What does it mean to be a data-driven company?
The use cases and interviews at #datalift reveal what it means to be a data-driven company, especially in Europe.
tl;dr - You need to build the company data infrastructure with intent, iterate on the best use cases for production, and empower your teams to build and deploy the products.
The #datalift use cases and interviews - by AI Guild members - represent an interesting sample of companies, for example Airbus, BMW, Delivery Hero, Klarna, MYTOYS, OLX, Société Générale, Siemens, and Uber. Whether corporate or startup, these companies have built up sizeable 'data operations' that shed light on the evolving productionization of data analytics and machine learning.
If you are investing in a data infrastructure, it is your chance for shifting to data-driven development. You achieve large-scale validation of your products and operation - like BMW has done in building a data factory that serves vehicle development, customer products, and company operations.
For your data infrastructure, scalable solutions are critical - like OLX needing to scale image classification and metadata extraction for, on average, >10 million images daily as well as for peak traffic.
Context is everything if the data infrastructure is to serve productionizing the use cases - like for Uber, where they test and valuate at scale for the specific use case, and develop the algorithms accordingly.
If you have use cases for selection, then deployment is what defines the use case - as Siemens Mobility shows, similarities in pre-production modelling do not usually translate to the production phase.
When developing use cases, it is important to know how to balance business imperatives and technical requirements - so when automating supplier payment for Société Générale precision was more important than recall.
A use case for automated decisions in real-time is uniquely valuable, and requires a focus on data and model quality - so Klarna is able to provide automated decision-making at the point-of-sale for the buyer and seller.
For customer-facing use cases, model monitoring is essential - for the Delivery Hero recommendation system this means e.g. adopting the algorithm to different use case scenarios, and protecting against mis-use.
When deploying the use case, the business value may be monitored - for MYTOYS to switch to an adaptive sorting algorithm reduced the bounce rate, increased conversion, and lifted revenue.
If you have data-literate people, you may upskill them at scale - like Airbus has done in training thousands of its engineers and other employees in data analytics and machine learning.
If you are interested in establishing your infrastructure and evaluating the use cases, please find support here: thedatalift.eu/platform.
Also, we are looking to build an ecosystem together with the companies most interested in helping close the gap to deployment. If your company is interested, please get in touch.