Smart Retail: the role of AI in the future of the industry
Nowadays Artificial Intelligence plays a crucial role as a driver of innovation and efficiency in numerous sectors; the retail industry is not an exception. Artificial Intelligence tools based on predictive or generative engines are particularly relevant in terms of decision support and process automation.
Predictive algorithms can offer various advantages in terms of time and effectiveness, providing decision-makers with information that facilitates their choices and supports them with an analytical approach. Similarly, these tools can be used to accelerate slow processes that do not fully exploit the operator’s skills to create added value for the company.
These advantages are observed in various business areas, so the benefits of adopting these measures are both horizontal and vertical for companies. In particular, the areas where such improvements are visible include:
- Production: using algorithms for optimal production planning, predictive maintenance, and energy optimization, generating reports on machinery status, or automating product design;
- Logistics: employing algorithms for delivery optimization and inventory management, generating reports on delays, and assessing the supply chain, including sustainability considerations;
- Marketing: customer clustering, churn prediction analysis, simulating the impact of promotions, scraping competitors’ prices, and generating optimal pricing, as well as automating marketing campaigns;
- Sales/E-commerce: utilizing algorithms for demand forecasting, up-selling, cross-selling, suggesting alternative products when items are unavailable, and offering customer recommendations based on their profiles.
The use cases must be aligned with the specific business context. Companies that operate on a make-to-stock basis cannot take actions at the production level, just as physical stores that do not have an online presence cannot leverage e-commerce-related benefits.
The potential of AI offers numerous opportunities. However, determining a clear starting point and creating a structured roadmap to benefit the company can be challenging.
In the following, we specifically present a case of how Intellico supported a company operating in the retail sale of organic/local products in selecting its priorities.
Case study: demand forecasting for retail sales of organic products
The project consisted of two phases:
- Service design workshop to identify AI application priorities within the company
- Implementation of the key use case (demand forecasting)
The service design workshop involved key stakeholders from various business functions, covering all critical processes (e.g., Procurement/Logistics/Marketing/Sales Network).
Through the workshop, each stakeholder had the opportunity to specify the areas in which they felt the need for AI enhancement and identify the potential risks associated.
It emerged that optimizing demand forecasting could create a differential value in response to:
- Business scalability needs: risks associated with the growth in supplier orders to serve an increasing number of sites and geographies
- Expected customer service standards: risks of losing customers due to product unavailability
- Monitoring and growth driving: limited capacity for analysis and understanding of sales results and management to intervene at a strategic level
During the second phase, stakeholders were tasked with creating a use case for the refined demand forecasting system. This included specifying the main user, outlining the value added, and discussing the consequences of not implementing the system.
Finally, the involved actors were asked to define intervention priorities, evaluating their relevance to the business and the complexity of implementation. This resulted in a matrix where the most relevant aspects for the specific use case were mapped according to the two criteria mentioned earlier. This allowed for the identification of elements to prioritize as they were both relevant (high-value quadrant) and feasible in the short term (low complexity).
From the matrix analysis, the following points were identified:
- Success criteria for the model: increasing the granularity and accuracy of forecasts by considering operational parameters (e.g., stock levels) and the peculiarities of different product categories (e.g., fresh products vs. long shelf-life items)
- Project success evaluation criteria: actionability, defined as the ability to provide forecasts that can translate into concrete short-term actions (e.g., products with a shelf life of 1-15 days)
- Data sources/features to be used in the model: selection of company data sources containing key parameters relevant to the use case.
Regarding the second point, it is important highlight that the implementation of the proposed solutions hinges on the availability of input data. Typically, the accuracy of algorithmic predictions is directly tied to the quality of the input data. This data may relate to customers, products, or external factors like macro-trends and competitor insights. How well this data is gathered plays a pivotal role in the success of the implementation.
The next step involved implementing the demand forecasting model, testing various options, including LSTM and XGBOOST. The algorithm was initially trained and tested on products with higher daily sales volumes and then extended to products with progressively lower sales volumes. This approach allowed for an agile path of iterative algorithm improvement in collaboration with the client’s data science team. This approach resulted in:
- Achieving an increase of up to 10% in demand forecasting algorithm performance compared to the baseline (per SKU);
- Enabling the data science team to understand the underlying logic of the algorithm’s evolution for its effective management and refinement based on changes in products, customers served, and markets.
