- The automation scale
- AI at the business level
- The advantages of automation
- 4 tips for implementing an automated forecasting solution
Artificial intelligence is a high-speed train that some innovative companies have already jumped on. It is a tremendous growth lever representing more than 12 trillion in potential earnings worldwide, every year. Within a decade, companies that adopt AI will see their positive cash flow increase by 122%. The podium of companies that perform in their markets will therefore be reserved for those who adopt AI.
We all know AI in many forms, in our daily lives, but also in our work. Chatbots, and some of our everyday applications (Netflix, Uber Eats), use AI to automate and personalize content, to reason better, to interact better with the consumer, or to classify better.
The application of AI can now help humans in their work decisions by using it in predictive analysis to better anticipate business.
In order for forecasts and processes to be as reliable as possible, it is essential to set up routines and rules, but also to learn from the data in a continuous way to be able to anticipate better. This is partly achieved by automating processes and forecasts.
The automation scale
You see here a graph illustrating automation, with level 4 being the most autonomous, real-time and prescriptive level; and level 1 being the basic level where the company will rather react to events, have an internal perspective and not be subjected to its ecosystem.
AI at the business level
AI is applicable to all businesses. It allows your company's business units to better collaborate with each other.
On the marketing side
When launching new products, the question arises as to how much the new product will sell. Artificial intelligence will allow you to analyze, automate and iterate the different scenarios to help you make the best decisions.
In the same way, on a pricing issue, artificial intelligence will allow you to answer the questions of pricing or product managers on the best price compared to the competition, or the adjustment of the price in real time.
Only AI allows us to make progress on these issues in a totally unprecedented way. It allows to predict sales with much more accuracy, much more reactivity, whether it is sales by store, by region or by channel, and obviously, to better choose the product assortment to better maximize the margin or simply the turnover on commercial operations.
On the supply chain side
AI is an precious ally in controlling the supply chain. For example, when it comes to replenishment for a procurement manager or supply chain director, humans are now subject to such a large amount of data that it is almost impossible to iterate thousands of relevant scenarios in Excel or in traditional APS, given the multiplicity of our data.
In fact, to calculate the right pricing and to optimize replenishment for more than 50,000 references and more than 300 points of sale, we have to take into account the multiple monthly promotions launched by marketing, and to integrate logistics business constraints: solving the equation in terms of maximizing the service rate and optimizing safety stock is a miracle. Better sales prediction thanks to artificial intelligence allows to securise its decision by integrating a multiplicity of scenarios.
The advantages of automation
Consideration of the environment
A manual forecasting system will usually not take into account exogenous data, trends in particular geographic areas, store types, seasonality, or cannibalization effects between different product types.
An automated system will allow you to understand trends even before you have access to all the contextual data. It will combine the company's history with exogenous and contextual data, allowing for improved accuracy. The system itself finds the correlations between the different parameters, which removes the bias that the human could add. In the case of manual forecasts, you have to be careful about the budget effect, which consists in trying to match the forecasts to certain business objectives, thus leading to biased forecasts.
More detailed forecasts
An automated system also makes it possible to produce models at the finest level, allowing an in-depth understanding of the behavior of stores, customers, products, and seasonality, thanks to the detection of trends by the algorithm.
A more efficient process
After collecting sales and historical data via extractions or API connectors, the automated prediction system will establish which type of model will be the most relevant and efficient on the set of topics to be predicted, working by iteration. The predictive platform can handle thousands of combinations much more efficiently and quickly than humans.
The automation of the prediction process therefore saves time and resources. Indeed, a spreadsheet-based forecasting system represents a significant workload to maintain Excel forecast files, operate them and run simulations. An automated solution produces forecasts instantly, allowing you to allocate the time and resources saved to higher value-added tasks. The predictive solution also allows for a gain in value since fewer errors are made.
IntegrationAn automated solution is designed to be as painless to implement and use as possible by functioning as an overlay that retrieves data from the client's existing IT system. The model makes predictions and then generates and sends them to the target systems.
These predictions can be consulted in different ways. They can be re-injected into APS (advanced planning system) systems, finance and cash management systems, CMS, BI tools, or even be sent back via an API that will integrate them into third-party systems. They can therefore be used at all levels of the organization without the need to transcribe or modify files or even to question existing business software.
4 tips for implementing an automated forecasting solution
1 - Measure your goalsIt is essential to focus on issues where the ROI is easy to achieve and where the gain can be measured with clear KPIs. The focus can be on a multitude of subjects: products, resource optimization, transportation, revenues, financial management, marketing operations.
It is also crucial to define the business metrics that you want to monitor. For example, one can use the resale rate, the value of the residual stock, or the MAE (margin average error).
2 - Collect the most exhaustive sales data possible
Machine learning models work with historical data so they can learn. It is recommended to have at least 3 years for the machines to learn especially when events like COVID disrupt the business. In retail, for example, one can use store data, sales data, sales receipts, available external data (weather, Covid, pollution, traffic) and marketing data such as store location to decipher the environment to try to characterize the impact of store location. All this is possible only if the data is available. If not, it would add time to the project and to the implementation of the solution.
3 - Prepare a scalable process
It is important not to consider the implementation of an automated solution as a test, but to directly design the application to see it deployed throughout the operational scope. This implies integrating the constraints of use, communication with the stores and the business from the start. There is often a tendency, in POCs (Proof of Concept), to reduce the scope, which means that we miss out on other elements that are only found in industrialization. So you have to keep this aspect in mind when designing an AI project.
Today, predictive intelligence has a very strong potential. We have extremely mature models, whether it be for anticipating customer demand, service levels, or increasing the capacity of employees, which allows us to give the models the means to be more precise in their predictions, in their forecasts, in their decisions and in the management of their data assets.
Finally, it is important to keep in mind that forecasting is not only a business, IT or innovation oriented subject, but that it concerns all these branches. It is therefore crucial to invest in them together from the start, with a single objective and a clear alignment on the subject.
4 - Anticipate user experience
The famous sentence of Martin Leblanc CEO of Icon Finder speaks for itself: "A user interface is like a joke. If you have to explain it, it's not that good."
The parallel can be drawn with forecasting solutions, as they respond to a need for ease of use. It is not up to the user to adapt to the forecasting system but the other way around. The format and the way in which the forecasts are to be delivered must not disturb the final recipients. This aspect must be taken into account from the very beginning of the project.
For this reason, Verteego has designed a system to extract Excel files for flat file forecasts that integrate with existing tools to be as painless as possible in operation and use.