Many times, we have heard from executives the following statement ‘I don’t believe in making forecasts based on past data’. Or ‘how is it possible to predict the impact of an event which has never happened’. While the capacity to predict the future may sound counterintuitive, it is a field where AI thrives.
In a business context, the most common use of AI is for predictions. AI can unleash considerable profits and performance to managers by enabling them to better plan ahead. One example : every company, when selling a product or service needs to forecast the demand in order to have an idea of how many products to produce.
The purpose of this article is to help managers understand how AI differs from traditional prediction approaches, and in which cases an organization should use AI instead of traditional forecasting.
How are forecasts traditionally done?
Traditionally forecasts are being made using different methods, but with the same goal: understand the market to predict the future. Let’s take a closer look at 4 different methods:
- Judgmental bootstrapping: this approach consists in translating an expert’s rules into a quantitative model by regressing the expert’s forecasts against the information that he used
- Game theory: this approach consists in watching the competitions or other key stakeholders of your ecosystem, and adapting your strategy accordingly;
- Extrapolation: this very common process consists in doing simple estimations based on the observational range. For example if the price increased by 2% within the last couple of years, we can assume this will be the same next year
- Simple regression analysis: is a causal approach, meaning that you will plot all data points and determine patterns between 2 variables, not more.
All of these methods start the same: gather as much information about the market as possible, because by understanding your target market it will be easier to estimate their demand.
Each of these methods give the experts the ability to make an educated guess on how the future will look like. But when speaking with clients, we hear a couple of constraints companies have with these types of methods. We have grouped these challenges into 3 categories. The below challenges can be found separately or can be combined.
- Dimensionality: These methods are limited when it comes to solving problems with different variables (ie: elements which impact a business problem, and therefore a prediction) . They have a very limited capacity to find hidden patterns in terms of the impact of a variable on the sought prediction. As a result, managers risk to make business decisions with a limited outlook
- Limited horizon: In the field of forecasts, the further you go, the harder it is to have an accurate forecast relative to a certain time horizon. For example, how will the demand for my product evolve in the next few months?
- Volatility : Accuracy can be very low in case of fast-changing environments as experts have a limited capacity to analyse multiple, complex data sources at the same time. As a consequence, managers are not able to have a reliable forecast to make decisions in real time.
Managers should choose the methods above in the following cases:
- In case of simple prediction problems where accuracy can be high;
- When there is a limited number of variables to be taken into account;
- In case of business cases which have a limited impact on company profits.
Examples of where these methods can succeed
The best use for the traditional methods is steady state situation such as the lunch rush through a cafeteria or predicting the number of tellers needed at a bank. In these situations, traditional forecasting work because past experiences are very useful for this type of forecasting.
What does AI bring to forecasts?
Due to the diverse challenges with traditional methods, AI has come as a blessing for many companies!
As explained, companies will traditionally make educated guesses on what the future holds based on expert’s expertise and previous decisions. Artificial Intelligence should be seen as a tool that will enhance this process and make it more performant and helps you create a bigger picture. Many companies have been gathering lots of data which can enable them to find hidden patterns and make better decisions. Artificial Intelligence has the power to process all relevant historical data and make a more precise forecast on how the future will look like. This helps businesses have a competitive advantage since they can act appropriately.
How AI differs
AI based predictions are different from traditional methods in different points of views. Here are 3 important ways :
- Learning from data: AI predictions are based on a key principle which is learning from data, even complex data. An AI model can find hidden patterns by making sense of a complexity and that is usually out of reach for humans. This model is also continuously improving itself by learning from the results of its predictions.
- Real time analysis of multiple data sources: Thanks to a high computational (calculation) capacity, AI enables companies to process data sources at a speed that is much higher than what humans could do. The data can be based on historical data, internal data or external data sources, depending on the project needs.
- Wide scope: AI gives companies the capacity to analyse higher and multiple quantities and longer time-lapses of data than any group of human experts could ever apprehend to have a better grasp of a problem. More data is not seen as a challenge, but as an opportunity to become better.
Added value for businesses
Based on our experience with predictive analyses we see the following added values that Artificial Intelligence brings to businesses :
- Higher accuracy: Because of its wide scope, AI enables companies to have more performing predictions. This is its core benefit in a business context. In a recent project, Kantify achieved to increase the accuracy from 70% to 98% in predicting the price of a raw material, this by finding hidden correlations that were undetected before.
- Understanding impact of variables: AI can detect variables that you didn't even know had an impact on a prediction, and quantify their impact for your business. In a recent prediction project on churn, Kantify found over 900 elements (called ‘features’) that had a significant impact on the churn rate of clients, and quantified their cost for our client. Of course, there were features that had a higher weight than others. On the contrary to what is sometimes said (the famous ‘black box’), AI models can be easily explainable for managers. In a business context, it is possible to develop an algorithm that can be understandable, by justifying in detail to users how and why the model recommends a given prediction or decision.
Managers should opt for AI predictions in the following cases :
- In case of complex, non-linear problems where many elements can influence a business problem or a prediction;
- In case of volatile environments, where immediate impacts of certain events can’t be easily measured;
- In case of expensive problems where higher accuracy can result into higher profits.
Price, churn and demand forecasting are areas where we have seen formidable impact with AI.
Price forecasting is when you forecast the evolution of the price of a certain product or material that you either sell or buy. This gives you the opportunity to better forecast your revenue streams. It can also be useful to adjust your production capacity or marketing actions.
Churn forecasting is forecasting how likely it is a client leaves your company in a certain time span. This is especially useful because a company can adjust their strategy towards their clients based on how likely they are to leave the company soon, and what are their reasons.
The need for demand forecasting is omni-present in every company. This type of forecasting determines how the demand for a certain product / service, or product family will evolve during a certain period of time. Demand forecasts can be used in different areas such as optimising warehouse management, foreseeing relevant promotion and sales actions, negotiate discounts with suppliers and manage cash-flows.
In all of these areas a lot of data needs to be analysed in order to make a forecast with high accuracy, and therefore gain higher profits.
Conclusions: combining AI prediction to human expertise
Traditional forecast methods are becoming outdated and outpaced due to easier access to large datasets and an increase in the available computational power. The more complex and changing the environment, the more traditional methods lose their value due to the low accuracy rate and being so time consuming.
Will AI replace experts? The answer is no.
In a business context, AI is there to augment managers so they can take more precise data driven decisions. Having a reliable prediction that everyone understands can help facilitate the decision on priorities and actions across departments, in a data driven way. Also, AI can help to valorise experts knowledge even more by combining their knowledge to an algorithmic prediction. This approach is particularly powerful in the case of rare events (rare phenomenons such as the firing of top management). Finally, a prediction is nothing without a well executed decision. AI simply gives the opportunity to executives and managers to make very well informed decisions and actions, thereby considerably increasing their chances in succeeding.
If you want to get monthly reads directly to your email inbox, sign up for our newsletter!