Datalab: small labs give rise to big projects

Data is the new Eldorado for business. It's a phenomenon that everyone, from GAFAMs and social networks to SMEs, is now aware of. Data can be a source of innovation, revenue and improved products and services.

Hence the emergence in recent years, within organizations, of structures dedicated to exploiting them. More or less advanced, they have gradually been integrated into the existing organization chart. Several may coexist in parallel, as may the technologies exploiting this data.

How can we get human resources and infrastructures to work together more effectively on this central theme of Data? How can we innovate and implement new applications as quickly as possible?

This is the purpose of a transversal laboratory dedicated to data: the Datalab. Here, we explain what a Datalab is, how it works, and which companies are taking advantage of it.

See all software Data Analytics

What is a Datalab?

Labs are all the rage. In a world where agility has become paramount, they bring flexibility and responsiveness to organizations. And they do so without the need to overhaul hierarchies or organizational charts.

Their aim is to foster innovation, by enabling experimentation.

A lab can be defined as " a structure that encourages the emergence of disruptive ideas, by isolating uncertain projects and carrying them forward without disrupting the existing organization" (Olivier Laborde).

This laboratory concept takes many forms. It's only natural that it should have been transposed to the field of data exploitation.

Today, a company can count on multiple Data professionals. These include, but are not limited to, the Chief Data Officer (CDO), the Data Scientist and Chief Data Scientist, the Data Analyst, the Big Data Architect and Engineer, the Master Data Manager, the Business Intelligence Manager, the Data Miner, the Data Protection Officer, the Machine Learning Engineer...

This enumeration immediately pinpoints the problem. All these functions are not attached to the same department, nor do they necessarily have the same objective or culture. They don't use the same software tools. They are not in constant contact.

What's more, other functions generate, handle, exploit or benefit from data. Marketing, for example.

How do you bring all these skills together, and combine their expertise, in an agile and efficient Data structure? By creating a Datalab.

The creation of a Datalab is also an opportunity to integrate new skills and talents. Some projects integrate a large percentage of new recruits from the outset.

The Datalab will have the advantage of operating like a start-up, or an incubator, within the organization itself. It doesn't replace what already exists, but enhances it in a different context.

Define a preliminary strategy

Even if the initial idea is to encourage flexibility and creativity, setting up a Datalab is best done with certain prerequisites in mind.

Strategically, if possible, priority objectives should be defined. This is primarily the task of management. Even if the Datalab's aim is to encourage intellectual ebullition and the emergence of creative projects, it is possible to direct its efforts. It's best to specify at the outset whether your objective is to diversify your business, improve customer service, collect new data, etc.

However, maximum freedom can also generate greater commitment and disruptive innovation.

Technically, you'll probably have to deal with the complexity inherent in Data Management and Big Data. In a large-scale enterprise, there may be multiple independent "silos" dealing with data, for historical, geographical or technical reasons. As a result, the technologies used to collect, store, process and use data may also be multiple and redundant, and require adaptation.

One of the advantages of a Datalab is that it encourages technical harmonization, perhaps even beforehand. The Datalab can lead to the creation of a Data Lake, if this is not already in place. What's more, since data quality and validation are of paramount importance, this can be an opportunity to verify these prerequisites.

Legally speaking, a great deal of groundwork also needs to be done. But this is always the case when data is to be used on a large scale. The heterogeneous origin of data means that they have not been collected in the same contexts, with the same objectives, and therefore not necessarily with the same initial legal constraints. This is a classic difficulty, not to be underestimated from the outset.

Don't overlook human resources

In addition to these three pillars, there is another difficulty to consider. It's inherent to setting up a cross-functional structure: the human factor.

You're suddenly going to bring together, within a sometimes informal entity, employees who have one thing in common: Data. But they may also be radically different. Nothing predestines them to form a happy band of friends united around a common goal.

In the Datalab, you may find profiles as different as a marketer, a data engineer, a salesperson and a coder specializing in Machine Learning. That's what a lab is all about.

It may therefore be a good idea to work with a consultant specializing in change management beforehand, or as a follow-up.

All the more so as there may be resistance to this change, which is quite logical. Employees may be more inclined to keep their ideas, skills and inputs to themselves, and use them within their own departments. So we need to make sure that the results of Datalab are rewarding for everyone.

A detail, but not the only one: the premises

As you'll see from the examples that follow, most companies that set up a Datalab (and probably other types of laboratory) do so in dedicated ^premises1.

There are several reasons for this:

• It's important that the premises reflect the freedom of the structure. This means open, bright, modular and fun spaces, with a minimum of constraints.
• It may also be important for the premises to be cross-disciplinary. This means that they should not be physically attached to one department (which would then take precedence over the others).
• Ideally, the chosen location can be completely new, designed for this project, even if this is obviously not within the reach of every company.

Drawing inspiration from success stories: 3 examples of Datalabs

According to Les Echos, two-thirds of CAC 40 companies are already equipped with a Datalab, so it's easy and useful to take an interest in them, to copy their model or avoid making the same mistakes.

Here are a few examples from various sectors.

Axa: an insurer at the heart of data

Axa's Data Innovation ^Lab2 was designed around an R&D team of... 4 people. Launched with 15 employees, it now boasts more than 70, to which must be added around 30 external or one-off participants.

Created in 2014, the Lab's objectives included the creation of car insurance contracts whose price would change according to the driver's behavior. It set itself 5 research objectives: fraud, claims management, analysis of driving behavior to reduce premiums for virtuous drivers, connected health, marketing.

Axa signed a contract with Facebook in 2014, at the time of the launch. In this way, the insurer no longer only manages its own data flows, but also external data with Open Data.

The projects that emerge from this laboratory are regarded by Axa as "agile satellites", IT-independent platforms at the outset. Later, if they are sustainable and prove their ROI, they can be integrated into existing systems.

AXA's Datalab project is highly significant, as it includes a "new talent creation" dimension right from the start. In fact, the insurer has allocated a budget of 180 million euros to train future employees. This has led to the opening of a "Digital Strategy and Big Data" chair at HEC Paris and another "Data Science for Insurance Sector" course at Polytechniques.

Finally, to take this innovative approach to its logical conclusion, the Datalab has also created an incubator. The incubator relies on the AXA Strategic Ventures (ASV) investment fund to support promising projects.

A total commitment to the digital transition. The insurer is thus in a position to help new projects emerge, to give them a reality, and to integrate employees trained for these tasks. A must.

SNCF, a Data Company that didn't know it was there

Would you have spontaneously cited the SNCF as an example of a Datalab? Probably not, and yet this creation is perfectly justified.

It was on August 29, 2018, that the heads of the various entities of the SNCF group presented the new stage of the digital strategy: building the company of tomorrow thanks to ^data3.

Indeed, the historic company is brimming with data. It's just a matter of being aware of it and making the most of it. First and foremost, the company's "data" assets include historical data such as timetables, data on interventions on 15,000 trains, 30,000 kilometers of track and 3,000 stations.

But more recently, it also includes the data that customers entrust to us through the services offered in stations, and the on-board experience (3G/4G connectivity, Wifi, etc.). The aim is to put all this data to use in decision-making, company management, performance and safety.

This led to the creation of a Datalab. Here, the aim is not so much to innovate as to provide access to data for all. The lab is a vehicle for sharing data. Potentially, any agent can connect to the DataLab and access datasets . Potential value creation will come from manipulation by agents.

The laboratory has been operational since 2018. By 2019, it had 350 datasets referenced.

It should be noted that Minilabs developed with École des Mines Paristech have already existed since 2010. Some of these Minilabs are also dedicated to the use of Data. One of them, for example, aims to anticipate the impact of climate change on rail ^networks4.

BNP: using Artificial Intelligence to innovate

As well as being 100% dedicated to data, the Datalab can also be a gas pedal for Artificial Intelligence. In particular, for the development of Machine Learning projects.

This is the path chosen by BNP. In the particular context of the banking market, it was the need for confidentiality that led the Group to internalize this structure. In this way, research, data manipulation and the creation of new applications can be carried out in complete security.

Among the projects that have come out of this laboratory is an astonishing translation application called "Translate". Initially, only three Data Scientists were needed to develop the PoC (Proof of Concept). Then, a reinforced Machine Learning team helped finalize it. This new tool, designed for professional documentation (contracts, reports, technical documents, etc.), quickly became an internal success.

At least a dozen other projects have already been launched. These include an automatic contract analysis system, a search engine, a chatbot, an emotional analysis tool, image analysis and character recognition.

BNP makes no secret of its ambition, through this Datalab, to acquire and develop extensive experience in AI-related fields. This would enable it to position itself as a credible player in this sector in the future.

See all software Data Analytics

💡 Key facts

• A Datalab is a structure dedicated to innovation around Data.
• Its purpose is to federate a wide range of human resources and infrastructures.
• It can be a transversal structure, or give rise to a more integrated creation in the organizational chart.
• It has been adopted by a large number of CAC 40 companies.
• It can be extended to related fields, such as Artificial Intelligence.