For fashion brands releasing large collections of products on a semi-regular basis, getting all those products to market online quickly is a key objective and allows faster revenue realisation.
However, with long product lines and even more variant options (particularly colours and sizes), brands have a massive task on their hands to transform their data into high quality customer-facing listings online.
All that data is currently manually processed, albeit with the help of multichannel platforms and Product Information Management (PIM) systems.
Why is machine learning the answer?
In the past, technologies have chipped away at the problem by providing interfaces that allow fashion retailers and brands to work across multiple channels and manage stock and orders. Multichannel technologies have stagnated since, however. Now, fashion brands are turning to newer techniques and deploying machine learning to transform their relationship with channels.
To understand how this can be automated and aided by machine learning, we need to split the process into two parts: enriching, and mapping.
Part One: Enrichment
This is the process of taking data from whatever source or sources it’s held in (typically an ERP, spreadsheets or PIM system) and making the product data customer-facing. This involves adding extra information, combining information from different systems and checking that the existing data is correct, relevant and in the right place.
In this phase, machine learning can be used to combine together disparate data sources and map them to a single standard, putting like with like and ensuring that discrepancies are ironed out. This creates a single data set for the team to work on, rather than piecemeal enrichment happening in different places.
Then, as the system learns and responds to data from channels, the enrichment process can be guided by recommendations based on performance data. That can mean suggesting keywords, title optimisations, adding or changing images, and much more.
The result is a massive increase in the speed of enrichment and higher quality listings thanks to the data-driven recommendations.
Part Two: Mapping
In the past, the problem with having a single data set for products was that they often ended up being listed on multiple channels, from websites to social media to Google advertising campaigns.
Those different channels have unique requirements and label the data differently. Previously, that meant either creating a whole new set of fields for each channel, or sending the same data to every channel and accepting the consequential performance penalties.
Part of the challenge is the complexity of mapping out the different product data structures that channels use, as well as constantly maintaining these, and having the knowledge to ensure that data is mapped correctly. Even assuming all of that is taken care of effectively, the time it takes to achieve perfect mapping makes it prohibitive for fashion brands who need to get their products online quickly.
Machine learning algorithms can analyse each channel’s product data structure, and intelligently map out the data to fit each structure. That means that users can both have their single data set and have optimised listings on every channel, without the tedious manual mapping or the performance downgrade from sending identical data to every channel.
How fashion brands can adopt machine learning
Fashion businesses are rushing to hire data science experts and frequently attempt to deliver their own machine learning projects, often with limited success after an initial PR fanfare.
Often this is due to failure to prepare and the lack of a coherent organisational strategy on AI and machine learning. For these in-house style programs to be effective, buy-in from every team is vital to enable access to data from across the organisation.
The alternative is to find purpose-built solutions using machine learning in pre-packaged applications. This allows fashion businesses to harness the power of machine learning and target business areas where it’s most useful without the pain and expense of developing the algorithms themselves.
Ultimately the correct approach will vary by company but in the same way that the rise of Software-as-a-Service (SaaS) platforms disrupted in-house hosted applications and services, we expect the trend of AI and machine learning applications to quickly hold a distinct advantage for retailers and brands.