When the forecast is always wrong: what CFOs need to rebuild in FP&A

A global consumer goods company runs a detailed 12-month forecast every quarter. By month three, the variance to actuals is already 15%. By month six, the planning team has quietly issued two "re-forecasts" that nobody outside finance ever sees. The original numbers are still being presented to the board. This situation is not unusual. According to research from Gartner (published in prior years but still widely cited across the industry), fewer than 25% of finance organizations report high confidence in their forecast accuracy. The gap between the forecast and reality is not a data problem. It is a design problem.

Why FP&A keeps producing the wrong numbers

The structural issue in most FP&A functions is not the model. It is the process around the model.

Most organizations still operate on a calendar-driven cycle: annual budget, two or three formal re-forecasts, and a monthly close that feeds into variance analysisvariance analysisVariance analysis compares actual financial results against budgeted or planned figures to quantify differences and explain why they occurred.View full definition →. This rhythm was designed for a world where business conditions changed slowly. In 2026, with supply chains still absorbing post-pandemic restructuring, interest rate environments that shifted dramatically between 2022 and 2025, and AI beginning to affect headcount planning in ways that are genuinely difficult to model, a quarterly re-forecast cycle is simply too slow.

The deeper problem is that most FP&A teams are still building forecasts by starting with last year's numbers and applying growth assumptions negotiated through a political process. Business unit leaders shade their revenue projections upward to look optimistic. They pad cost budgets to give themselves room. Finance aggregates these inputs, applies a corporate overlay, and calls it a plan. The result is a number that satisfies the internal approval process but has limited predictive value.

Driver-based forecasting is the alternative that most CFOs are familiar with in theory. The idea is to model the business through its actual operating variables: units sold, customer acquisition costcustomer acquisition costCustomer Acquisition Cost (CAC) is the total sales and marketing spend divided by the number of new customers gained in a period. It measures how efficiently you grow.View full definition →, average contract value, headcount by function, capacity utilization. When a key driver changes, the financial output adjusts automatically. FP&A teams at companies like Unilever and Siemens have been building toward this model for years, with varying degrees of success. The difficulty is not the concept. It is agreeing on which drivers actually predict financial outcomes, and maintaining the discipline to update those drivers with real operational data rather than estimates.

Rolling forecasts present a related challenge. Many CFOs advocate replacing the annual budget with a continuous 12 or 18-month rolling view. This is directionally correct, but it only works if the organization genuinely abandons point-in-time budget targets as the primary performance metric. If the rolling forecast is simply layered on top of an unchanged incentive structure, teams will manage to the original annual number and treat the rolling forecast as a secondary reporting exercise.

What this means for the CFO

The CFO's role in FP&A is not to build better models. It is to make better models possible by changing the organizational conditions around forecasting.

That starts with data access. A forecast is only as good as the operational signals feeding it. In most large organizations, the data that finance actually needs, pipelinepipelineAll active sales opportunities across the stages of the sales process, together with their combined potential value and probability of closing.View full definition → conversion rates, customer churncustomer churnChurn rate is the percentage of customers or revenue lost over a period. It measures how fast a business loses its existing customer base.View full definition → by segment, procurement lead times, sits in systems that FP&A does not have direct access to. Closing that gap is partly a technology question, but mostly a governance question. CFOs who have successfully improved forecast accuracy tend to have made formal agreements with CROs, COOs, and HR leaders about which operational data points will be shared with finance on a regular cadence, and in what format.

The second implication involves how uncertainty is communicated. The standard forecast is a single number. This is useful for certain purposes, including external guidance and incentive targets, but it systematically hides the range of outcomes that finance has actually modeled. Presenting scenarios, a base case alongside a downside and an upside with explicit assumptions attached to each, changes the conversation with the board and with business unit leaders. It forces the organization to think probabilistically rather than treating the plan as a prediction.

CFOs at mid-size companies sometimes resist this approach because they worry it signals a lack of conviction. The opposite is true. A CFO who can say "our base case assumes 4% volume growth, but here is what the business looks like if we land at 2%, and here is what changes operationally in each scenario" is demonstrating more analytical rigor than one who presents a single forecast and defends it.

The third area is talent and tooling. AI-assisted forecasting tools from vendors including Anaplan, Workday Adaptive Planning, and Oracle (each with commercial interests in positioningpositioningThe mental space you want your brand to occupy in your target customer's mind relative to alternatives.View full definition → their platforms as forecasting solutions, so vendor claims about accuracy improvement should be treated with appropriate skepticism) have matured considerably. The more important point is that these tools require FP&A teams with the skills to design driver logic, interrogate model outputs, and translate financial scenarios into operational language. That skill profile is different from the traditional FP&A profile, which rewards Excel proficiency and month-end speed.

Practical moves worth prioritizing

• Audit your current forecast against actuals for the past eight quarters. If variance consistently exceeds 10% at the EBIT line, the process needs structural change, not incremental refinement.
• MapMapUsing software to automate repetitive marketing tasks and campaigns, enabling personalisation at scale across channels like email, web, and social.View full definition → which operational data sources are not currently flowing into your financial model, and identify the one or two that would most improve predictive accuracy. Start there rather than attempting a full data integration program.
• Replace at least one formal re-forecast per year with a scenario review: no new point estimates, just explicit discussion of which assumptions have changed and what the range of outcomes now looks like.
• Be deliberate about what "good forecast accuracy" actually means for your organization. Hitting within 5% at the annual level may mask large monthly swings. Define the metric that matches how your business actually makes decisions.
• Assess whether your FP&A team's incentives reward accuracy or reward budget attainment. These are not the same thing, and most organizations measure the wrong one.

The best FP&A functions in 2026 are not the ones with the most sophisticated models. They are the ones where finance and operations have built enough trust to share real data, have honest conversations about uncertainty, and adjust quickly when conditions change. That is a management problem, and it belongs on the CFO's agenda.