At the beginning of 2019, Fashion Data started the development of a new technological platform, Jarvis, combining Big Data and AI to support the data-driven transformation of the Fashion Cube brands (Pimkie, Jules, Orsay, Grain de Malice, Rougegorge and Bizzbee).
Creating a new technical stack from scratch is always risky and time consuming. But in the end, it was obvious to the whole team because the stakes were so high for the success of our mission.
Notre mission ? To put smart data intelligence to work for fashion retailers to accelerate their transformation towards a new Zero Waste Fashion Business model.
In other words, it consists of designing, implementing and deploying data and AI-powered business solutions & applications that enable operational teams to significantly increase the economic and ecological efficiency of all company processes.
Three examples of data challenges to reveal for a retailer to become eco-efficient:
- To anticipate consumer expectations and accompanying them on an omnichannel and responsible purchasing journey.
- Only make and buy that a brand is able to sell, without wasting or selling off.
- Intelligent management of stocks in each shop, taking into account the local specificities of stores.
These challenges are not only technical, they are also functional (retail, and particularly fashion, is undergoing a major transformation) and organisational, because it is necessary to support the new business gestures and ensure that they are understood, adopted and applied, which are essential conditions for the creation of additional value promised by data.
BUT WHY DESIGN A NEW DATA PLATFORM?
Developing and operating these new data solutions at the service of the business requires systematically crossing and combining all the data produced by each company, which generates a veritable combinatorial explosion that only Big Data and AI can address correctly. However, these technologies, which were not available to our clients, forced us to start by designing and feeding 6 data lakes as quickly as possible before we could create our algorithms and deploy our first business applications.
Anyone who has already set up a data lake knows that the market offers a very large number of solutions and tools for building a Big Data platform and that it isn’t always easy to find your way through all the promises made by the publishers. Experts will forgive this shortcut, but a complete data platform requires:
- An ETL to collect, transform and export data within the company. The most classic are Talend, ODI, Informatica or Datastage
- An object storage space (S3, Bucket GCP) to archive all data, whether structured (files, logs, etc.) or not (images, videos, customer emails, etc.)
- A high-performance database to query and analyse data. These include Big Query, Snowflake or Redshift of AWS
- Computing power to run algorithms in Python for example
- Data visualisation tools such as Tableau Software, Qlik, Looker or Data Studio
- Algorithm design environments such as Dataiku, Databricks or Alteryx
- But also tools to catalogue, profile and correct the quality of data such as Collibra or Zeena.
As this list is already very long, we will not mention the tools for monitoring, supervising security, access, managing RGDP compliance, etc.
In short, as you will have understood, a Big Data platform often resembles a jigsaw with a hundred or so pieces that must be patiently and intelligently combined to have a coherent, functional and controlled whole. In addition, building a Data Lake requires an initial budget of around €150k (licences + setup) and takes between 6 and 8 months to be fully operational. Don’t expect to load data before or even industrialise a first machine learning score.
BUDGET ET TIMING
As these investments in time and money were not compatible with the objectives of Fashion Data, we adopted a radically different approach allowing us to divide the cost of a data lake by 5, the setup time by 10 and the time needed to load, transform, enrich and activate the data of each sign by 3.
To achieve this objective, we have made the following structuring choices:
- Work in Data Ops mode, i.e. in DevOps mode but adapted to the problems of Data
- Make SQL our default development language, both for loading, transforming and enriching data and for analysing it, building algorithmic models, enhancing data quality and supervising the platform.
- Drastically reduce the complexity of integrating technical bricks by using only the services natively available in the Google cloud.
- Ensure that the platform is scalable and very fast, whether it is to develop a flow, test the data, run a transformation or apply a statistical model.
- Event-driven to seamlessly handle both batch processing and real-time data streaming.
- Remove all risks of manipulation of PII by integrating native data protection mechanisms
JARVIS, for Just A Retail Versatile Intelligent System, was born. The synthesis of the many years of experience of Fashion Data’s experts, we had a Big Data and AI platform capable of rapidly industrialising all our data pipelines as well as controlling the quality of our data, our algorithmic models, the construction of a data catalogue and all this at very high speed.
Over the past 18 months, Jarvis has continued to gain new functionalities to further simplify its use, extend the functional richness of processing, add IN/OUT connectors, and strengthen its integration with our customers’ supervision and monitoring tools.
Today, Jarvis runs over 3,000 streams daily and transforms over 100 TB of data everyday. It is the engine at the heart of all the algorithms, business applications and dashboards used daily by our clients.
The challenge taken up at the beginning of 2019 is now a formidable weapon to accelerate the Data Driven transformation of retailers. But we want to go further and make it possible for as many retailers as possible to design their own tailor-made data platform.
To embody this new stage, we decided to give Jarvis a new identity: Tailer. Anagram of Retail and a wink to the world of fashion (My Tailor Is Rich!), this new name reflects our desire to offer the best data platform designed for retailers, natively integrating all the issues of commerce and distribution.
We will continue to evolve Tailer to provide ever more powerful and innovative retail applications and solutions. So this is just the beginning!