When thinking of financial forecasting and the utilization of AI, the concept of time series is most often used. Time series forecasting is a statistical method to predict future values based on previously observed values. These kinds of models often make use of the one-way ordering of time so that values for a given future period are expressed as a derivation of past values.

It is true that time series based financial forecasting works best in a stable environment when the dynamic between variables can be reliably captured across time. In such conditions, some of the more skillful Financial Controllers can produce good enough estimations without AI, using Excel models. But in a black swan event, the unpredictable nature of the situation renders both time series and smart Excel models useless. It drives the need for agile and fast scenario planning, and this is where AI very much outperforms any other method, including Excel.

• Achieve fast and reliable scenarios by letting AI to do the heavy lifting

Preparing for more than one “official plan” has historically helped companies to manage their risks and asses their options in normal times. But it’s especially when facing high uncertainty, scenario planning become crucial in navigating a crisis effectively.  It’s important that reliable scenarios can be generated fast, given the operational lead time there often is for production- or project planning. Having a good scenario planning model in place that rapidly provides insights into best case / worst case type of scenarios helps a company understand and prepare for the possible financial implications, including the need for funding, reorganizations, or the need to delay certain investments.

For many companies, creating fast and reliable scenarios has proven to be a difficult goal, but now it can be much easier achieved with the use of AI. Like in other planning models, manual estimations are often biased and slow, since they require coordinated input from multiple parties. Embedding AI in this process helps automate the fine calculations while dramatically improving precision, requiring only to enter the key assumptions of the scenario at hand. While companies still have to invest time to train the AI to understand logical correlations between (mostly) non-financial inputs and their financial outcomes, once this is done the AI provides for a very powerful ally in the creation of different scenario models in a fraction of the time previously required Think about expectations related to covid infections, weather situations, or financial market indicators – it’s becoming increasingly easy to input these expectations into a well-trained AI model to see how they would impact the company’s financial forecast.

Today, one thing is certain: there will always be uncertainty. Companies can leverage AI to help deal with that uncertainty and prepare for the most probable outcomes.

^[1]www.investopedia.com

1. Does the CPM solution determine the “best fit” algorithm automatically or does a data scientist need to identify it?

Answer: A major development in tFivehe CPM technology field is that software can now be simply fed with data and it can identify the “best fit” algorithm by itself. It does that by picking an algorithm from the library that explains best the relationship between provided parameters. For the few percent of cases with very advanced requirements, a data scientist can still develop a custom algorithm and upload it to the platform. For most companies, especially those at the beginning of the AI journey, the out-of-the-box feature is more than enough and certainly a great first step.

2. How many variables on average are needed for a correct prediction?

Answer:  The number of independent variables depends entirely on the complexity of the business’s market forces. An AI driven forecast implementation should be regarded more like a journey than an exercise to get the perfect fit right away. Starting with a few common sense and easily available variables that influence your market is enough (ex: GDP, unemployment, consumption trends etc) to start seeing improvements compared to a manual forecast process. With time, the team can experiment with more variables and further improve the model.

3. Does a black swan event as COVID impact forecast/ budgeting powered by AI?

Answer: Black swan events such as the Covid pandemic are impossible to predict during a regular corporate forecast, but companies using AI driven forecasting are much better at reacting to such events. In crisis times, given the uncertainty, one single forecast may be insufficient to mitigate the array of possible risks. In these cases, the best practice is to create multiple scenarios by adjusting some of the levers in your forecast model. After a short while, a measurable driver of the crisis that impacts your company’s performance can be identified (such as COVID infections rate) and incorporated into the forecast model, from then on things can resume to business as usual. As for budgeting, AI used right can enable more swift and coherent re-allocation of resources in crisis situations and avoid paralysis by analysis types of situations.

4. Have companies begun to take black swan events more into consideration after experiencing the COVID pandemic?

Answer: Before COVID, only a small percentage of companies were planning for more than one scenario. As the pandemic struck, the interest has since risen across the board. In most organizations the finance department was expected to play a significant role in containing the impact of the disruptions and safeguard the value of the company. Scenario planning is an effective tool to mitigate risks associated with low risk & high impact events, both before and after the occurrence of the event. AI can make scenario generation easier, more accurate and less time consuming, giving the finance office time to model an appropriate plan of action.  

5. If you need at least 5 years of historic data, does it only work for “stable” companies?

Answer:  There are many ways in which data can be considered “unstable”. If the data is not consistent across the organization in terms of definitions and timing, a certain amount of cleaning can be performed.  On the other hand, if the company has been acquiring new businesses or making changes in the product portfolio, existing models for similar products or high-level models can be leveraged until there is enough data for a more granular approach. In most cases there is a way to start using AI at less “stable” companies, but each case has to be assessed individually.

Closing words:

Similar to other business functions that have already heavily adopted AI, it’s a certainty that the Office of Finance will go the same path. The gains in time and accuracy are making the use of AI in finance processes a compelling proposition. If you have more questions, wish to discuss how AI can impact your processes or set up a webinar for your team, please reach out Vadim Stoian.