One of the most persistent — and costly — mistakes in modern FP&A is treating forecasting and planning as the same activity. They are fundamentally different disciplines with different purposes, timeframes, and success criteria. Conflating them is one of the primary reasons AI and ML forecasting tools underperform their potential.
Kaemingk's article resonates particularly strongly with organisations that have invested in AI and ML forecasting tools but seen disappointing results. In most of these cases, the root cause is exactly what this article describes: the models were trained on budget-contaminated data, learning to replicate management aspiration rather than identifying genuine business patterns. The fix isn't better AI — it's cleaner data separation.
The Core Distinction
A forecast answers: 'What will happen?' A plan answers: 'What do we want to happen, and how will we get there?' A budget is a financial expression of the plan — the targets and resource allocations that follow from strategic choices. These are three different questions, and they require three different processes.
- Forecast: objective, probability-weighted estimate of future outcomes based on current data
- Plan: aspiration-driven target that reflects strategic goals and management commitments
- Budget: resource allocation that funds the activities needed to achieve the plan
Why Mixing Them Breaks AI and ML Tools
Machine learning forecasting models are trained to predict future states based on historical patterns. When organisations feed them 'budgets' as training data — numbers that were shaped by management aspiration rather than observed reality — the models learn to replicate biases rather than identify genuine patterns. The result: AI that confirms what management hoped for rather than what the data suggests is coming.
The Rolling Forecast as the Solution
Decoupling forecasting from planning means adopting rolling forecasts — continuously updated views of expected outcomes that are explicitly separated from budget targets. Rolling forecasts have no commitment embedded in them. They exist to improve decision-making, not to hold managers accountable. This distinction is essential for honest forecasting.
- Rolling forecasts are updated at least quarterly, often monthly
- They extend 12-18 months forward regardless of fiscal year
- Variance analysis compares actuals to forecast, not to budget
- Management incentives are tied to plan achievement, not forecast accuracy
Building the Separated Process
Implementing true separation between forecasting and planning requires both process changes and cultural shifts. Finance teams must resist pressure to adjust forecasts to meet budget expectations — a practice known as 'sandbagging' or 'stretch targeting'. This requires explicit leadership commitment that forecast accuracy, not optimism, is valued.
When you train an AI model on budget data, you're not teaching it to predict the future — you're teaching it to predict what management hoped would happen. Those are very different things.— Nate Kaemingk, FP&A Trends (Forecasting vs Planning, Jun 2025)
Practical Implementation Checklist
- Conduct a data audit: separate your historical 'actuals' from historical 'budgets' in your forecasting system and ensure ML models are only trained on actuals data
- Create separate reporting views for plan vs. forecast: leadership should always be clear whether they're looking at a target (plan) or a prediction (forecast)
- Establish a formal rolling forecast process with a defined cadence (monthly or quarterly) and a single owner — the Finance Director or VP FP&A should own forecast integrity explicitly
- Change your variance analysis practice: report actual vs. forecast (prediction quality) alongside actual vs. budget (performance accountability) — each serves a different purpose
- Communicate the distinction explicitly to business unit leaders: explain that the rolling forecast is not a target and that producing accurate forecasts (even unfavourable ones) is valued over optimistic ones
- Audit your incentive structures: if managers are rewarded for 'beating budget' but not for 'accurate forecasting', you have a structural incentive for sandbagging that no process change will fix
Separating forecasting from planning and budgeting is one of the highest-ROI changes an FP&A team can make — and one of the most culturally difficult. The teams that succeed are those whose CFOs explicitly champion forecast accuracy over forecast optimism, creating the psychological safety for finance professionals to call the future honestly.
Key Takeaways
Forecasting ('what will happen') and planning ('what we want') are fundamentally different activities
Conflating them causes AI/ML models to learn biases instead of patterns
Rolling forecasts, decoupled from budgets, restore forecast integrity
Variance analysis should compare actuals to forecast — not to budget
Cultural change is as important as process change: rewarding accuracy over optimism
ML models trained on budget data learn management aspiration, not business patterns — the data contamination is the root cause of most AI forecasting disappointments
Separate reporting views for plan vs. forecast are a structural prerequisite for maintaining the distinction
