Math+Design to transform a Core Business activity

What we have learned helping Euler Hermes' datalab

By MFG Labs
04/27/2020
What we have learned helping Euler Hermes' datalab

Conference “a datalab at the heart of a business transformation”

Monday March 9, 2020, Marie David, chief data officer at Euler Hermes and Julien Laugel, chief data scientist at MFG Labs were invited by the group of alumni of Arts & Métiers specialized in Consulting (GP24), to present a return of experience.

After a quick introduction of Euler Hermes, world leader in credit insurance, Marie David explains the missions of her datalab and in particular the mandate she obtained to transform the “credit limit request” process:

This process is at the heart of Euler Hermes’ activity and is based on the work of an expert: the referee (or underwriter). He must make a decision in a very short time (of the order of a minute) without being able to rely on relevant data or algorithms to gauge risk, expected profitability and customer history at the same time.

The mission focused on the design and production of a model that optimally responds to customer limit requests.

The design phase focused on getting a good understanding of the following three points:

  • Risk / reward arbitrage on different time scales
  • The challenges of the policyholder (customer satisfaction / dissatisfaction),
  • The current process and, in particular, its governance

The design phase took place in 5 stages:

  1. Field survey
  2. Data mapping
  3. Specification of the problem
  4. Definition of the “Target Operating Model”
  5. Choice of the right modeling approach among three tracks (Rule based + Machine Learning, Machine Learning with Simulation or Optimization Function)

The prototyping phase of the model took place in agile mode, by parallelizing several tasks

  1. Definition of technical architecture and its implementation
  2. Raw data qualification and construction of business objects
  3. Construction and calculation of KPIs for models
  4. Iterations on models

Mission Outcomes:

For low risks, decision automation (Benefits: gain in arbitrator time, gain in client time, optimized response).

For medium risks, implementation of a model recommending a decision, allowing the referee to have the last call and to validate the optimal arbitration proposed by the algorithm. The referee benefits from the power of recommendation algorithm to make better and faster decisions.

For significant risks, setting up an “intelligent” BI via dashboards to assist arbitrators in complex decisions with advanced indicators

Below is an example of a dashboard implemented showing the credit history with a customer. It helps the referee to quickly get an overview of the past activity with the customer.

Key Takeaways:

  • A transformation and value creation oriented datalab requires:
    • Sponsorship at the highest level (ExCom)
    • Demanding but pragmatic and delivery-oriented leadership
    • A strong budget, indexed to the creation of value generated by the projects
  • The contribution of a design team allows:
    • The reverse engineering of expert knowledge
    • The business user to be a stakeholder and initiate changes
    • The establishment of a single mental representation that helps all actors to be aligned
    • The definition of possible interactions between business / algorithm in the decision process
  • Data engineering for a complex project requires:
    • A Top-Down approach to define the right level of data abstraction
    • A Bottom-Up approach to analyze and filter this data, define the right scope and validate our understanding
    • Patience to bridge the gap between raw data and business objects. The complexity of the data should not slow down the delivery of value; the understanding of data and the business is gradually enriched
    • To build an infrastructure and a data pipeline that guarantee data quality, consistency, traceability and a single version of truth for all
  • The subtle art of parallelizing tasks that are usually sequential:
    • In order to quickly obtain results enabling a validation of the approach, it is necessary to perform in parallel data structuring and modeling activities
    • Although essential, agile methods show their limits in such a context because modeling is an experiment with its batch of uncertainties.
  • The challenges and advantages of modeling by optimization function (cost function):
    • Expressing business interests that are difficult to quantify in mathematical terms is a challenge that requires a strong expertise
    • It is impossible to integrate all the business specificities into a single formula, which could be difficult to accept for experts
    • Understanding how to articulate and balance objectives of a different nature (risk, profitability, customer satisfaction, etc.) requires a rigorous experimental approach
    • The optimization approach is easier to understood (than for instance ML models) by the various stakeholders of the project - explainability being a key subject in this decision making process.
    • Such an approach works well with the analysis of examples. It is therefore possible to work hand in hand with the end users on the model’s behavior, and to involve the end users in the project while enriching the business knowledge of the project team at the same time.

Why did you call on MFG Labs?

  • The project is complex and requires a second look to quickly move in the right direction
  • The project requires broad skills in applied mathematics, especially in optimization and modeling, which go beyond the usual skills of current data scientists
  • The datalab did not have all the skills necessary for the design and implementation mission, in particular in process mapping (service design team) or support (experienced profiles in consulting and data science)

About the Speakers:

Marie David, Chief data officer at Euler Hermès (and head of datalab)

After a career in financial services, Marie David led several teams dedicated to Big Data and AI in the banking and insurance sector before joining Euler Hermes as Chief Data Officer, where she developed and industrialized Machine Learning models. She is an engineer graduated from Polytechnique and ENSAE. Marie David is also co-author of the book “Artificial intelligence, the new barbarism”, published by Editions du Rocher in October 2019.

About Euler Hermes, word leader in the Credit Insurance

N ° 1 worldwide in credit insurance. Euler Hermes protect your assets and support your business on a daily basis so that you can seize market opportunities, wherever they are. With over 90 years of experience in collecting and analyzing information, EH knowledge of risks is unmatched. Euler Hermes is a subsidiary of Allianz, rated AA.

Julien Laugel, Chief Data Scientist at MFG Labs

A long journey in various sectors of finance has developed in Julien Laugel’s sense of the “value” of information, the delicate art of making data speak… and transforming them into tangible assets! Chief Data Scientist of MFG Labs, he is now an expert in big data management systems, the interactive exploration of these datasets and in the implementation of machine learning algorithms. He studied actuarial science at the Institute of Statistics of the Pierre & Marie Curie University (ISUP).

About MFG Labs

MFG Labs is a consulting and implementation company expert in data and artificial intelligence. We help our clients to model their strategic or operational challenges, and to solve them thanks to tailor-made solutions using AI and data.

See other news