In recent months, Dutch Detailresult Group (DRG) and Israeli Shufersal signed multi-year contracts with MIcompany to use AIRPro to further optimize their promotion strategy. For both retailers, this is a continuation of their partnership with MIcompany.
The foundation for the partnership between DRG and MIcompany was established in 2016 when they joined forces to evaluate and predict the effectiveness of DRG’s promotions. For this, they needed to develop a methodology that would be generic enough to apply to all promotions, and at the same time, capture the unique behavior of over 30 thousand different products. By applying the extensive AI knowledge and experience of MIcompany to DRG’s detailed and rich data, and integrating DRG’s specific retail expertise, the challenge was solved. This resulted in a granular methodology, in which 10 different value drivers that influence promotion effectiveness were identified.
In 2017 Israeli retailer Shufersal joined AIRPro, after which the promotion effectiveness methodology was further improved. It was applied to the specific Israeli environment, with more promotion pressure and strong holiday-driven seasonality.
For both retailers, the use of AIRPro led to significant changes in their promotion strategy. More than 200k promotions were evaluated, and based on granular insight into the underlying drivers and real effect, more than 5% of promotions were stopped due to their negative contributions. Many other promotions were changed to maximize impact, for example, by adjusting the discount, placement, or the level of media coverage. Overall, this led to an average increase of 20% to 40% margins on those promotions.
“With the AIRpro as a compass, we have set-up procedures to optimize our promotion management process, together with MIcompany. As our promotion management results have a direct impact on our company results, this was crucial. We have stopped throwing money and energy into irrelevant efforts, and have seen major improvements in our promotion results. We will continue improving our performance using AIRPro”
– Ilan Talon, Shufersal’s Head of Commerce
To enable retailers to continuously steer promotion effectiveness, AIRPro is directly integrated with existing systems and processes. Therefore it can easily be used by all colleagues, including those without technical or statistical knowledge.
“Thanks to the integration of AIRPro with our internal promotion system, category managers can easily combine the operational side of promotions with value optimization. Furthermore, the output of AIRpro is directly fed back into our systems and used in commercial discussions, the logistic process, and further analysis within the analytics department”
– Thijs Nootenboom, Manager Advanced Analytics at DRG
Since the start of the collaboration with DRG and Shufersal, multiple innovations have been implemented, both functional and technical. In 2017 the Model Management solution was created, continuously tracking the performance of all underlying models. Based on this tracking mechanism, many model improvements have been implemented over the past years, adjusting for changes in the market, both small and large (such as the COVID-19 impact). In 2018 the Supplier module was added, calculating the value of promotions for the supplier, compared to its promotion participation. This information is used to create a win-win promotion for both the retailer and the supplier. In 2019 the Lost Sales component was developed, giving insight into the demand for promotions by correcting for missed sales due to insufficient supply.
AIRPro runs completely in the cloud, using SnowFlake as the DWH and running all other components on AWS. This further standardizes the data architecture and all model components compared to the previous on-premise solution. This innovation reduced runtimes of processes by >75%, simplified data exchange processes and decreased costs.
Shufersal is the largest supermarket chain in Israel, with over 2 million weekly customers, in 275 stores. In addition, Shufersal operates a shopping website, a credit card club and a pharma chain (BE). Its extended private label has over 1700 products, featuring among other groceries, cleaning supplies, beverages and fish.
Detailresult Group (DRG) is the overarching organization combining the strengths of the supermarket chains Dirk van den Broek and Dekamarkt. It is a unique combination of two family businesses, founded by Dirk van den Broek and Dirk Kat respectively. Weekly they serve over 2 million customers in more than 200 stores.
AIRPro is an AI-based promotion optimization tool, enabling retailers to optimize promotion decision-making, reduce lost sales and product waste, and support supplier negotiations. To this end, AIRPro evaluates the real promotion effect based on 10 different effects, both direct and indirect. This sophisticated and data-driven method gives retailers far more accurate insight into their pure promotion uplift than any other approach.
“Calculating the actual added value of a promotion is very complex due to the different effects it has, short and long term, on the promoted product and its substitutes. With AIRPro, we are able to calculate the actual generated value of our promotions and gain surprising new insights”, says Talon
Furthermore, the application uses a combination of machine learning algorithms that extract patterns from previous promotions to predict the ROI of a new promotion upfront. This real-time prediction enables the user to try different promotion settings and determine the optimal mix. This intelligence is combined in a user-friendly software solution, which can be used by, amongst others, category managers, data analysts, and the logistics department.
Talon: “Our category management teams evaluate their promotions based on the actual value created, reviewing AIRPro-driven new gains and avoided losses. Also, new promotion choices are based on effectivity predictions”.
AI cannot grasp all retail-specific details. Therefore, AIRPro is has a unique ability to create synergy between AI-driven recommendations and user input. It deploys a couple of features, which creates two-way feedback, allowing both the algorithms and users to continue improving over time. First, AIRPro gives a reliability score for any new prediction, indicating whether subject matter input is needed and allowing for changes in the prediction if the reliability score is insufficient.
Secondly, AIRPro also contains the Model Management functionality, which provides insight into the algorithms’ realized prediction performance. This is monitored over time, and models are regularly updated and upgraded to constantly ensure model robustness.
Thijs Nootenboom, Manager Advanced Analytics at DRG: “Based on our past experiences, we are well aware that models, especially within retail, can show a steep decline in accuracy if maintenance is neglected. The Model Management functionality provides us with a clear view of the model’s reliability and gives us the confidence that results can be trusted”.