How an AI-Driven Markdown Tool Enabled a Fashion Retailer to Optimize Its Seasonal Sales Strategy

The traditional markdown process has always been a labour-intensive process that requires plenty of human attention, functions on bulky Excel-files and is therefore prone to mistakes and suboptimal results.

That is why an established fashion retailer turned to AI-driven analytics & automation to master the challenging balancing act of matching discounts with inventory levels. Here is how.


The client is an established fashion retailer in Belgium and is rapidly expanding in multiple countries within Europe. With brick and mortar shops, its unique e-commerce platform and use of third party e-commerce platforms, the organization has recently grown its footprint through multiple sales channels.

To optimize the operating result of the company, the client decided to make the transition towards more data-driven decision-making and the automation of an essential part of its operations, that is the seasonal markdown process.

How? By embracing smart algorithms, machine learning and AI-techniques.


Its a hard day’s work

The traditional markdown process has always been a labour-intensive process that requires much human attention to calculate the markdowns - one Excel-file in, out the other - and introduce them in the different sales channels.

Every weekly cycle, a team needs to collect and organize all new data, analyze and process that data in record time and present further discounts in time for the next cycle.

A typical Monday for the retailer unfolds as follows:

  • Collecting all data from the brick-and-mortar shops, e-commerce website and third-party e-commerce platforms.
  • Organizing and structuring the different data sets to support the analysis phase
  • Analyze sales trends and the impact of previously defined markdowns on rotation levels.
  • Define the markdown strategy for the upcoming week, taking into account the business objectives set at the beginning of the season. This process is repeated for each country.
  • Collect and structure the newly defined markdown values for all different sales channels
  • Communicate and implement the adjustments to all channels and countries.

This process contains a lot of manual and repetitive tasks, and due to the short deadline, such often leads to suboptimal results. The team hardly has the time to focus on high value-added tasks. Little time is left to evaluate, for instance, additional data elements such as weather impact, footfall data and product characteristics.

In essence; one is tasking employees to function as machine calculators under substantial pressure. It is a way of working that hinders any company looking to scale its activities.

Suboptimal decisions that leave value on the table

Moreover, the abovementioned way of working is a risky business. Typically, items are bought once -or maybe twice- a season in a given quantity, which the purchase teams project feasible to sell. If the items do not rotate as expected, they put the retailer at risk, locking up capital in inventory.

Here comes the daunting balancing act. To free up that locked capital, the retailer can use markdowns to boost the rotation speed of an item. However, a challenging balance needs to be found:

  • High discounts - low margins - lower unused inventory: High discounts might help to get rid of the excess product inventory. Yet, the retailer will be selling at a lower margin than expected, leaving money on the table.
  • Low discounts - high margins - higher unused inventory: Lower discounts may not be enough to get rid of the excess inventory, but can result in maintaining a higher margin than when pushing out at higher discount levels. Due to limited cut in prices however, the risk of having more unsold inventory at the end of the sales period is more present.

Additional elements have their part to play. No-one wants to shop in an empty store. Therefore, retailers with brick and mortar sales channels need to ensure that their markdown strategy is tweaked carefully, verifying that it does deplete the shops’ inventory in the blink of an eye.

One also has to be wary of emotional decision-making. In the traditional manner of setting markdowns, finding the said balance is often driven from a more emotional mindset than a rational mindset. For example, the purchasing team ’hopes’ the product will increase in popularity the upcoming weeks, as they ‘still have a positive feeling about it’.

This traditional retail process is in desperate need of more data-driven decision-making. Retailers have tons of valuable data available that they are not using effectively. Machine learning and AI-techniques can provide technological support to make sure the team is no longer leaving value on the table.

Why the traditional discounting process is ready for a change

  • The current way-of-working still contains a lot of manual and repetitive tasks;
  • People need to collect, organize, process and analyse data in record time;
  • Team members often function as machine calculators, performing tasks humans are not good at;
  • Little time is left to evaluate, for instance, additional data elements such as weather impact, footfall data and product characteristics;
  • The process is prone to emotional decision-making.


To help a retailer manage this daunting balancing act, we introduce the power of machine learning and a set of tools in the field of artificial intelligence.

The key focus is to remove both the repetitive and data-crunching related tasks from the team that handles the markdown process. Such enables employees to dedicate their attention to tasks where humans outperform calculation machines; strategic & creative thinking & decision-making.

Making this distinction streamlines the markdown process and optimizes the results during the discount seasons. Employees and machines work hand-in-hand, each focusing on the most value-adding tasks, in a sustainable manner.

Step one: data capture and collection

First things, first. Everything starts with mapping and gathering the most useful data. Together with the retailer, we define the datasets and data sources required and set up an automated integration that enables the AI-algorithm to source the data from one central place. As the data gathering and structuring is automated, the chance of data errors, due to data handling, declines and no longer requires unnecessary human interaction.

The most commonly used datasets are sales and stock data. We can enrich those datasets with product information, store information, weather information, competitor and market information, or others to get better insights into the buying behaviour of customers.

Step two: capturing business rules

The second step is capturing business rules. Each country, sales channel, brand, or other has its own set of business rules. The retailer, at one point, decided that these are essential from a strategic and legal perspective and should be safeguarded.

The AI-algorithm must safeguard these same rules to ensure that it generates a feasible output. Together with the client’s team, we first map out the all-important business rules. We subsequently source them towards the algorithm and ensure that we cross-validated them afterwards.

In essence, these business rules help the algorithm move within clear boundaries that are defined by the team. What are some examples?

Well, they include, for instance

  • The minimum and maximum allowed depth of a markdown;
  • Price ladders;
  • No-touch periods, e.g. markdown can only be adjusted once a week;
  • Legal constraints in a given country;
  • Projected profit targets for each product group.

Step three: AI-driven markdown generation

Step three focuses on creating a true prediction machine. The solution we build applies an algorithm that crunches the gathered data and searches for the most optimal markdown per product, per market, per time-period. The models’ most important task is to predict appropriate markdown levels that maintain a fair balance between optimizing the rotation rate and depleting the inventory.


Why are these models so transformative?

Well, the solution is capable of digesting significantly more datasets than the human brain. Its role in the markdown process is to detect patterns that are not visible for the human eye and influence decisions that optimize the retailers’ topline.

The model is simply more consistent than an employee and eliminates human error in the application of business rules. It does what it is tasked in record time, relieving the team members from the tasks they just not excel at.

What is the final output of the model?

Well, the tool suggests the discount levels to apply, generated automatically and visualized in the format, familiar to the markdown team. It allows the team to review what is proposed swiftly. It can look at the situation from a more strategic perspective, consult additional information and make manual adjustments for the suggestions they do not feel comfortable with. The final output is subsequently sourced to and applied on the different sales channels.

Meanwhile, the model will in turn learn from these additional inputs, minimizing the need for the team's intervention during the next markdown cycle.

Redefining the tasks of the markdown team

The application of machine learning and AI-techniques is so transformational; it enables the retailer to redefine and reorganize the tasks of its markdown team. Said team does no longer need to put time and effort into data gathering and data crunching. The developed model will take over these time-intensive tasks in a faster and more efficient way which is less error-prone regarding the business rules.

With supporting data visualization, the team can now focus solely on the evaluation of the proposed markdowns and manage the exceptions where they do not agree with the proposed markdown. The model will take these additional inputs into account and learn, creating a feedback loop that makes it even more accurate and taking even less time from the team.

How can an AI-driven tool make such a difference?

By applying AI-techniques, a retailer ensures that no value is left on the table by removing both the repetitive & non-strategic assignments from the employee's task list and handing them over to a prediction tool.

The tool gathers the data, applies business rules and suggests the discount levels to apply. The markdown team can swiftly review, what is proposed. It can looking at the situation from a more strategic perspective and making manual adjustments where needed.



The goal of the AI-driven markdown model was two-folded:

  1. Optimize margins while ensuring a controlled & gradual depletion of each sales channel’s inventory;
  2. Reduce the time spent by the discounting team on gathering, processing, and introducing markdowns on different sales channels, making time available for more impactful strategic analysis.

So, how did we do?

By applying the AI-driven model, the retailer was able to secure an overall margin increase of more than 4% in comparison with the old markdown process.

Moreover, the team is looking at a 50% reduction in time spent after integration of the said AI-solution. This time is now available to focus on company expansion and other more value-added and fulfilling tasks.

How might we help you?

If you are interested in discovering how artificial intelligence and machine learning can assist a retailer in facing its pricing challenges, be sure to get in touch. Together with our team of retail and AI experts, we can guide your team in preparing an appropriate strategy, making technical choices and setting up a tool, generating valuable results in record times.

+ Find out more about our work in Retail & E-commerce

Get in touch with our retail experts!