For businesses, the use and analysis of data has become an important driver of their growth. But developing a data project is a complex process that requires certain criteria to be transformed into real success. Companies must create innovative products that are adapted to the uses of their customers in a much more regular and accelerated way.
Today, it is the main vehicle for maintaining or increasing market shares in a constantly changing economic context. In-depth analysis of market data and consumer behaviour makes it possible to anticipate future needs. With the aim of identifying the uses of the future, expected by customers or breaking with the market, a company can make the best operational decisions to define its strategy and position itself in its market segment.
As a result, data analysis becomes more important than ever as an engine of growth. In 2020, corporate investment in data projects is expected to exceed $ 203 billion globally. But at a time when many claim to be data-driven companies, many data projects are still failing. These failures are, for the most part, due to redundant and well-known causes. Following are the recurring pitfalls that it is essential to avoid.
An approach exclusively focused on technology
Many data projects focus on the implementation of technological solutions (Data Lake, installation of Hadoop clusters, use of NoSQL database …) without even worrying about their purpose, i.e .,the needs or uses to be predestined. Investments are, in fact, focused on IT and not on business and generate little business input. Not all of these projects are going to waste as they contribute at least to the increased technological skills of the IT teams, but they bring little value to the company.
Results built on non-industrializable models
Data Scientists have the necessary knowledge of models (predictive analytics, machine learning, etc.) but generally have little experience in terms of development, especially in the industrial environment. The scripts they provide are regularly not very exploitable by IT teams. On the IT side, if the teams master the aspects of industrialization, they encounter some difficulties in understanding the cogs and sequencing of models proposed by Data Scientists, they do not always understand all the constraints.
Most of the data initiatives are thus concluded with unsuitable and inexpensive results, both for the analysis of the data and for the deployment of the treated use cases. Designing industrial and automated methods requires joint learning for both the IT department and the Data Science team. This is a prerequisite for the generalization of effective deployments.
A lack of hindsight, analysis and preparation of the company
Companies setting up data projects often make the assumption that the models and approaches to be implemented will be similar to those they already know with Business Intelligence projects, thinking finally that there is only technology and tools that change. This drives them to keep the same organization and the same design rhythms as those of their historical projects.
Data science is a discipline based on a prospective approach, with stages of exploration and trial and error. It is not possible today to carry a project like we did ten years ago, on the basis of predetermined studies, without phase of discovery and by not jointly doing business, Data Scientists and IT.
Low collaboration due to cultural differences internally
Companies, especially large groups, are now very small. There are few synergies between different departments and all operate according to their own cultures and habits. However, confusion can develop when the business lines, the data science team and the IT department each work at different levels of objectives, without complementarity between them. Companies are still lacking in maturity with regard to new approaches and methods of work, both for the organization of work and for the processing of data. To remove this obstacle to innovation that persists, the different teams must learn to work together more.
An organization not mobilized
Top management, if there are many expectations, is still too little involved in data projects. But the success of the data initiatives also depends on a suitable investment in the management of the company. This organization must be carried by the entire structure – from the management to the operational teams – so that all the actors concerned are aligned on the same objectives. It must be articulated around a community of transversal practices capable of raising the skills of all the teams, from business to IT.
A data project is a complex initiative, requiring both architectures adapted to deal with large amounts of data (structured or unstructured), to use advanced statistical techniques (proven algorithms, integration of the predictive dimension, d learning) and access to multiple sources of information (stored and managed internally or externally, or shared or accessed through Open Data or Open API initiatives).
But the key to success lies elsewhere and lies in the structuring of the approach. Nothing like going through different stages (ideation, framing use-cases, experimentation) – well-defined in time and dictated by specific objectives – to avoid falling into the usual pitfalls brought to light earlier. Carrying out these three steps will validate the business inputs of a project before entering the industrialization phase, with maximum chances of success!