Tapping Into New Markets with AI-Driven Innovation and Process Optimization

In the summer of 2022, one of the leading manufacturing companies in the world requested Aigoritma to build a business case for the digital transformation of their Retail division in Turkey. This division oversees sales operations in one of the company’s key emerging markets. 

With more than 30 brands and immense amounts of data, it would have been incredibly challenging to keep up with all of the company’s strategic initiatives for 2023. One of their main goals was to win over another one billion consumers around the world — and the Turkish market in particular — through innovation and process optimization.

Retail and AI

Human intelligence and domain expertise are no longer enough. What is needed is their application speed for overcoming constantly emerging new challenges in the Retail industry.

The number and variety of these challenges keep everyone related to solving critical business problems caused by these disruptors under too much pressure. 

The ever-growing consumer demand for always-accessible products—with detailed and personalized descriptions, tailored to their needs, offerings, and promotions, corresponding fully to their habits and lifestyle throughout the whole buying journey. The modern consumer wants to have user-friendly online channels with various payment methods and immediate delivery without any delays—and available on their mobile. It’s not hard to understand, but it’s pretty difficult to organize. 

Early AI adopters within the Retail industry are already harvesting the fruits of a frictionless customer experience by winning the majority of their wallet share — while still decreasing costs by optimizing operational efficiencies. But it’s not too late to join the Avangard of the industry. 

Opportunity

With more than 80K employees across 150 countries on five continents, this company has been using AI for its primary markets for some time. But when the time for emerging markets came, finding skilled data scientists was challenging. Setting prices, managing inventory, and maintaining proper headcount, to prevent loss of sales opportunities and higher costs for urgent working schedules—combined with increasing sales volumes and without a single Data Scientist on the Turkish team—would have been “Mission Impossible.” 

The Aigoritma team and the Data & CRM Manager of said company decided to focus on their two main priorities:

• Improving customer experience 
• Increasing sales volumes with ML Studio Recommender System to generate personalized product offerings for their online shopping channel with machine learning 

Recommendation engines help retailers with e-commerce channels create individual product recommendations based on purchase history, clicks, views, and more sophisticated customer buying behavior analysis. This solution helps maximize conversions from the main page to the payment, improve customer satisfaction, reduce shopping cart abandonment, and allocate future marketing budgets wisely. 

The ML Studio Solution

There are several different approaches when pursuing Personalized Product Offering Engines. In this approach,  identifying the best fit for a particular client by testing various strategies is crucial. For instance, if a particular product has become popular in the last week for some age groups, it won’t necessarily be a good offering for other ages. So, depending solely on one strategy is not considered best practice. But in a resource-constrained world, testing more solutions might be a luxury not everyone can afford.

The original dataset shared by the client company contained 589,881 transaction records that took place in a 10-month period in 2022. The training dataset for this machine learning project consisted of features such as:

• Customer static attributes (demography, age, and city) 
• Customer dynamic attributes (average spending, min spending, max spending, average basket size, number of transactions, and number of transactions for a particular product—each calculated for various time windows) 
• Product static attributes (price) 
• Product dynamic attributes (min price, max price, discount, number of purchases,  popularity—each calculated for various time windows) 
• Customer-item similarities 
• Item-item similarities 
• The target variable was whether or not the purchase occurred

Since ML Studio is a data-centric platform, the final dataset consisted of transactions that happened before, along with transactions that have not happened. The reason for this is because we are proposing additional products and downsampling those negative samples to create a balanced dataset. This combination balances the dataset and provides more accurate outputs. After the Data Preparation part, Machine Learning Algorithms were ready to learn from the fully prepared historical data. 

ML Studio blended this data into four different strategies in the POV for the given Manufacturing company by using Automated Machine Learning to provide their customers with the most suitable product recommendations. These strategies were empowered with the best-fitting machine learning algorithms. They used past transactions and similarities among user items and generated a set of products, ranked by relevance and demand for each and every customer in the dataset. 

Product offering strategies ML Studio tested were:

• Repurchases 
• Collaborative Filtering (User-based similarity) 
• Content-based Filtering (Item-based on similarity) 
• Trending Items

To validate the results, ML Studio Data Scientists split the past week’s transactions for each customer and scored how many transactions our recommendation algorithm could predict. ML Studio’s result for this POV was an increase of 28.72% per average basket size. 

If you’d like to see how Automated Machine Learning could optimize your business and bring you closer to your organization’s goals, feel free to contact us and get similar research and a POV.

Get in touch for a personal demo.