"
INTRODUCTION
The massification of data opens the wayfor a new domain of competitiveness, which simultaneously threatens companiesand offers huge potential for innovation. However, companies whose businessdoes not have new information and communication technologies at its core must undertakemajor transformation schemes, in order to use their data to obtain acompetitive advantage. For many years, and via a network of industrialpartners, the Centre de gestion scientifique (Scientific management center) hasbeen working firstly to clarify the keys to successful transformation throughdata and artificial intelligence (recurring obstacles, impediments, risks,etc.), and secondly to provide methodologies for transformation via AI, andapproaches for integrating and industrializing the predictive model developmentprocess. This paper situates the big data wave in its historical context, andconnects it to the digital transformation. It then seeks to clarify the reasonsfor the difficulties that many companies encounter, before advancing someexplanatory hypotheses which also serve to indicate possible solutions.
Akin Kazakci / Associate Professor, Center for Management Science, Mines ParisTech – PSL
The big datatsunami is flooding through every sector. While the concept (and associatedconcepts such as new AI, data sciences and predictive analytics) continue tocause both rapture and terror, a growing number of experts (and non-experts)are certain in their assertion that no company will be spared by the datarevolution.
True as thismay be, such prophecies do nothing to explain how this transformation will takeplace. To my knowledge, there is no significant research highlighting theconsiderable difficulties that companies are already encountering in theirefforts at digital transformation via AI and data. Let us take, for example, ahot topic which is currently attracting great interest: predictive maintenance.Now let us imagine the situation in which your average maintenance manager mustcope with this upheaval. Normally, their job description includes drawing upthe maintenance schedule, monitoring execution of the plan, and managingintervention and risk management teams. There is no denying that in manyindustries and companies, this is an uncomfortable position, which involvesdealing with challenging emergencies and hazards, inadequate staffing levelsand resources, and work tools (e.g. CMMS) that became obsolete long ago,abandoned and replaced by simple Excel spreadsheets and paper solutions.
Now, thesemanagers find themselves in the front line of the transformation effort. Theirhierarchies involve them in data usage projects. They are called upon by manyconsulting firms and start-ups. Every day, they hear a foreign vocabulary (deep learning…). They feel threatened by the risk oflosing control and frustrated that others do not understand their work, or thatthey themselves do not understand the technologies they will soon be expectedto use. Can we rely on them to conduct this transformation successfully andeffectively?
In truth, inthe two dozen big data projects I have been involved in over recent years, Ihave noticed a very high mortality rate. From industrial maintenance to thesupply chain, from insurance to law, from aeronautics to transport andmobility, the difficulties are the same, and they have nothing to do with datascience. Since I am the only element that all these projects have in common, wemight infer that that I am the cause of these failures. In the followinganalysis, I will try to construct an alternative hypothesis.
What is the source of the problem?
To understandthe nature of the problem, we must first situate the big data wave in itshistorical context.
Firstly, weshould note that this is not the first wave of its kind.
From the1950s and up to the end of the 1980s, a first wave of rationalization based onstatistics and operations research caused similar movements to those we areseeing today. Thus, in “Management in the 80s,” Leavitt and Whisler werealready describing the evolution of management, which was making increasing useof “techniques for processing large amounts of information” using “statisticaland mathematical methods,” which would allow “higher-order thinking throughcomputer programs.” The astonishing thing about their text is that it waswritten in 1959! Already, in the 1950s, we can identify exactly the same dream:better management methods thanks to information allowing improveddecision-making in operations.
Of course, wecould argue that things are different today. Aside from the quantity andavailability of data, and of better systems and algorithms, over time, astrategic dimension has come into play. For example, the certainty that thecompetition will force the hand of directors and compel them to take thesubject seriously. However, none of these aspects indicates a change of nature:the phenomenon remains similar, even if the scale has increased significantly.It would therefore be informative to examine how successfully these companieshave taken advantage of the first wave to boost their competitiveness.
History tellsus that very large research programs were launched in academia and thatbusinesses were quick to set up internal statistics and operations researchdepartments, recruiting the best specialists of their time. However, despiteimpressive scientific and technical progress, over a period of 30 years, itbecame evident that the vast majority of the operations research projectslaunched did not yield the anticipated results. In a text dated 1979 (beforethe management of the 1980s dreamed of by Leavit et Whisler could be accessed),Ackoff announced that “Operations Research is dead even though it has yet to beburied.” The paper, which is among the many works on the crisis of operations researchin the 1980s, cites severalfundamental reasons. The most important is now very visible in today’s big dataprojects: a severe discrepancy between the technical substance of OR(algorithms, technical performance criteria, etc.), and the context ofintegration and organizational usage of these tools. At the time, like today,big data experts very often did not understand the organizational issues (evenif they were recruited by the organization in which they were intervening), andthe companies did not have the necessary knowledge to organize the integrationof the tool in everyday work.
Thisconfrontation very frequently leads to a dialog of the deaf, where the partiesinvolved do not understand each other’s languages, priorities and goals.
A question of renewing businesses
Intraditional industry, data are generated by a business process.
For a companyseeking to define the issues connected with exploiting data, the first point tounderstand is the nature of the link between the value proposition of thiscompany and the data it can generate. In fact, for companies such as GAFA andtheir likes, data lie at the heart of their value propositions: the companyexists because it has been able to develop the skills and technologies requiredto exploit these data. These companies are not so much digitally transformed asdigital-born. In contrast, for companies in the right part – traditionalindustry which was not born of information technologies – the value propositionis more often based on tangible elements (products, infrastructure, etc.). Inconcrete terms, this means that data and data processing are not at the heartof a product or a key skill: at best, they are a derived and not an essentialproduct, which is very often unexploited. They are a by-product which isneither desirable nor undesirable, and which is not taken into account in valueand competitiveness considerations.
This orphannature of the data means that it is impossible for a data scientist to beoperational immediately and independently of the organization: their knowledge andexperience do not include determining the value of these data (i.e. theprediction target and the gains offered by this information). In such aprocess, they will at best play a supporting role, but not a leading one. Toidentify the leader, we simply need to ask: who is the owner of the processthat generates these data?
The answerwill, without exception, point to a business.
Dualitybetween “work tools” and “business processes”
To moveforward while avoiding the traps that have been well-known for more than half acentury, we first need to broaden the restrictive framework offered by the conceptof big data, instead seeing digital transformation as a key concept.
From big datawith the most sophisticated infrastructures and algorithms, to the digitizationof the simplest paper resources, the quest is the same, and has been so for 70years: the renewal and management of information systems, taking into accountthe fundamental duality between “tools” and “work.” Basically, there can be nowork without work tools and the reorganization of work requires work tools tobe redesigned. It is impossible to rethink and transform a business (itsprocesses, its performance indicators, its methods, etc.), without rethinkingand transforming its underlying tools. This well-known duality in managementsciences becomes fundamental when, under the pressure created by big data, thefocus of innovation gradually shifts from the products to the processes.
The businessline is powerless and impaired when it comes to considering its owntransformation via data.
Only thebusiness line could master the intimate duality between a work process and itstools. Only the business line can conceive its own transformation. This is ageneral proposal, going beyond the question of data and AI. However, as we haveseen, the transformative power of formal models of rationality (artificialintelligence, operations research, etc.) and more generally of new informationand communication technologies, has been known to us for several decades. Wemight therefore expect that the business would be used to it, and would alreadyhave the keys to such transformations (methods, approaches, experiences, goodpractices, culture, etc.).
In reality,it is the business line that is powerless. More and more often, it is incapableof understanding the value of its own data, incapable of implementing measureswhich will allow it to design suitable tools and systems to access this value,incapable of managing a client-supplier relationship for the implementation ofthese tools, because it can neither write (nor even conceive) the specificationsfor these tools. In short, the business is behind from the start, because it isso incapable of envisaging the transformation of its work processes and tools.
(Images from internet and copyrights belong to original authors)
Akin Kazakci
Akin Kazakci is an Associate Professor at the Center for Management Science of Mines ParisTech - PSL. He specializes in Innovation and Technology Management, with a dual technical competence in Design Theory and Artificial Intelligence. His early work focuses on logical and mathematical foundations of design theory. More recently, he worked on the generative potential of deep learning models and computer simulations of design reasoning and creativity. Since 2013, in partnership with the industrial world, he has conducted a research program on the transformation of businesses by the AI and methodological approaches for the valuation of the company's data.
Copyright Chuàngkàn Bālí © 2017.
This content is under Creative Commons Attribution 3.0 License. You are free to share, copy, distribute and transmit, under the condition that original source is declared.
Images from internet and copyrights belong to the original authors.
Let’s feel the pulse of the present and future together!
Click “閱讀原文” and feel the pulses from the world.
閱讀更多 創瞰巴黎 的文章
