The new tenets of Financial Forecasting

In keeping with the latest trends, before writing this article I checked with ChatGPT to see what AI had to say about financial forecasting.

What (would a human say) is a good forecast?

If the above question was asked to a CFO, Head of FP&A, VP of Sales or Chief Customer Officer, the answer would almost always coalesce around the following; accurate and comprehensive but flexible and reactive to changes within the operational drivers of the business. As a follow-on, some would observe that the forecast should be fast and efficient to run.

In pre-pandemic survey run by McKinsey, 40% of 130 CFO’s surveyed stated that their current forecast was neither accurate nor efficient.  In post pandemic conversations with Metapraxis customers, an increasing common theme for CFOs has been a reflection on the inability of their current forecasting processes to deal with the increased complexity of externally impacting operational factors. As a result, ad-hoc forecasting is undertaken to support decision making that is then divorced from the core forecasting process.

Were McKinsey were to re-run their survey today, it could be realistic to expect that 50%+ of CFOs would now reflect on the inaccuracy and inefficiency of their processes.

So even though the is a consensus around what a forecast should be (and how it should be achieved), why are businesses unlikely to be able to achieve them?

Disconnected from strategy & finance focussed

The traditional model followed by FP&A teams when creating forecasts is to focus on the traditional finance inputs and outputs of the business; build a P&L for the operating areas of the business (business units, brands, products, geographies etc). Focus on the operational costs of the business, review sales performance and compare versus the budget. Dependent on the variance and the time of year, reallocate cost or revenue across the remainder of the year. Produce a balance sheet and a cash flow statement. Review with the company executives, agree and sign off. Repeat the same process in a quarters time. A recent report from Deloitte highlights the reason why; at the beginning of the 20th century, management decided that it was too much effort to have separate reporting for internal and external audiences. They believed that it was too time consuming (and expensive) to have a difference between the two.

Whilst this process does serve the basic needs of any business, it represents a missed opportunity by focussing on measures which are by nature backwards looking and too far down the chain to provide insight into the reasons why certain results have been achieved. This normally is given via anecdotal commentary or via ad-hoc analysis produced to support a certain set of results. An example for a manufacturing business maybe reporting a rising COGS across business units, geographies, or products, with only written support explaining that raw material costs have risen due to external economic factors. In this example the analysis is deep (in that it covers the depth of the dimensions of the business) but is also shallow, not providing the empirical insight into a core driver of performance – a historical trend view of the material costs being incurred would have shown this sooner.

The primary issue that this approach has created is that forecasts are far removed from the actual strategy and operations of a business. This disconnection means that core non-financial measures, or operationally significant metrics are deprioritized (or even ignored) for discussion. This is not a new observation, W.E Deming highlighted in 1984 that one of the deadly diseases that management suffers from is only using the “visible” figures to operate the business.  Deming believed that this was one of five hindrances that prevented businesses from achieving transformation (Deming was concerned with transforming western businesses to match the level of performance achieved by the Japanese post WW2).

Building better practises for forecasting teams

In choosing to break away from the traditional models of financial forecasting, what are the new practises that FP&A teams need to follow? The below provides an overview of the new core tenets that should be followed in the creation of forecasts, but not exhaustive as each industry and business within an industry will have its own particular aspects which should be covered. These should be applied to the below, based on a finance teams experience to complete the forecast.

Tenet 1 – Define the purpose for the business

Most businesses operate multiple forecasts, even a moderately sized business may run a sales forecast, operations/demand planning forecast as well as a finance forecast (this may then be further divided between a sales/revenue recognition forecast and a financial forecast). Similarly, each may also create and use its own plan values. Specifically in the creation of a financial forecast, careful consideration should be given to ensure that information is re-used efficiently, that logic is re-used particular to a business function, so that the data created can be recognised and acknowledged by those who it should be used by.

Answering the following questions will provide a start to define the forecasts intended purpose.

• What is the time horizon that the plan needs to cover – 6/12/18/24 months?
• How often should the forecast be refreshed – monthly/quarterly?
• Who will the forecast inform within the business?
• What decisions will be taken using the information?
• Based on the decisions to be made, what operations or activities should be carried out by the business (and therefore should be measured)?

The above questions should form the basis of function wide consultation, generating a consensus and buy-in to a forecast that will be produced. It should also enable teams to discuss the options for rationalization and the reduction of the variations of forecasts which exist within the organization.

Tenet 2 – Driver based forecasts for both internal and externally influenced aspects of your business

Bias in forecasting is one of the biggest reasons why inaccuracy and frustrations are experienced. By nature, they are hidden from view and are included at almost every level of entry. This is especially true when creating forecasts from the bottom up, each layer of the hierarchy will undoubtedly massage their numbers for any number of reasons. In areas of the business such as Operations, Manufacturing, Sales (transactional or high-volume sales led), Treasury and Transaction processing (AP, AR). In these areas, building a driver-based model built around the mathematical relationships between the core operating activities of that area of the business and the outcomes creates a forecast which is driven by inputs (empirical) rather than beginning with individuals’ assumptions (and biases).

Driver-based forecasts are also advantageous in these areas as they can be rapidly refreshed when new input data becomes available. For example, a software business can rapidly refresh its forecasted revenue based on using their latest sales pipeline, or number of downloads. A professional business services forecasting its people costs can use an average cost per person per level (and by type; permanent and contractor) and its planned headcount numbers to create the overall cost to the business. In both internal examples, forecasts can quickly be changed, and the outcomes reviewed.

For driver-based forecasting based on externally influenced aspects of the data, external datasets should be used. In areas such as treasury this data will be well known and accepted within the business (yield curves, GDP & inflation forecasts) however careful consideration needs to be given to using external datasets. For a consumer goods business, internal sales data plus a dataset from the wider market can enable the creation of an accurate market share %, which trended over time can be used to create a baseline for likely future performance. Changes in the market can then be reflected in your forecast. However, selecting this data needs careful consideration; is it accurate historically and do I believe that it will remain so for the forecasting period?

Tenet 3 – Return on investment should drive forecasts in remaining functions

Outside of areas in which drivers can be applied (and which should be evaluated on operational metrics), other areas of the business should be forecasted based on an expected return on investment. This is a challenging prospect for many, how do you generate a target ROI for a HR department? Whilst on the surface this may appear challenging, when broken down it is a simple equation – any department within the business consumes costs and resources. In this example, the cost of the HR teams and their systems. The activities that they perform have a financial outcome which can be measured; a training program for employees that results in an improved conversion rate in sales calls. The increase is revenue generated can be weighed against the cost of the activity to ascertain the ROI. Similarly, the cost of employee churn can be measured, the cost of recruitment, the time taken for an employee to become productive. For forecasting, these calculations can be reversed, allowing you to create a forecast of costs.  

Other have described this as choice-based or investment-based planning. It has gained popularity within organisations which practise zero based budgeting (AB-InBev has been taking this approach since 2005). However, some have argued that this form of forecasting should not be used for functions such as Legal or Finance, and instead forecasting for these areas of the business should be driven by an allocation or % of the overall budget. The issue with taking this approach is that it encourages the thinking that certain areas of the business have a necessary or inevitable cost to the business. Instead, by focusing on the ROI there is a necessary shift to thinking about the benefits the department will bring for the spend.

Tenet 4 – Avoid benchmarking as a method of validation

The temptation to use external consultants to provide a validation of forecasts is strong, further external validation of a forecasting methodology and the forecast itself results in a de facto approval. Having access to benchmarked metrics from the industry creates a feeling of safety – if you a line with the industry you are probably doing the right thing, or maybe you are outperforming? Either way, external benchmarking exercises are too often used to validate decisions and ideas which have already been taken.

This must be tempered however, some benchmarks are an entirely necessary and cannot be ignored (dependent on the stage at which the business is at this will vary from those benchmarks used by potential investors to those use by shareholders to judge performance). Like the selection of external data to support operational, driver-based forecasting, care must be taken in the selection of datasets. When provided as part of a consulting exercise, or if purchased, the integrity of the data should be verified and understood. If to be used as a onetime validation of a forecast, the providence of the data should be checked (volume of companies surveyed, length of time surveyed for, consistency of calculation…).

Final thoughts

The four tenets mentioned above summarize an approach to be taken which is reliant upon having an underlying technology platform to perform the work within. If there is one key message that this article delivers, let it be that for those businesses which want to improve their forecasting, the first step is to stop using Excel or Google Sheets. All the above (and much of the additional functionality that could be implemented – artificial intelligence, machine learning etc) requires a volume of data which make these (very useful and venerable in Excels case) tools redundant. Embracing the right technology platform allows you to adopt the tenets, with the emphasis on the areas of most need for your business.

For some, accurate sales forecasting based on a high-quality sales pipeline will be the most important aspect of a forecasting platform. For others, the ability to model the changing cost of raw materials, import taxes and transportation costs will matter more. Before selecting a dedicated FP&A technology, detailed conversations should be had with the vendor to understand if the platform supports the areas of forecasting most important to the business. Not all platforms are alike, ensuring that the forecasting capability provided matches the specifics of your business is key to beginning your journey towards improving your organisations forecast capabilities.

Finally, what does AI say?