《Wavestone ：2025數據與AI雷達：掌握數據與人工智能轉型的10大挑戰研究報告(英文版）（30頁）.pdf》由會員分享，可在線閱讀，更多相關《Wavestone ：2025數據與AI雷達：掌握數據與人工智能轉型的10大挑戰研究報告(英文版）（30頁）.pdf（30頁珍藏版）》請在三個皮匠報告上搜索。

1、10 challengesto master your Data&AI transformation in 20252025 DATA&AI RADAR2Summary.and whats next?Preparing for the future by putting people at the heart of transformation.and whats next?Preparing for the future by putting people at the heart of transformation27CHAPTER 1The rise of AI:time to scal

2、e up and industrialize governance1.Industrializing AI 2.GenAI at scale:moving beyond PoC to realize the opportunities offered by generative AI3.AI governance:increased complexity to address all the underlying issues4.Data/AI acculturation on a large scale,to accelerate innovation and prepare for the

3、 future479512CHAPTER 2Unlock the full potential of Data5.A federated organization to unify Data-related roles,standards,and practices6.Democratize the use of Data for as many people as possible7.Data storytelling,or the art of making Data speak for itself13141720CHAPTER 3Data governance and quality

4、remain key concerns8.Successfully combining Data Quality and Data Observability9.Governance of unstructured Data:a growing problem for organizations10.Systematize the measurement of the value generated by Data212224253Data&AI:what can we expect in 2025?The year 2024 has been a challenging one for Ch

5、ief Data Officers/AI leaders,many of whom now also hold the title of Chief Data&AI Officer,along with their teams.The rise of AI,particularly generative AI,increased interest from executive committees(ExCom)in Data&AI,and evolving regulations have put Chief Data Officers under pressure to tackle mul

6、tiple challenges.Theyve had to address these emerging priorities while excelling in their traditional role:ensuring the company maximizes the value of its data assets to enhance performance,decision-making,and competitiveness.To help you prepare your strategies and roadmaps for 2025(and beyond),Wave

7、stone has analyzed the major trends that are shaping the daily responsibilities of Chief Data Officers and the core challenges they will face in the years ahead.Wavestone supports the CDOs/AI leaders of major companies and public institutions on all geographical regions.This radar reflects the assig

8、nments carried out during the year and the changes observed by all the firms Data&AI experts.We have thus identified the 10 hot topics for the period ahead.4The rise of AI:time to scale up and industrialize governance51.Industrializing AI To take AI projects all the way to production and integrate t

9、hem at the heart of business processes,companies are rethinking their operating model.The stakes are twofold:Define an organization,roles and responsibilities that provide all the skills needed to autonomously deliver a product throughout its lifecycle;Define practices to ensure proper implementatio

10、n of AI,from design to maintenance in operational conditions.To date,the first challenge has mostly been addressed by setting up AI Factory-type organizations.These are specific to each organization,but we can outline a state-of-the-art AI Factory in 3 main functions:The first aims to address the Da

11、ta/AI ecosystems,build relationships with publishers and derive maximum value from the opportunities offered by the market;The second is to deliver use cases,both for experimentation and industrialization;The third aims to steer the Factorys activities,both upstream(managing business demand,driving

12、proposals to address new use cases)and downstream(facilitating adoption,driving change and promoting AI and its potential within the organization).The second challenge lies in delivery practices.An ML model is not enough to deliver value;it needs to be designed for deployment in production.All too o

13、ften,model creation is decoupled from deployment,and designed for experimentation.There is no design authority,no development framework,no automated tasks,no versioning.Moving towards AI industrialization means deploying partially automated model and infrastructure management practices that are desi

14、gned for production use.MLOps aims to address these issues.CHAPTER 1Despite the rise of artificial intelligence in the business world,around 85%of AI projects still fail to reach production.This figure decreases slightly from year to year,but remains relatively high,and only the most mature organiza

15、tions manage to bring it down drastically.6Following on from DevOps,MLOps is a set of practices aimed at unifying development activities(Dev)and operations(Ops),to continuously deliver and manage AI models,from development to monitoring,via acceptance and deployment.While certain teams are now incre

16、asingly proficient in initial model production,the next challenges to be addressed will be advanced,automatic performance monitoring,drift detection and orchestration of continuous retraining to maintain model relevance.The challenge for Data Scientists is no longer to confine themselves to model de

17、velopment,but to acquire ML Engineering skills to ensure the implementation of these practices.And for organizations,the challenge is to define a coherent,shared framework across the organization,based of course on best practices shared by the market,and to drive change to ensure its application.Ill

18、ustration gnrique dune AI Factory ltat de lArt en 2025MLOps:the key to managing AI models in productionCHAPTER 172.GenAI at scale:moving beyond PoC to realize the opportunities offered by generative AI2024 saw the rise of GenAI.Our study showed in May 2024 that 74%of organizations had already begun

19、work on implementing generative AI.PoCs have been launched since 2023 and have led to convincing results.Nevertheless,we note that many companies are still stuck in the PoC phase and have not yet been able to move on to the industrialization stage.Some projects are still PoCs more than a year after

20、their launch!“Generative AI has made decision-makers PoC-phobic.To avoid this situation,we recommend:Define simple rules for all PoCs launched within an organization.For example,PoCs must be limited in time and budget,undergo an initial arbitration at the end of the first milestone,and can potential

21、ly be extended once if the initial results are convincing.If the expected results take a long time to be demonstrated,its likely that the project will have to be terminated,and the focus shifted to other use cases;Manage a single portfolio of GenAI PoCs(even if delivery is decentralized),and collect

22、ively arbitrate which PoCs deserve to go from delivery to scale(demonstrated value,optimal return on investment,etc.).These must be limited in number,and a collective effort must be made to deliver these projects as effectively as possible.A technological strategy to be carefully thought outIts temp

23、ting to want to move fast to take advantage of GenAI and gain a competitive edge quickly.Some companies have entrusted data scientists with the keys to the house,in order to promote innovation and time-to-value.However,this strategy can be fraught with risks:the fact that use cases are delivered loc

24、ally,conceived as a single unit,with relatively little flexibility and scalability.All this at a time when the entire GenAI ecosystem is evolving very rapidly,and innovations abound(new models,new capabilities,etc.).How,for example,can we ensure that we will be able to make the most of a new model a

25、vailable on the market that offers improved performance,without having to rebuild everything?Worse still,some companies,out of a sense of having fallen behind their competitors,quickly forged partnerships with certain tech giants to reassure decision-makers.These choices have sometimes led to vendor

26、 locking situations,in which an organization finds itself limited in its ability to innovate,explore alternatives and optimize its orientations for specific use cases.In addition to flexibility,the financial equation can also be degraded in the long run.Companies that have thought through their medi

27、um-to long-term strategy have thus taken their desire for robustness and technology diversification into their own hands.For example,many are building model-agnostic GenAI platforms and architectures.Such a platform makes it possible to host,train and monitor a variety of LLMs,enabling the LLM best

28、suited to each situation to be used to derive optimum value.Flexibility is guaranteed,as is adaptability and the ability to accommodate a freshly published model.CHAPTER 18An AI Agent is a system designed to reason with complex problems,create action plans,and execute these plans using a series of t

29、ools.Unlike traditional computer applications,these agents have advanced reasoning,memorization,and task execution capabilities.These agents can,for example,solve complex problems(e.g.,generate project plans,write code.),perform self-criticism by analyzing their own outputs,use existing tools and IS

30、,and even perform inter-agent collaboration.These Agents are made up of several elements:1.A Core Agent:the central element integrating all processing functionalities.2.A memory module:stores and retrieves information to maintain context and continuity over time.3.A set of tools:external resources a

31、nd APIs that the agent can use to perform specific tasks.4.A planning module:analyzes problems and develops strategies to solve them.Materialization in the form of Agents CHAPTER 193.AI governance:increased complexity to address all the underlying issuesThe rise of AI is also prompting decision-make

32、rs to strengthen governance,by grasping all the issues at stake.These are many:risk management,compliance,sovereignty,ethics,carbon footprint,are just some of the issues to be addressed through this governance.A number of issues need to be tackled at the same time,in order to establish a sustainable

33、 approach to implementing artificial intelligence.An organization adapted to address AI-related challenges holisticallyOrganizations have begun to set up an organization and roles to govern AI-related topics.All of them are facing a challenge:managing to strike the right balance between having a glo

34、bal,centrally-controlled view,while not bridling local initiatives and innovative business line initiatives.The key concern is therefore to manage decentralization of the IA initiative portfolio.To achieve this,we need to:Define clear roles and responsibilities between central and local levels;For c

35、ross-functional functions(CDO,DPO,AI Factory,etc.),define each persons responsibilities and specify boundaries and interactions.Implement multi-level governance:A strategic body,involving members of the executive committee,to apprehend all the subjects inherent in AI,particularly the impacts on HR,p

36、artners,customers.;An operational body,to manage the portfolio,highlight local initiatives and encourage the dissemination of successful initiatives throughout the organization.This body enables us to maintain an exhaustive view of the portfolio,so as to better manage risks and ensure compliance wit

37、h the AI Act(see below).A growing need for trusted AITrusted AI refers to artificial intelligence designed and deployed in such a way as to guarantee high levels of transparency,safety,fairness,and respect for human rights and ethical values.This implies that AI is developed in compliance with rigor

38、ous standards to avoid uncontrolled bias,protect user privacy,and ensure robustness in the face of errors or cyber-attacks.Trusted AI is also explainable,enabling users to understand how and why it makes certain decisions.Finally,it includes responsible governance,where designers and operators assum

39、e responsibility for its impacts,while integrating control and supervision mechanisms to prevent abuse or malicious use.Setting up trusted AIs requires the involvement of a variety of players:Ethics and CSR teams,to define an AI policy that embodies clear principles and is aligned with the organizat

40、ions values;Engineers developing the models,in order to provide explainable and documented modelsHR teams,to help set up training modules so that teams apply these instructions;Businesses and Data scientists,to analyze biases and monitor results;CISO and DPO,to guarantee security,test vulnerabilitie

41、s and carry out controls;Chief Data&AI Officers,to steer the overall approach.CHAPTER 110The AI Act:Getting your organization started on complianceThe AI Act was enacted and came into force in August 2024.The Act aims to ensure that artificial intelligence systems and models marketed within the Euro

42、pean Union are used ethically,safely and in a way that respects EU fundamental rights.The AI Act thus creates regulations applicable to artificial intelligence systems and models being commercialized and marketed.Research activities with no commercial objective are not affected.All suppliers,distrib

43、utors or deployers of AI systems and models,legal entities(companies,foundations,associations,research laboratories,etc.),headquartered in the European Union,or when headquartered outside the European Union,who market their AI system or model in the European Union.The level of regulation and associa

44、ted obligations depend on the level of risk presented by the AI system or model.There are 4 levels of risk,and 4 levels of compliance:AI with unacceptable risk:AI systems and models with unacceptable risk are prohibited and may not be marketed in the European Union or used for export;High-risk AI:hi

45、gh-risk AI systems and models must be CE marked to be marketed;Low-risk AI:low-risk AI systems and models must be subject to information and transparency obligations vis-vis users;Minimal-risk AI:minimal-risk AI systems and models can comply with conduct measures.Special obligations apply to generat

46、ive AI and to the development of general-purpose AI models*(e.g.LLMs),with different regulations depending on whether the basic model is accessible or not,and on other subsidiary criteria(computing power,number of users,etc.).AI will be implemented gradually over the coming years:February 2,2025:Art

47、icle 5 takes effect,banning AI systems with unacceptable risks;August 2,2025:regulations on general-purpose AI models will begin.The EU AI Office,already established,will oversee governance and regulatory procedures.Sanctions for non-compliance will also begin to apply;August 2,2026:general applicat

48、ion of the AI Act begins,with the exception of article 6 paragraph 1 on high-risk systems;August 2,2027:applied rules will be extended to high-risk systems.The European Commission will publish practical guidelines and examples of high-risk AI systems by February 1,2026.How to get started on complian

49、ce?Start with an assessment of the AI systems in place and in the process of being set up.As a reminder,Wavestone has published,with France Digitale and Gide,a practical guide to enable companies to understand and apply European AI law.CHAPTER 111Risks and cybersecurity,all too often forgotten in pr

50、ojectsWith all the buzz around artificial intelligence(AI),organizations are facing unprecedented threats that go to the very heart of these models.New attacks are taking shape,such as poisoning(modifying training data to trick it),oracle(hijacking AIs to make them reveal things they shouldnt),or il

51、lusion(making AIs believe things that are false but invisible to humans).New risk assessment and protection measures need to be put in place.In the short term,the priority is therefore to secure business projects using AI,particularly in the following stages:Classification of AI use cases according

52、to regulatory criteria(refer to the future European AI Act)or the NIST(National Institute of Standards and Technology)AI risk management framework;Definition of the responsibility matrix and governance for validating use cases,taking into account cybersecurity,transparency,privacy,bias and ethics;Im

53、plement specific measures when necessary,either by integrating them directly into the project design,or by implementing new AI security products that are starting to appear.CHAPTER 1124.Data/AI acculturation on a large scale,to accelerate innovation and prepare for the futureOne of the main obstacle

54、s to the adoption of innovations remains human resistance.AI is no exception,and its adoption remains a major challenge,not least because of the fears and misunderstandings it arouses.It is therefore becoming crucial for companies to acculturate their teams to AI,by demystifying this technology and

55、explaining its applications in concrete terms.Demystifying AI:Reassuring and informing about what it can(and cannot)doThe first challenge of acculturation to AI is to dispel the misunderstandings and fears that surround it.For many,AI is still perceived as a mysterious,even threatening technology,ca

56、pable of replacing humans or making uncontrollable decisions.It is therefore essential to demystify AI by explaining what it can do,but also its limitations.For example,AI is extremely powerful when it comes to processing large quantities of data and automating repetitive tasks,but it lacks human aw

57、areness and intuition.Reassuring teams on these points helps prepare them to collaborate with these technologies,rather than fearing them.Daring to explore technical concepts:the challenge of making applications concreteBeyond this demystification,its important to dare to go further in pedagogy by e

58、xplaining to business teams what the technologies underlying AI are in concrete terms.AI managers:dare to explain to your business teams and executives what OCR,LLM and clustering are.By repeating these explanations and taking a step back,youll be able to make the application of AI in business proce

59、sses tangible.Employees will then be able to understand how these technologies can be integrated into their day-to-day work,identify the processes that could be transformed and the pockets of value to be exploited for the company.No Data,no AI!Finally,acculturation must insist that magical AI doesnt

60、 exist,and that it requires training models on quality Data.And to involve business units in Data management.CHAPTER 113Unlock the full potential of Data145.A federated organization to unify Data-related roles,standards,and practicesEveryone now agrees that Data is a key asset for the success of org

61、anizations.Yet its potential often remains under-exploited,due to organizational silos,knowledge,data control,accessibility,and interoperability.Heterogeneous practices also make collaboration between different teams difficult.To meet these challenges,Data governance remains the watchword.An operati

62、onal model covering organization,roles,and responsibilities,as well as operating modes,remains a must-have for all organizations.And beyond theory,thinking about implementation through concrete practices that can be understood by all players is a strong differentiator,where many organizations still

63、confine themselves to description without tangible application of the elements defined.This operational model covers several key themes:1.A new organization,federated around Data domains,in which businesses regain control of their DataFor a long time,IT and Data management have been the mainstays of

64、 Data management.They are still doing so in the least Data-mature organizations.This poses several problems:Business units lack autonomy in controlling their Data,and become relatively dependent on IT or Data departments for access to their Data,delaying the time-to-market of analyses and projects;O

65、n the other hand,IT Departments are sometimes held responsible for the quality of Data over which they have no control in order to develop the processes used to generate and process it.Attempts to remedy the situation often involve reprocessing the stock,without addressing the flow problems at the s

66、ource of data creation.The most mature organizations have taken steps to give the keys to their Data assets back to the business.This means involving business functions in Data management and making them accountable for these activities.For example:Businesses(re)become accountable for Data managemen

67、t within their scope(Data mapping,documentation,quality monitoring and implementation of corrective actions,etc.);The Data Office plays the role of orchestra conductor,defining the framework,tools,policies,practices,and standards.It is also responsible for supporting the business units in their deve

68、lopment,through training and coaching.The challenge is to give the business units back the keys to their Data assets.CHAPTER 2152.A unified role repository to facilitate know-how deduplication,recruitment and career managementIts not uncommon to find organizations that cant tell you how many Data em

69、ployees they have.Datas unique position at the crossroads of business and technology is a major contributing factor.The consequences are manifold:Difficulty in managing the workforce and strategic workforce planning;Difficulty in managing skills and maintaining them at the state of the art for these

70、 profiles,against a backdrop of rapidly evolving technologies and associated know-how;Recruitment and internal mobility can be laborious,with candidates unable to understand the activities behind these profiles and career paths unclear;Data-related operating methods and responsibilities that need to

71、 be finely defined on a perimeter-by-perimeter basis,and whose understanding by the rest of the organization remains opaque.So,beyond a coherent organization,Data/AI roles are tending to be streamlined towards a common,shared repository across the organization.HR teams are fully integrated into this

72、 approach.3.Common standards and practices throughout the organization For this federated organization to function effectively,it is essential to define and unify Data management standards and practices.This includes common rules for data governance,cataloguing procedures,documentation standards,as

73、well as security and compliance policies.In addition,an organization-wide Data catalog is defined,which describes business concepts via a Data glossary,documents Data in the Data dictionary,and records the transformations undergone by key Data(Data lineage).Unified standards ensure Data consistency

74、and quality throughout the organization.They also facilitate system integration and interoperability,making Data more accessible and usable by all.For example,a standardized Data model enables teams from different areas to collaborate more easily,share insights,and multiply the value created.Whats m

75、ore,these standards facilitate Data sharing,not only within the organization,but also via other subsidiaries,and even with third parties and external partners.CHAPTER 216The Financial Data Access(FiDA)regulation proposed by the European Commission in June 2023 aims to create a legal framework for th

76、e access and use of consumer financial Data.It is part of the broader Open Finance strategy,extending the rules already introduced by the Payment Services Directive(PSD2),which only concerned payment accounts.Essentially,FIDA will make it possible to:Greater Data transparency,with clear and transpar

77、ent communication on how customer Data is used and shared between financial institutions;Fine-grained customer consent,giving customers the ability to grant,manage and withdraw consent for data sharing;Enhanced security,by implementing strict security measures for the protection and processing of fi

78、nancial Data;Standardization of user Data and technical interfaces to accelerate Data-sharing capabilities.FiDA compliance will therefore require the implementation of rules for accessing and sharing very specific customer Data,and will contribute to a sometimes forced rise in the maturity of financ

79、ial services players in their Data sharing capabilities.Focus on FiDA,the new European regulation that will make a major contribution to the growing maturity of Data sharing in financial services:CHAPTER 2176.Democratize the use of Data for as many people as possibleThe framework is thus set,with an

80、 organization,roles/responsibilities and operating modes.All that remains is to prepare the Data and make it available to the greatest possible number of people.With the explosion of Data and the acceleration in its use,a new paradigm has emerged over the years to facilitate its access and use.The c

81、oncept of the Data product has thus emerged,and is now being adopted everywhere.A Data product can be generically defined as a product that facilitates an end-goal through the use of Data.The term thus encompasses various items:Technical foundation type products,i.e.technological capabilities that r

82、epresent the foundations for managing and leveraging Data,and whose construction and evolution are carried out in product mode(e.g.,a RAG platform,the Data catalog.);Data products,where the Data itself is offered as a ready-to-consume product(e.g.,customer repository,sales per perimeter,etc.);Analyt

83、ics products(e.g.,BI dashboard,recommendation engine,scoring model,etc.).In particular,and by misuse of language,the Data product is assimilated to the concept of“Data as product.It is in this sense that the Data product is defined through the Data Mesh as defined by Zhamak Dehghani(see below).This

84、Data product has several basic features:Discoverable:Data is sorted by business domain(then by sub-domains,families,business objects,etc.)and stored in a marketplace.Potential consumers are thus able to see what Data is available,read its description,and request access to it if necessary;Self-descri

85、bing:products are documented and users can independently understand what they contain(via product definition,metadata description,Data freshness,scope,etc.);Trustworthy:the quality of this Data is monitored,and metrics are shared so that consumers can judge whether it is in their interest to consume

86、 this Data.In addition,responsibilities are specified:an owner is identified,as are the teams in charge of Data stewardship activities;Easily accessible:mechanisms are in place to easily manage access,and consumption is facilitated(via APIs,for example).The aim of these Data products is to build up

87、a wealth of Data,the value of which no longer needs to be demonstrated,and to factualize this value creation through multiplied use cases.a.Data Products to build up a wealth of Data ready to be exploitedCHAPTER 218Beyond the creation of data products,it is essential to make them easily accessible t

88、o maximize their value.For this,a data marketplace plays a central role.It enables these data products to be centralized,organized and distributed intuitively and securely.By providing a platform where users can search,explore and acquire relevant datasets,the data marketplace promotes transparency,

89、efficiency and optimal use of data.It thus becomes a strategic lever for democratizing access to data and accelerating innovation within companies.The market is very dense in terms of available data marketplaces.These tools are rich in functionality,with the ability to address complex issues.The dow

90、nside of these tools is that they can also be seen as complicated to grasp by the Business Units,who are the stakeholders who need to be convinced of the benefits of this marketplace.In particular,we recommend implementing such a marketplace in product mode,involving a panel of end-users in the proc

91、ess of choosing a solution,designing and implementing the product roadmap.This approach is the cornerstone of the change management process,and will strongly encourage the support of the business units.Data Mesh:A new approach to Data-driven businessb.Self-servicing and the data marketplaceThe Data

92、Mesh concept is emerging as a solution to the challenges posed by centralized data architectures.Traditionally,organizations have tended to aggregate all data in centralized data lakes,but this approach has shown its limitations,particularly in terms of scalability,governance and agility.Data Mesh i

93、s an approach designed to decentralize Data,while accelerating its sharing and consumption by empowering producers and consumers.The Data Mesh proposes a federated organization where each domain or team takes charge of its own data as a product,while respecting common standards to guarantee interope

94、rability and quality.This decentralized approach enables teams to work autonomously,while ensuring overall cohesion thanks to a federated governance framework.Data Mesh is thus based on 4 pillars,the concepts of which have been detailed above:Organizing Data assets into Data domains;Data management

95、as a product;A self-service platform for accessing and managing Data;Federated governance to govern it all.What are the aims of the professions?Reduce time-to-Data by eliminating bottlenecks(everyone can share their Data easily)while making Data more reliable(e.g.by avoiding Data recopies);Move from

96、 a“Data on demand logic to a“Data as a product logic,where Data is designed to be easily consumable.These are the kind of results that will enable us to move towards a true democratization of Data.CHAPTER 219The profile of Citizen Data Scientist is emerging within organizations.Without being a posit

97、ion or even a formal role,it lies at the intersection between business expertise and Data science.Without being a specialist in advanced Data science techniques,this professional has a solid understanding of business issues and accessible analytical tools.Thanks to Data products made accessible via

98、the marketplace,and via platforms that democratize the use of Data science via no/low code(Alteryx,Dataiku,Knime,etc.),the Citizen Data Scientist can manipulate,analyze,and interpret Data to generate insights and make informed decisions.They thus play a key role in democratizing the use of Data,faci

99、litating the independence of business units in their operational Data analysis activities,and ultimately contributing to the adoption of Data-driven practices within organizations.c.Citizen Data Science,or Data science made accessible to business,with fewer technical prerequisitesCHAPTER 2207.Data s

100、torytelling,or the art of making Data speak for itselfMany companies have recently launched ambitious Data training plans,notably by deploying Business Intelligence(BI)tools among their teams.While this training has enabled employees to familiarize themselves with the what-i.e.the technical use of t

101、he tools and access to Data-it has often left out the how:how to translate this Data into usable,impactful information.Today,the aim is to go one step further by training teams in Data storytelling.Its no longer just a question of manipulating Data,but of learning how to tell a relevant and convinci

102、ng story based on insights.This skill is becoming crucial to reinforce teams ability to extract the full strategic value from data,by transforming figures into informed decision-making levers.How to train your teams in Data storytelling:Understanding the basics of storytelling:Before diving into the

103、 Data,its important to train participants in the fundamentals of storytelling.This includes narrative structure(the key message to be emphasized,the analysis and methodology that led to that conclusion,and the elements of analysis),the construction of powerful messages,and how to captivate an audien

104、ce.These skills will help make sense of the Data;Focusing training on the target audience:Data storytelling focuses not only on core Data populations(Data analysts,Data scientists),but also on key functions that are frequently called upon to add value to Data(strategy,marketing,operations.).Training

105、 teams to adapt their discourse to the stakeholders they are addressing(management,customers,etc.)also helps maximize the messages impact;Going beyond the numbers and choosing the right visualization:Teams need to be supported in interpreting Data,extracting relevant insights,and translating them in

106、to concrete recommendations.They also need to be supported beyond that,particularly in choosing the right graphics or the right way to present the Data.This point is often overlooked and tends to water down the message we want to put across.This notion of Data storytelling must therefore become an i

107、ntegral part of the training courses already in place for the target populations.CHAPTER 2Data governance and quality remain key concerns228.Successfully combining Data quality and Data observabilityIf theres one theme thats invariably at the top of the list of Data issues to be addressed within an

108、organization,its Data quality.To hear project teams,Data scientists and Data analysts tell it,all their problems stem from the unreliability of the Data they are supposed to be handling.These Data quality issues remain a major concern,with only 37%of our customers claiming in 2024 that their efforts

109、 to improve Data quality had been successful.Data observability is on the rise for a number of reasons.The proliferation of Data sources,combined with increasingly complex IT infrastructures,makes continuous monitoring of Data flows indispensable.With so many interconnections between different Data

110、bricks,it is difficult to maintain manual,non-automated supervision,as this would risk leaving room for reactive action in the face of potential incidents affecting Data pipelines.Data observability takes a proactive approach to quality monitoring,quickly and accurately detecting anomalies in Data f

111、lows.It aims to provide complete visibility over the health of Data and systems,enabling teams to be the first to know of any problems,their origin,and how to correct them.In contrast to a static approach to Data quality,this discipline focuses on real-time monitoring of flow performance and the sta

112、te of Data throughout its lifecycle,from source systems to downstream analysis.Using these tools,teams can anticipate Data problems,an essential element in ensuring the reliability of information.These tools enable automated monitoring,problem alerting,Data comparison,root cause analysis,and SLA tra

113、cking,giving teams a better understanding of end-to-end data quality.For Data-driven organizations,that leverage their Data to generate insights,build machine learning models,or drive innovation,a Data observability solution is crucial to ensure that Data remains a valuable asset rather than a risk.

114、CHAPTER 323How can we distinguish between Data observability and Data quality?Data observability supports Data quality,but the two concepts address different aspects of Data management.An organization can have very good Data quality without an observability tool,and conversely,implementing this tool

115、 does not guarantee excellence in quality management either.A real-time monitoring tool is essential for rapid,detailed analysis of the performance of Data ingestion flows.Todays observability solutions are numerous and offer increasingly powerful and comprehensive functionalities;However,it is esse

116、ntial to complement this monitoring with a system of automatic,personalized alerts.These alerts will enable immediate detection of incidents,and rapid intervention by operational teams.They need to be parameterized according to different levels of criticality to guarantee an appropriate response to

117、each type of problem;Setting up an observability system requires precise definition of the steps to be followed,as well as roles and responsibilities at each stage of the process.This includes the teams in charge of monitoring indicators,those who will receive alerts in the event of a problem,the ac

118、tions to be taken,and the people to mobilize depending on the seriousness of the situation.It is essential to co-construct this process with all stakeholders to ensure effective implementation and buy-in from all teams.Pure players are entering the market to address this issue(Sifflet,Montecarlo,IBM

119、 Databand,Anomalo,etc.).The traditional players in Data Quality Management(Informatica,Talent,etc.)are also gradually addressing this issue.It should also be remembered that,although this tool will help to address pain points,it alone cannot remedy all Data quality issues.And its up to Chief Data Of

120、ficers to remind all their organizations of the famous adage:“Data quality is everyones business.Tools for Data observabilityCHAPTER 3249.Governance of unstructured Data:a growing problem for organizationsThe rise of AI,and Generative AI in particular,is putting the spotlight back on the governance

121、of unstructured Data,which has always been the poor relation within organizations.Indeed,Data lakes and other EDMs have accumulated increasingly varied Data over time,and RAG use cases require the ability to manipulate reliable,usable documents.On the whole,organizations are still in their infancy i

122、n this area.In particular,2 topics are currently being addressed by Data Management teams:1.Operationalize unstructured Data practicesManaging unstructured data requires specific practices.Unlike structured Data,which takes the form of organized tables,unstructured Data(documents,images,videos,etc.)

123、requires particular attention to metadata,which provides a minimum of standardization for its content.Good metadata management not only makes it possible to classify and retrieve information more efficiently,but also ensures that documents are up to date and that obsolete information does not interf

124、ere with decision-making.For example,it is good practice to systematically date documents,both in the metadata and in the document content,so as to deal with conflicting information and consider the most recent as the authoritative information.2.Using AI to strengthen Data governanceWe know that Dat

125、a governance is key to obtaining high-performance AI algorithms,but on the contrary,AI also helps improve Data governance.They automate Data categorization,cleansing and anomaly detection,reducing manual intervention and the risk of human error.Examples include:The use of OCR capabilities,placed at

126、the front end of operational systems such as EDMs,to recognize,categorize and transform documents into structured Data,which is easier to govern;The use of NLP models(or even LLMs)to extract structured information from unstructured documents.CHAPTER 32510.Systematize the measurement of the value gen

127、erated by DataFinally,once the projects have been launched,its now a matter of factualizing the value they generate.Indeed,very substantial investments in Data and AI have been decided this year by executive committees.This subject is mechanically highly visible to ExCom members,who want to get thei

128、r moneys worth and measure returns on investment(ROI).The subject is nonetheless ambivalent,since Data is still often perceived as a cost center by decision-makers.It is therefore necessary to distinguish between Data and AI use cases,and the enablers that enable these use cases to be implemented.On

129、 uses:Use cases make it easier to factualize-at least for some of them-the return on investment of these initiatives.Financial gain can be achieved in two ways:through increased revenues,or through cost avoidance.We recommend:On direct gains:to be rigorous in drawing up business cases and monitoring

130、 gains,by involving the Finance and Controlling teams in this process so as to be able to transcribe these elements into the P&L;On indirect gains:to define KPIs,financial or otherwise(e.g.Customer NPS)to factualize the added value of a use case,without necessarily having to transcribe this added va

131、lue in accounting terms,as it cannot be translated into a P&L.And to assume that some of them(e.g.BI dashboards)cannot be transcribed into ROI,despite their key importance for business activities.Note:Even for direct gains,some are trivial to transpose into (e.g.cost avoidance due to fraud detection

132、,or FTE gains due to automation).However,others are more complex to measure(e.g.targeted marketing use cases),and require the implementation of AB testing mechanisms,the features of which must be thought through right from the design stage.CHAPTER 326On enablers:Finally,the value of Data and AI enab

133、lers is ever more complicated to factualize.Indeed,its difficult to directly measure the contribution of actions to improve quality or modernize organizational and technological foundations.For example,the implementation of good Data governance improves information processing,sharing,and management,

134、enabling the identification of critical organizational processes,improved decision-making,and overall optimization of operational efficiency.The same goes for the technological foundation:a Data platform on the scale required constitutes the cross-functional asset needed to deliver use cases handlin

135、g increasingly varied,voluminous and complex Data,while ensuring greater speed of execution,deployment or maintenance.The paradox is that these enablers now account for the majority of long-term Data/AI investments,but decision-makers expect their intrinsic value to be factualized by short-term ROI.

136、We recommend that CIOs,CDOs and CTOs:Factualize the value of enablers by proposing other,more qualitative management indicators(adoption KPIs,compliance with recommended usage,data quality targets,etc.);Over the long term,acculturate decision-makers up to COMEX level,and business units,to the need t

137、o invest in enablers;Rapidly capitalize on early successes,so as to convey the value generated through concrete,measurable examples.Note:To show their credentials and win(and keep)their trust,they must also succeed in factualizing the ROIs of the use cases they address,in order to be heard by decisi

138、on-makers.CHAPTER 327.and whats next?Preparing for the future by putting people at the heart of transformation28A shortage of talent in Data and AIOur analyses show that the number of Data/AI jobs posted online increased by 102%in 2024.Faced with this imbalance of supply and demand,which is tending

139、to increase as talent is also attracted by prestigious,high-paying positions abroad,companies are getting their act together on several fronts to attract and retain talent:Contribution to school relations and student training pathways within institutions training future Data and AI talentReskilling

140、and upskilling existing resourcesDefinition of career paths to offer attractive career paths to technical profiles in particular,who are not all destined to occupy managerial positions but to support technical expertise and technological strategyInsert a short explanation of the chart on the rightEr

141、at pellentesque adipiscing commodo elit at.Laoreet non curabitur gravida arcu ac tortor dignissim.The impact of AI on employment:a long-term strategy to think about nowAI will also have an impact on employment in general.Its potential for automation will profoundly transform many sectors of activity

142、.Some professions will be heavily impacted,requiring a re-evaluation of skills and associated tasks.Against this backdrop,it is crucial for companies to adopt a proactive stance,and to start planning their workforce now.In concrete terms,this means:Evaluate the impact of AI on the various businesses

143、,particularly in relation to the first industrial AI projects planned and the associated roadmapsImplement forward-looking employment and skills management,and plan,upstream,the reskilling of impacted employees to new,more sustainable functions that are less impacted by AITo manage change,with the i

144、nvolvement of social bodies and in a spirit of transparencyCDO Radar methodologyThe CDO Radar presents a selection of the leading topics for Data and AI professionals,which they are likely to handle in their business.The visual is organized into dials that delimit key themes:A,B,C,D,E.Each of these

145、themes is divided into three maturity levels:Mature,Trending and Emerging.Mature level topics can and should be mastered by any CDO;Topics categorized as Trending are beginning to be handled operationally;initial feedback can be shared;The Emerging level covers subjects that are still little-known,i

146、n movement or for which there are no obvious solutions.Identifying them enables you to anticipate future developments and prepare for their arrival in your organization.WHATS NEXT29AuthorsChadi HantouchePartner,Wavestonethis link redirects you to this persons linkedin profileGhislain de PierrefeuPar

147、tner,Wavestonethis link redirects you to this persons linkedin profileStphanMirAssociate Partner,Wavestonethis link redirects you to this persons linkedin profileRomain LevanManager,Wavestonethis link redirects you to this persons linkedin profileContributors:Guillaume Le Floch,Sina Movaghar,Ibrahim

148、 Sail.30ABOUT WAVESTONEWavestone is a consulting powerhouse,dedicated to supporting strategic transformations of businesses and organizations in a world that is undergoing unprecedented change,with the ambition to create positive and long-lasting impacts for all its stakeholders.Drawing on more than

149、 5,500 employees in 17 countries across Europe,North America and Asia,the firm offers a 360 portfolio of high-value consulting services,combining seamlessly first-class sector expertise with a wide range of cross-industry capabilities.Wavestone is listed on Euronext Paris and recognized as a Great Place to W