A good forecasting process should allow organisations to have a better understanding of where they are headed based on current actions and plans. With this visibility, finance leaders can make and enable better decisions such as identifying and acting on opportunities for growth or risks to strategic plans; the reallocation of resources throughout the year; and how to ensure cash needs are met.
There is no single best practice to achieve this – a good process depends on the characteristics of the organisation and sector as well as how the forecast is being used. However, there are leading forecasting methods that are almost universally applicable to businesses. These are outlined below along with features of a good forecasting process and the key challenges and aspirations relevant to many organisations today.
What are the leading forecasting methods?
Driver-based forecasting models causation, using metrics (drivers) that materially impact the financial measures of interest. This allows more accurate and explainable forecasts, helping focus the organisation on drivers of performance and facilitating scenario analysis. Most organisations (94 percent according to FP&A trends) have now partially (50 percent) or fully (44 percent) implemented driver-based forecasting for income statements.
Integrated (or extended) forecasting refers to the integration of financial forecasts with strategic and operational forecasts. Achieving this requires a collaborative, cross-functional approach and results in business ownership of forecasts and alignment on cycles, assumptions, and metrics. AFP reports that 75 percent of organisations have partially (58 percent) or fully (17 percent) aligned planning between finance and the business / operations. Where relevant, changes in the external environment should also be monitored, anticipated, and reflected.
Organisations should also integrate:
- Forecasting across the 3 financial statements (income statement, cash flow and balance sheet) to ensure material impacts have been reflected, consistency and ease of updates. Surprisingly, only ~17 percent of organisations currently achieve this using integrated drivers – instead requiring manual updates to the balance sheet and cash flow. For smaller organisations, balance sheet forecasting may not be critical, but income statement and cash forecasting should still be carried out and integrated.
- Short-term and longer-term forecasts (for instance by connecting summary outputs from detailed forecasts into a simple long-term model), to track against strategic plans and easily evaluate multiple scenarios.
Rolling forecasts extend the time horizon by 1 period with each forecasting cycle, and when combined with a driver-based approach, achieve earlier visibility of future impacts of current actions and events (and form an early view of budgets). Appropriate time horizons vary by sector but a 12 to 18-month detailed (feeding a 3 to 5-year long range) forecast is typical. Traditionally forecasts would only extend to year-end, and despite this severely limiting the value of the forecasting, many organisations continue to do this.
While rolling forecasts in the strictest sense can sometimes prove cumbersome, similar approaches (e.g., monthly reforecasting but adding a new quarter every 3 months) can still ensure a meaningful time horizon is always covered.
What makes a good forecasting process?
Efficient, fast and frequent
Your forecasting process should be as efficient as possible, allowing your forecasts to be produced quickly and frequently (as well as being cost-effective). In terms of speed, forecasting in 2-5 days or less is generally good practice. However, that’s not the case for many, with APQC reporting a median time of 16 days and AFP reporting only 8 percent of FP&A teams can reforecast easily. Quarterly or monthly frequency is likely to suffice in many cases, though specific elements such as sales or cash might be updated more frequently as appropriate. Regardless, the process should be as efficient as possible to allow updates whenever required.
FP&A teams typically spend 46 percent of their time manually collecting, combining, cleaning and manipulating data. Largely focused on obtaining actuals (e.g. for metrics which may not be in the general ledger/ERP), this is often exacerbated by poor integration between operational and financial data. Finance teams should aim to eliminate this task, so that actual data is readily available or generated automatically – effective data governance across the organisation is therefore a key enabler and this is a higher priority in lager organisations with more data
Similarly, the process of producing the forecast should be as streamlined as possible. A clear, standardised process for updating the forecast (aligned with the business), can help achieve this (including policies on eliminating low level adjustments). For larger organisations, APQC recommends using a centralised support team to handle the routine/transactional elements. While Excel is the most used forecasting tool and may be appropriate for smaller organisations, when multiple stakeholders and/or FP&A team members become involved or when large data volumes are involved, other systems/methods such as cloud planning tools (or even building models in Python/SQL) should be considered.
Given the data and modelling skills demanded, this is a key area where FP&A may need to hire specialist skills into the team and/or collaborate closely with technology / data functions.
Unbiased and automated
Forecasts should provide an honest and unbiased view of direction based on current plans and should be independent of budgets/targets. One view of best practice is using an automated (naïve) forecast as the starting point for the forecast process. This ‘automated first draft’ can be used for discussions with stakeholders, adjustments to the forecast being captured along the way and allowing more time for value adding conversations.
CIMA found this to be effective in de-biasing stakeholder input and facilitating more efficient and productive conversations. For instance, in order to justify deviations from the automated forecast, stakeholders were prompted to think through how they would achieve those results. Issues were also uncovered earlier, giving the company more time to act. However, there was still a need for leveraging local knowledge about extraordinary events and structural changes, and incorporating recent and tacit information.
A best-in-class solution would even allow multiple models (including machine learning-based) to be evaluated for best fit and automatically improved over time. As well as achieving dramatic cost savings, reforecasts in minutes rather than days, and enabling better decisions, these forecasts often prove more accurate than those incorporating stakeholder judgement.
While requiring technical skills that not every FP&A team possesses (and might not be realistic for smaller organisations), various companies have already automated their forecasts (including the 2 studied by CIMA, Microsoft, and indeed some banks were prompted to develop the capability to meet needs for rapid stress-testing scenarios). This approach would most suit larger organisations that have their data in order, and is likely to become increasingly common as the relevant technology and skills increase in availability.
Was this article helpful?