Why Nephelai

Our value proposition

Spare the time of your Front and Middle Office teams

  • Reduce the time spent to identify and correct errors in financial transactions.

  • Speed-up reconciliation process and PnL or NAV validation.
  • features
    Reduce your operational risk

  • Get rid of fat finger errors which are a real threat to your reputation

  • Reduce errors on electronic orders which incur direct losses.
  • features
    Comply with regulations

  • Reduce own capital requirements to cover operational risk (BASEL III).

  • Comply with 24/48h deadline to confirm transactions (EMIR / Dodd-Frank act VII).

  • Improve client order handling (MiFID II).
  • In average, each middle office using our solution save
    20% of her time

    Our detection rate on fat fingers is above

    Artificial Intelligence to facilitate trade processing

    Nephelai is an answer to booking errors on orders and financial transactions, which incurs financial losses and overwhelm Middle Office teams. Our solution drastically reduces its consequences by identifying them directly at the moment of capture in the booking tool and while it is being processed by operations.
    Our solution assists three user profiles :

    Front Office

    Our assistance solution facilitate order and transaction capture by traders, allowing to detect booking errors before they propagate through the processing chain and improve the quality of data submitted to the Middle Office.

    Middle Office

    Our solution assist Middle Office at the validation stage thanks to several risk indicator to identify at a glance suspiscious fields. Furthermore, our solution is an additional safety net on electronic transactions.

    Operation manager

    Nephelai produces activity monitoring report for the chief of operation, who can visualize in real time volume of transactions, error rates, alert not yet addressed by the team and anticipate pic of activity.


    Key Benefits

    How it works ?

    Human and Artificial Intelligence

    Nephelai is a human adventure, carried by a team of engineers passionate about finance and computer science. We have been working for three years on the development of statistical learning algorithms specifically for the detection of booking errors in financial transactions. Our solution relies on modern technology components, to warrant security, robustess and high performance while having a limited impact on your own infrastructure.

    How Machine Learning works

    Supervised Learning

    Supervised learning uses historical transactions and the knowledge that some of them were wrongly booked to create statistical models which will is then used to determine whether new transactions are correct or not. This approach is particularly useful to detect frequent booking errors, like when a trader tends to mix up 2 subsidiaries of the same counterparty with similar names.

    Unsupervised Learning

    Unsupervised algorithms learn from historical transactions as well, but this time using only the validated version of them. Transactions are clustered into groups of similar objects. When a new transaction is captured, it is compared to the various groups and if it does not belong to any of them it is classified as an anomaly, which may be a booking error or simply a rare event. This approach allows the detection of fat fingers, like an error on the direction of an FX transaction.

    Tour de force

    The secret sauce

    Out solution is able to perform spectacularily because all the ingredients necessary to a successful machine learning project are there.

    A team with complementary skills

    Machine learning is the art of finding statistical schemas within the data observed in the past to anticpate future evolutions. It requires three set of skills to make it happen successfully: competencies in applied mathematics, in system architecture and deep business knowledge. The first is obvious, the second is key to orchestrate ressource consuming algorithms and the last is key because only the confrontation between the data and the business reality allows to extract its very substance. It is because our team gathers all these competencies that this product is so successful..

    A well defined use case

    Generic algorithms work well on a school use case but suffer once confronted with a real use case. That's why we have been working for several years now on algorithms specifically adapted and created to solve that one problem which result in much better error detection scores than generic algorithms.

    A complex problem, but with users who have their habits

    In addition to a large set of data available for the machine to learn from, Machine Learning requires the existence of statistical schemas, which can be translated as user habits. It happens that markets have their structure and traders have their habits.

    Structured and audited data

    Result precision is greatly improved when data are structured, which is the case of financial transactions. The audit trail of each transaction is necessary to the application of supervised learning algorithms.

    A fine volume of data

    It is obvious that statistical learning is more relevant when the quantity of data available is large. However, what our algorithms discerns are habits of trader and structure of markets. If these are little diverse, a few dozen past transactions are enough to get reliable predictions.

    About Nephelai


    Nephelai is the association of 5 experienced and passionnate engineers around an idea carried by Eleonore de Vial. When she was a consultant in charge of implementing Middle/Back Office solutions, Eleonore had been in contact with many Middle Office team of banks and asset managers. The assessment is simple, the majority of tasks accomplished by Middle Office teams revolve around management of errors, either directly spending time correcting them or upstream, their identification via numerous reconciliation process (confirmation, cash, position and trade reconciliation, valuation control, performance control..). and while each software has its own set of controls and reconciliation functions, which themselves requires time consuming set up, they end up being like butterfly net letting many errors going through unoticed for a long time.
    This experience, associated with the emergence of machine learning in related areas is born the founding idea of Nephelai.


    The founding team of Nephelai is composed of 5 associates with diverse competencies and experiences. We cumulate in particular 35 years in financial software creation, creation of a former successful startup, 3 years study of mathematic applied to the error detection problem and advance knowledge of financial markets.

    de Vial

    Eleonore de Vial

    Chief Executive Officer


  • Initiator of the machine learning based solution to detect booking errors on financial transactions at Misys.
  • 10 years experience at a financial software vendor (Misys) as a Middle/Back Office consultant and then as Product Director in charge of the Middle/Back Office roadmap of Sophis and FusionInvest.
  • Formation

  • M.Sc. Electronic Engineering Queen Mary University (Londres)
  • Engineering Diploma from Institut Supérieur de l'Electronique de Paris.
  • team

    Romain Mangeret

    Chief Finance Officer


  • 10 years experience at a financial software vendor (Misys) as Front Office consultant and then Product Director in charge of the roadmap of the asset management solution (FusionInvest).
  • Formation

  • Ecole Polytechnique, Paris
  • team

    Jeremy Dupre

    Chief Technology Officer


  • 12 years experience at a financial software vendor (Misys) as developer SDK and then Technology Director in charge of all capital market solutions of Misys.
  • Formation

  • Ecole des Ponts, Paris
  • team

    Xavier Guenard

    Chief Science Officer


  • 3 years experience as chief data scientist at Misys on the Machine Learning error detection project.
  • Development of a machine learning solution to detect intrusion in French military systems.
  • Formation

  • M.Sc. Mathematics Versailles University
  • Etudiant Ph.D à Polytechnique, Paris
  • team

    Frederic Minot

    Chief Operation Officer


  • Founder of a start-up IT, 200k users connected at all time, sold in 2016 to an american venture
  • Product Manager at Sagemcom
  • Formation

  • Ecole Polytechnique, Paris
  • Contact

    We would love to hear from you

    A question? we will get back to you as soon as possible.

    • contact@nephel-ai.com