Data Driven Merchandise Planning: When Data Designs The Collection

Data Intelligence

,

Fashion

Data Driven Merchandise Planning: When Data Designs The Collection

Andrea Ardizzoia | Nov 18, 2020
  • The disparity between the items in a catalog and the ones that actually turn out to be successful is a constant problem in Fashion Retail and leads to reduced sales margins, wasted natural resources, and extra costs incurred by unsold and returned items.
  • Though this is becoming increasingly strategic and complex, merchandise planning risks depending too much on the experience of a small group of experts which companies in the sector contend for.
  • Machine learning technology has been on the market for many years now. A more recent phenomenon, however, is the huge computing capabilities that low-cost accessibility cloud platforms have for applying predictive models.
  • Machine learning can now take you from largely experience-driven merchandise planning to data-enhanced merchandise planning.

The fashion industry is made up of a succession of collections and capsule collections in continuous development. However, many of the items that make up these collections remain unsold, and cause reduced sales margins, wasted resources, and extra costs incurred by unsold and returns. This has an impact both economically speaking and on the brand’s value and sustainability. 

Added to these intrinsic features is the fact that Fashion is one of the industries to have most felt the effects firstly of the Hong Kong protests, then the COVID-19 pandemic, and later the Black Lives Matter protests in America.

The attention to customer needs and the resulting success of a collection has never been as important as it is today. All this complexity makes the role of merchandise planners become even more critical and with a direct impact on company margins.

 

Traditional Merchandise Planning

Which item should be added to or removed from the catalog? How likely is a SKU (stock keeping unit) to be canceled or returned? Should that item be included in the continuous collection? What price should that SKU be assigned? What is the best time to release that particular SKU?

These are just some of the questions a merchandise planner addresses on a typical work day.  Their strategic value for the success of the collection and of the company itself is indisputable, but what information do you have available to make these critical decisions with?

Using the traditional (top down) approach means starting the analysis from sales targets and turnover projections and defining Mixes, Categories, Classes and sub-Classes, Groups and Assortments.  Over time, this has been replaced with a bottom up approach that not only takes targets, but also historical information, new markets, new product categories, and new pricing strategies into account.

Technological innovation has made it possible to fine-tune these approaches and analyses with data mining tools and specific merchandise planning solutions capable of processing a large amount of historical data that relates to performance and past collections, customer habits and the trends of the moment.

All of this in support of that which is the other big strategic asset after data: the merchandise planner’s professionalism and experience. If this is part of the role on the one hand, on the other hand, the company cannot risk depending on the experience of a small group of people, contended for by the competition, to make such strategic decisions.

 

Predictive Merchandise Planning

The most recent technological innovations have meant machine learning (ML) has begun to be applied to real business cases on a massive scale. In fact, while ML technology has been available for many years, the huge computing capabilities of cloud platforms for applying predictive models offered at low accessibility costs is a more recent phenomenon. Such models can now be applied to improve processes with the aim of making increasingly accurate predictions.

In this context, applying ML to the bulk of data produced in Retail scenarios, especially in the Fashion industry, has a lot of potential to improve company performance. Merchandise planning is one of the most beneficial processes available today.

Starting out by analyzing historical performance data, such as data related to orders, returns and unsold items, and cross-referencing it with the fundamental features of the SKUs (extracted from structured and unstructured sources, such as Bills of Materials, material compositions, text and images) it is possible, for example, to make a prediction about the potential sale of new SKUs that have never been produced before, to rationalize the number of SKU variants during catalog creation, to identify hidden similarities between these items and carryovers, and to predict the categories of items most at risk of order cancellation or return.

 

Machine Learning in the fashion industry: Traditional Merchandise Planning vs. Predictive Merchandise Planning

 

Promoting the adoption of Data Driven Merchandise Planning

Having both the experience and professionalism of merchandise planners and cloud-enabled machine learning on the other makes for two essential assets that need experienced professionals to amplify their combined value.

Professionals who are able to provide the Fashion company with the necessary tools and skills to support the adoption of data-driven merchandise planning through:

  • A natural language processing (NLP) engine designed for the Fashion industry and capable of reconstructing the “abstract” features of products by starting from unstructured data such as texts and images from the catalog and the online portal.
  • A logic for processing and cataloging structured data and not product data, so as to reconstruct similarities between collections. This logic is used as a basis for projections of new product behavior in the future.
  • A machine learning engine to train specialized predictive models for each brand and make predictions about the performance of new products on the market compared to the behavior of past product combinations that were more or less similar in different ways.
  • Interfaces for exporting reworked and enriched product data and predictive data for visualization and analysis with external reporting tools.
  • The use of micro-service and serverless cloud architecture to ensure scalability.

In conclusion, it is important to emphasize that the point of machine learning is not to replace coveted key figures with years of unquestionable experience and professionalism, but to provide them with all the information that can be extracted from the available data in order to help them make highly complex and strategic decisions for the company’s success.

From merchandise planning driven by experience, to one driven by data.

 

Andrea Ardizzoia
Connect on LinkedIn

 


This article is part of the series #TechedgeTalks: views and thoughts from our Industry Leaders on the main market sectors. Don't miss the next one - subscribe to our newsletter!

Other episodes of the #TechedgeTalks series:

New call-to-action

Subscribe!