Case Studies




Key questions




Stores chain Performance Management

Learn how a leading company in the hearing care Industry improved their retail chain profitability.

Maximizing the profitability of a retail chain through an interpretable, generalizable and efficient Machine Learning strategy

A global leading company in the Hearing Care industry needed to improve their accuracy in the measuring the retail chain performances, as well as a strategy to boost revenues over time, yet without increasing costs or undermining service quality. Such necessities needed to be met through the optimization of a series of KPIs identified by the client through their experience over time. Furthermore, a “white box” solution was required – that is, a result generation process that is interpretable in each of its stages. aHead Research proposed a modeling solution for the retail chain that could evaluate performances and was based on an unsupervised learning method for the segmentation of the stores, linear and non-linear regression models to model interaction between the KPIs, and a mathematical optimization engine providing optimized KPIs to guide the improvement of the store chain. The need for “interpretable” results that were coherent to the business flow, as well as the presence of data noise, makes it necessary to design ad-hoc algorithmic solutions. Among the necessities that might need satisfaction: objectively identifying critical KPIs for each store of the retail chain and the desirable target to reach, providing a robust estimate of the complex correlations between them, guaranteeing a solution where it is possible to apply “what-if” analyses, enabling a full exploitation of the model by a domain expert to create evaluation scenarios for specific KPIs.

Key questions:

  • Is it necessary to cluster similar stores to guarantee a fair comparison? If so, what criteria should be followed clustering?
  • What objective criteria make it possible to identify critical KPIs for each of the stores?
  • For a given KPI, how can a target value be quantified? How can we guarantee that such value is achievable?
  • How can we estimate correlations in case of data noise?
  • What is the most accurate class of algorithmic optimization problems?
  • Are the above-mentioned solutions truly “interpretable”?
  • Is the solution fast enough for on-demand, What-If analyses?
  • Is the solution generalizable and reusable by other businesses?

Such questions can be answered through the employment of specific algorithms within the context of an overall strategy aimed at guaranteeing transparency along the entire decisional process.

On the one hand, the optimal value of each KPI needs an accurate calculation and solid theoretical bases as to guarantee revenue maximization. On the other hand, all parts of the decisional process need to be objective, explainable, coherent to the business flow and computationally efficient, ensuring a concrete implementation of the solution as a Decision Support System in a corporate context.





The problem

Identifying critical KPIs and suggesting a feasible span of improvement for each store.

Identifying critical KPIs and quantifying an achievable span of improvement that is also coherent to the business flow is a complex management activity. The decision is based on both objective performance data and the personal know-how and experience of the manager, elements that are intrinsically objective.

Large quantities of KPIs -that may be increase by the addition of new ones- and the complex correlations between them makes the task increasingly challenging. Once a decision is taken, its outcome can be evaluated only by quantifying its effects on the revenues in the following trimesters and, in any case, the decision drivers are not entirely “explainable”, in that they are based on the intuition that the manager developed with experience. For these reasons, it becomes necessary to transform the decision-making process in something that is fully objective, interpretable, and automatic.

The solution

Our solution to maximize the profitability of the retail chain.

The Data Intelligence Team of aHead Research developed a Python-based solution which implements a Decision Support System (DSS) based on Machine Learning models and a mathematical optimization engine. In particular, the solution is made of four sequential blocks:

  1. Hierarchical clustering – an unsupervised clustering of the stores.
  2. Critical KPI selection and estimation of target values – for each store, selection of KPIs for improvement (critical) and identification of a reachable and coherent target value.
  3. KPI correlation model – an artificial intelligence model that captures the complex correlations -i.e., relationships graph – between actionable levers and KPIs at the specific level of a singular store.
  4. KPI optimization – a mathematical optimization engine to provide optimal, reachable values for critical KPIs, with the aim of maximizing the revenues of the specific store.

Thanks to this technology, it is possible to provide concrete support to managers for major strategic decisions related to the management of the retail stores:

  • Identifying main areas for improvement in a quantitative and unvarying fashion across the stores through the analysis of historical data.
  • Establishing feasible and coherent improvement targets.
  • Understanding the complex interactions between KPIs and conducting What-If analyses to simulate the quantitative impact of variations on key KPIs.

The benefits

The benefits gained from the optimization of critical KPIs

The proposed solution was successfully applied to the large store chain -around 500 stores- managed by the global leading company for Hearing Care in Italy. For each store, a comparison was made between revenues from the 2ndtrimester of 2021 -without the support of the AI strategy- and revenues from the 2nd trimester of 2022 -with the support of the AI strategy. To enable a fair comparison, the growth trend was deducted from the overall revenues of 2022. The results show a significant growth in revenues for almost each store, an improvement made possible by the ability of the AI strategy to enhance the current state of the retail stores and identify optimal improvements for the critical KPIs. The suggestions made by the solution were positively evaluated by the managers, both in terms of critical areas identified and the feasibility and coherence of the proposed corrections.

Moreover, the level of generality of the AI strategy and its independence from the type of business analyzed makes the technology adaptable to -almost- any other type of store chain. The sequential steps and interpretability of the applied Machine Learning models make it possible to understand the way in which results are generated. Such transparency is particularly convenient for businesses, in that the proposed solutions not only need to be efficient, but also easy to explain to stakeholders. The AI strategy is parsimonious in the number of parameters it employs and does not require specific hardware or long processing times. Therefore, it represents a useful simulation system, where dozens of different What-If scenarios can be tested to evaluate the impact on revenues of different KPI corrections.