driver and outcome data prepared so the relationship can be tested rather than assumed, alongside automated actuals preparation. Model design, driver selection and the assumptions in the forecast remain decisions for your FP&A team. Free tool ### AI Cost Estimator for the CFO Office Four inputs. A first year range, a timeline, and three delivery models compared. No login required. [Open the estimator](https://chatfin.ai/enterprise-ai-tools/ai-cost-estimator-for-cfo-office/) ## Frequently Asked Questions ### How many drivers should a model have? Usually five to ten, covering the majority of the cost base, with the remainder forecast on a simple basis. Comprehensive models require inputs nobody maintains, and the maintenance burden is what causes them to be abandoned rather than any flaw in the approach. ### What if our best driver is only available quarterly? Then it is not usable in a monthly forecast without an assumption in two months of every three, which reintroduces exactly the guesswork the model was meant to remove. Either fix the measurement cadence or choose a different driver. ### How do we test a driver? Against several years of history, checking whether it explains movement in the outcome better than time alone and whether the relationship is stable across periods. Correlation without a stable relationship produces confident forecasts in the wrong direction when conditions change. ### Who should own the driver data? The operational team that produces it, with a stated obligation to deliver it on the close cadence. Where the data comes from a team with no stake in the forecast, it arrives late and inconsistently, and the model degrades until it is abandoned. Start from what you measure ## The model is limited by the data, not by the method. Listing candidate drivers with their sources, eliminating anything not measured monthly and testing the survivors against history produces a model that survives contact with a reporting calendar. If you want a first year range for driver data extraction and reconciliation, the estimator gives one in about a minute. [Open the estimator](https://chatfin.ai/enterprise-ai-tools/ai-cost-estimator-for-cfo-office/) [Book a Demo](https://chatfin.ai/talk-to-us/) Related Reading CloseThe close that starts before the period ends [FP&AScenario planning stopped being an annual exercise](https://chatfin.ai/blog/scenario-planning-stopped-being-an-annual-exercise/) [DataThe chart of accounts was designed for people](https://chatfin.ai/blog/the-chart-of-accounts-was-designed-for-people/) [DataMaster data is the automation project](https://chatfin.ai/blog/master-data-is-the-automation-project/) Explore this cluster - [AI FP&A and Forecasting: Automated Planning, Budgeting, and Variance Analysis](https://chatfin.ai/solutions/fpa/) Put it on your own numbers ## See what this looks like in your ERP. We will walk through the workflow behind this article on the system you already run, using your own data, and show what a first phase would actually cover. *✓*Runs on the ERP you already have, on the version you already run*✓*Every action prepared as a draft that a person approves before it posts*✓*A first phase scoped against your real workflows, not a generic packageSchedule a Demo Confidential & private Please enable JavaScript in your browser to complete this form.Please enable JavaScript in your browser to complete this form.Name *Email * MessageSubmit![Loading](https://chatfin.ai/wp-content/plugins/wpforms-lite/assets/images/submit-spin.svg) This article is general information for finance and accounting teams. It is not legal, tax, accounting or investment advice. Dates and thresholds are given as published at the time of writing and several of these timelines have moved before, so confirm the current position with your advisers and with the relevant authority before acting. Photos by Mikhail Nilov, free to use under the [Pexels License](https://www.pexels.com/license/) (commercial use, no attribution required). © ChatFin  ·  [Blog](https://chatfin.ai/blog/)  ·  [Contact](https://chatfin.ai/talk-to-us/)  ·  [chatfin.ai](https://chatfin.ai/)