Solution fitting : AI for retail is not one size fits all.
With numerous AI powered retail solutions available, it is a complex task for retailers to select the solution that brings the most value to their business. Kirahvi, a demand forecasting tool developed by Jalgos adapts to each clients business context in order to produce high quality custom results.
The use of AI within the retail industry is expanding every year. That is a projected 35% increase from 2019 to 2025. But using AI to respond to business needs is not one size fits all. Retail innovation spans from chatbots to product recommendations, inventory management, price optimization, among other uses. Furthermore, a given solution selected based on intuition alone may not produce results that are operationally or financially relevant to the company. Retailers are tasked with sorting through an abundance of AI solutions and gadgets in order to select, source, test, and analyze the impact of the results.
Kirahvi, a demand forecasting tool to optimize your inventory replenishment strategy
Kirahvi, a comprehensive demand forecasting solution by Jalgos enhances inventory replenishment strategy through optimizing restocking processes for delegated stores. Using proprietary prediction machine learning models, Kirahvi systematically adapts its technology to best fit each client in order to produce a demand forecast for a given week, month, or season with acute accuracy. But demand forecasting is not the end all be all. Instead, the value extracted from demand forecasting is what is essential in measuring the impact of Kirahvi.
With accurate demand forecasting, retailers can optimize their inventory replenishment strategy in the following ways :
Decrease the need for product discounts due to overstocking or approaching expiration dates
Decrease in missed revenue due to out of stock articles
Reduction of transportation costs related to restocking products in stores
In the pursuit of cost reduction for retailers, a varied approach gives the best results. In addition to its demand forecasting model, Kirahvi offers complimentary add-on features in order to extract the most value from clients’ available data resources.
Price optimization & product recommendation as features necessary to reinforce the value of sales prediction estimations
The features present in Kirahvi are used interdependently to reinforce the impact of sales prediction results. An optimized inventory replenishment strategy is reliant on additional insights and metrics to function at maximum efficiency. Price management of articles, and product recommendations are necessary insights used to reinforce the effects of sales prediction estimations to ensure an equilibrium between the supply and demand of products.
The following add on features were conceived through a need based and iterative process during the development of Kirahvi.
The price optimization feature analyzes historical sales data in order to determine the most relevant price per article at a given moment in time.
Kirahvi’s product recommendation feature creates article clusters based on customer buying habits to give insight into marketing and retail-experience strategies.
With tangible business value as the core objective, AI powered solutions must be capable of capitalizing on client data resources meanwhile taking into consideration their business context in order to maximize impact.
Demand forecasting success case: the adaptation of a sales prediction tool for a multinational lingerie retailer
The notion of solution fitting is multifaceted. Not only is it necessary for retailers to choose the best type of solution for their business, but the engine within the solution must take into consideration contextual differences from one client to the next.
The following use case illustrates the process required behind producing accurate and unique sales prediction results for a mid-size retailer.
This particular multinational lingerie retailer operates with more than 200 stores in France, and wished to integrate Kirahvi in order to obtain precise sales predictions for up to 85,000 products. High heterogeneity in sales patterns prevented us from straightforwardly training Kirahvi on such raw data with the risk of poor results.
Using data clustering to produce custom sales prediction results
A crucial step in Kirahvi’s prediction pipeline consists of clustering articles and stores whose customer buying behavior can be deemed as similar. By creating larger aggregations of articles and stores where buying patterns are similar, Kirahvi’s model is capable of delivering accurate sales predictions. Our clustering engine produces results that are unique for each client and make up a core block of the prediction model. This is key in order to produce custom and accurate prediction results per client, as opposed to generic model training.
By creating these clusters of articles and stores, useful business insights emerge from the data allowing us to confirm the accuracy of our clustering model. Some business insights include :
Stores located near the Atlantic coast have similar sales patterns
Stores located in the city center are clustered together and therefore experience similar sales patterns.
High-end lace products have similar sales patterns with other ‘seductive product lines’
These business insights extracted from the clustering process of the client’s data, not only serve to confirm and nourish our sales prediction model, but allow the client’s business teams to benefit from precise and insight driven monitoring to aid in decision making processes.
A ‘product recommendation’ feature to complement sales prediction results
Kirahvi has an advanced article/store segmentation method made up by two levels of analysis. It takes into consideration aggregated data of stores and articles as mentioned above, and also relies on a robust analysis of customer buying specificities (how much customer X has spent for a given article in a given shop at a given time). Customer buying specificity data allows for high accuracy in clustering results, however requires a detailed and systematic tracking of customer data from the retailer.
Customer buying specificity data is a core building block of Kirahvi’s product recommendation system, where products are recommended to customers based on an analysis of other customers with similar buying patterns.
An example of product recommendation logic is the following :
Customer A has spent a high amount of money for article a
Customer B has spent a high amount of money for article b
Clustering reveals that ‘a’ and ‘b’ articles are similar therefore, it might be interesting to recommend article b to customer A and article a to customer B.
The recommender system allowes us to identify clusters of products that have similar stylistic characteristics (combination of colors, shapes, typologies) among more than 10,000 articles by using more than 400,000 client profiles.
Kirahvi’s recommender system creates a double opportunity for value. The recommender system itself serves as an additional functionality that the client can benefit from to increase the use of data in their retail strategy (targeted email campaigns, product recommendations for their website). However the customer buying specificity data also increases the robustness of the clustering process and final sales prediction results.
Solution fitting in the quest for value is a multidisciplinary process in the retail sector
The adaptability of Kirahvi’s algorithmic motor in the form of data clustering, new feature proposals, and multilayered data analysis highlight the level of customization necessary to compute accurate and meaningful results.
Solution fitting for the conception of business focused products is a multidisciplinary process. The careful construction of machine learning methods, the layering of prediction models and optimization algorithms based on micro-economic phenomenons exemplifies the complexity of retail focused product development in the quest for value.
Through its unique machine learning approach, Kirahvi was designed to respond to these constraints and continues to evolve to tackle the retail industry’s challenges.
