《IIA Research：2023分析預測和優先事項（英文版）（10頁）.pdf》由會員分享，可在線閱讀，更多相關《IIA Research：2023分析預測和優先事項（英文版）（10頁）.pdf（10頁珍藏版）》請在三個皮匠報告上搜索。

1、 TOM DAVENPORT,CO-FOUNDER&ADVISOR,IIA BILL FRANKS,CAO,IIA BRIAN SAMPSEL,VP OF ANALYTICS STRATEGY,IIA DECEMBER 2022 2023 Analytics Predictions and Priorities RESEARCH BRIEF RESEARCH&ADVISORY NETWORK COMPLIMENTARY EDITION Copyright 2022 International Institute for Analytics 2 2023 Analytics Prediction

2、s and Priorities is the season for yearly reflection and planning for what is to come.Welcome to the 13th annual Predictions and Priorities from the International Institute for Analytics,where we look at the latest analytics trends and the most pressing analytics challenges currently facing organiza

3、tions.For most corporations and society at large,2022 was a gradual return to normalcy if that word has any merit anymore but was not without its challenges.Workforce trends were unpredictable,from the Great Resignation to Quiet Quitting and substantial layoffs at large,influential technology compan

4、ies.However,the overall job market remains strong.The Federal Funds Effective Rate has gone from near 0 to over 3.8%,causing some pull back in investment and some increased uncertainty.Geopolitical instability disrupted corporate strategies and sent ripple effects across the global supply chain.IIA

5、has distilled hundreds of client conversations over the year and curated opinions from the largest expert network in the field of data and analytics to put forth five predictions and priorities as we head into 2023.For this years research brief,we thought it would be fun to add a bonus IIA Expert pr

6、ediction courtesy of Burtch Works.IIA has always admired Burtch Works knowledge of the D&A industry and were glad to have them join us in bringing out the crystal ball for 2023.We are pleased that our annual Predictions and Priorities research brief and the associated webinar have become among IIAs

7、most popular content every year.This year,were continuing the tradition of augmenting each of our predictions with some specific priorities for leaders and practitioners to focus on as they attempt to address each prediction.As always,each priority provides specific guidance as to how to best prepar

8、e for,and adapt to,its corresponding prediction.PREDICTION DATA GOVERNANCE GOES IN A NEW DIRECTION IN ORDER TO SCALE Data and analytics leaders will try new approaches to data governance to get business functions to take on more responsibility and decentralize the practice.Special attention will be

9、given to improving the ease of data consumption and building governance into platforms,rather than the typical approach of begging for behavior change.By building governance into platforms CDOs are making it impossible to ignore.Such an approach changes the tone from regulation to enablement,which m

10、any users will find more appealing.Despite years of effort and numerous initiatives,many organizations are still struggling with their data governance efforts.With the amount and variety of data continuing to grow,it is getting more difficult to scale the historical ways of doing data governance.For

11、 example,many data governance approaches are still highly centralized with a one-size fits all model;data and analytics governance practices are not maturing as quickly as they are in adjacent areas of analytics and data management;and the efforts still have a greater resemblance to compliance(appea

12、sing regulators)than they do governance(increasing efficiency).While success remains elusive,interest remains high.Everyone wants the benefits of a good data governance program such as the ability to make better decisions,removal of unnecessary data duplication,regulatory compliance,and risk managem

13、ent.A quick check on Google Trends shows that search activity for data governance is on an upward trajectory.We have continued to see a high level of interest in the topic from our clients in the Research and Advisory Network.T 1 Copyright 2022 International Institute for Analytics 3 2023 Analytics

14、Predictions and Priorities Many chief data officers are also wanting to spend less time on data governance initiatives.Even though the establishment of effective data governance programs is still one of the primary responsibilities of the CDO,many are finding it difficult to demonstrate the value of

15、 their position without spending greater amounts of time on value creation initiatives such as analytics and AI.Governance also doesnt have a great connotation outside of the enterprise data office.A few have even begun to adopt other terms or have eliminated the use of the word“governance.”To furth

16、er complicate matters,it is difficult for many to see an obvious connection between untrusted/ungoverned data sets in a warehouse or dashboard and its impact on a business decision or outcome.This lack of visibility between data and outcome helps explain why business leaders seem disinterested in th

17、e work of governance and stewardship.PRIORITY TRY ALTERNATIVE APPROACHES TO DATA GOVERNANCE Lets consider some of the objectives of data governance and some alternate approaches.Privacy-Enhancing Technologies These are techniques designed to allow data scientists and other researchers to gain insigh

18、ts from sensitive data without ever having access to the data itself.Many of us are likely familiar with techniques such as hashing.Rather than show a credit card number,well show a hash of that number.Differential privacy is one of these techniques,ensuring that anyone seeing the results of an anal

19、ysis will make the same inference about private information,even though the information was not included in the data.Availability and Data Observability Data observability has become a hot topic recently.This is borrowing some of the best practices from the DevOps space and applying them to data pip

20、elines.These often involve automation to ensure that data is moving in a timely fashion and getting to where it needs to be on time and provides notifications if those activities do not happen so they can be addressed in near real-time.Accuracy and Data Quality Solution providers are beginning to in

21、vest in automated data quality services.These services help ensure that the data being produced and delivered from the warehouse or data lake is of high quality no duplicates,no missing values,etc.This is a step beyond observability and can help prevent downstream issues in reports or ML models befo

22、re they are discovered by end users.Usability and Data Products The application of a product management approach to data has been gaining steam and shows no sign of abating.This approach encourages all contributors ranging from analytics to data scientists to engineers to collaborate to ensure that

23、what is being produced addresses a vital business need.This approach encourages the team to always think about the end user and to develop a product that is not only going to be adopted by the end user,but also be engaging for them.This may mean delivering insights from a dashboard or delivering mod

24、el results via an API.1 Copyright 2022 International Institute for Analytics 4 2023 Analytics Predictions and Priorities PREDICTION REUSABLE DATA SETS RISE IN POPULARITY Companies will increasingly employ reusable data assets that can be applied to a variety of analytics and AI purposes.The objectiv

25、e of creating these assets is to increase the productivity of analysts and data scientists by reducing the amount of time spent wrangling data.This is also a means of facilitating effective data governance.Some companies,including Charles Schwab,are creating enterprise-level“data marketplaces”for bo

26、th internal and external data assets that can be easily accessed and used across the organization.AT&T is another company that has created many reusable data assets;these can be easily accessed through a semantic search.A user seeking data on customer churn,for example,can easily find data to analyz

27、e for that purpose.AT&T has also created a machine learning feature store with thousands of easily accessed features.Reusable data assets are only one aspect of a broader focus on asset reuse that many organizations are using to improve their analytical productivity and digital capabilities.They are

28、 creating and reusing software objects,APIs,analytics variables,machine learning features,and datasets.Not all data can be reused,but most organizations can take considerable strides in that direction without going overboard.PRIORITY CREATE A PROCESS FOR DEVELOPMENT OF REUSABLE DATA ASSETS Here are

29、a few key considerations when creating a process for reusable data assets:The first step in creating reusable data assets is for organizations to assess their demand for data and decide what types of assets need to be created.This can be done by surveying analysts and data scientists or by examining

30、 existing analytical applications.Then the hard work of data management begins.A companys data management professionals can focus on capturing,storing,cleaning,integrating,and ensuring quality of the datasets to be reused.Once the assets are created,access becomes the primary priority.Organizations

31、need to publish catalogs for the data assets and make them easily accessible.Several vendors offer AI-based catalog development tools that can make this aspect of the process easier and less labor-intensive.2 2 Copyright 2022 International Institute for Analytics 5 2023 Analytics Predictions and Pri

32、orities PREDICTION GENERATIVE AI GROWS EXPONENTIALLY OVER THE NEXT YEAR Generative AI will be one of the fastest growing forms of AI for making creative work more productive and effective.It is software that creates content using complex machine learning models.Its basic approach is to predict the n

33、ext word based on previous word sequences,or the next image based on words describing previous images.Using this form of prediction,generative AI can produce a wide variety of text and image outputs,including blog posts,program code,poetry,video,and artwork.Also called transformers,many leading tech

34、 firms have created a variety of large language and text-to-image models;the greatest difficulty for companies adopting them may be to choose among the many options.Some online communities and open-source providers have also created generative models.Leading examples of the technology include OpenAI

35、s GPT-3,ChatGPT,and DALL-E-2,Googles BERT and LaMDA,and Meta/Facebooks BlenderBot 3.Business applications of generative AI include:Marketing:Blogs,social media posts,web copy,sales emails,ads,advertising and marketing images,and product descriptions;Code generation:Given a description of a“snippet”o

36、r small program function,some generative programs can produce code in a variety of different languages;Conversational AI:Large language-oriented generative models are increasingly being used at the core of conversational AI or chatbots,often with greater levels of understanding of conversation and c

37、ontext awareness than current conversational technologies;Knowledge management:An emerging application of generative models is to employ them as a means of managing,recalling,and synthesizing text-based(or potentially image or video-based)knowledge within an organization.PRIORITY EXPLORE THE POTENTI

38、AL APPLICATIONS AND PITFALLS OF GENERATIVE AI FOR YOUR BUSINESS Given the emerging market of generative AI,lets consider where to begin:Companies wanting to develop their own generative models from scratch should be aware that such models generally require very large amounts of data for training and

39、 large volumes of computing power,typically in the cloud.GPT-3,for example,was initially trained on 45 terabytes of data and employs 175 billion parameters or coefficients to make its predictions.Each training run costs several million dollars in computing expense.Since few companies have the data a

40、nd processing capabilities or budgets to train their own models of this type from scratch,they can customize existing models.Once trained on a large corpus of text or images,a generative model can often be“fine-tuned”for a particular content domain with much less domain-specific data.As few as 100 s

41、pecific examples of domain-specific data can substantially improve the accuracy and relevance of a generative models outputs.3 3 Copyright 2022 International Institute for Analytics 6 2023 Analytics Predictions and Priorities Companies should train their employees on how to make generative AI work e

42、ffectively.In almost every case there is still human activity required,both at the beginning and the end of the process.Humans kick off the process by entering a short textual prompt into a generative model in order to have it create content.Generally speaking,more creative prompts yield more creati

43、ve outputs.Most users of these systems will need to try several different prompts before achieving the desired outcome.After the model has generated some content,it will need to be evaluated and edited carefully by a human editor.These systems raise numerous questions about what constitutes original

44、 and proprietary content.There also will be dramatic and unforeseen implications for content ownership and intellectual property protection.Companies should engage with their internal or external counsel to establish policies that minimize intellectual property risk.PREDICTION SYNTHETIC DATA REACHES

45、 CRITICAL MASS When first hearing the term“synthetic data,”it is easy to dismiss it as a far-fetched concept.However,todays synthetic data capabilities have become a critical component of many real-world data science solutions.This is because synthetic data does not mean randomly generated data.Rath

46、er,synthetic data is carefully engineered to mimic the real data it is intended to enhance.Today,synthetic data can be generated so that it is realistic enough to enable finer tuning of models that will be applied against real data.When it comes to traditional data,such as transactional data,synthet

47、ic data generators can be directed to maintain the statistical properties of a real dataset across important dimensions.That means that analysis of the synthetic data focused on those dimensions will be useful to understand the dynamics of real-world data.This is invaluable when sample sizes are sma

48、ll.It is also invaluable when highly sensitive data,such as medical or financial records,need to be analyzed.By using synthetic data,valid patterns can be found while still protecting the privacy of those whose data was used to generate the synthetic data.In other situations,synthetic data is the on

49、ly feasible way to train models.Autonomous vehicle systems,for example,have used games like Grand Theft Auto for model training purposes for years now.While the roads in the game arent fully realistic,they are realistic enough to help models learn how to handle a lot of common situations.It is also

50、much safer to have the models experience accidents and road hazards in a video game environment than on a real road.Inventive use of synthetic environments to train AI models is continuing to spread and the autonomous vehicle example is no longer isolated.From digital twins to generative AI and more

51、,synthetic data is hitting critical mass and most companies will be making at least some use of it very soon.Expect a big uptick in 2023.4 Copyright 2022 International Institute for Analytics 7 2023 Analytics Predictions and Priorities PRIORITY INCLUDE SYNTHETIC DATA IN YOUR STRATEGIC PLANS The rele

52、vance and prevalence of synthetic data is going to increase because of the value that it can drive.Now is the time to incorporate synthetic data into your organizations data and analytics strategy.If sensitive data is a large part of your organizations focus,then synthetic data should be front and c

53、enter in the strategy.There are not only ethical issues to be concerned about when working with sensitive data,but an increasing number of legal issues as well.On top of that,consumers are becoming more concerned with how their data is being used.Synthetic data can enable the analytical breakthrough

54、s needed while avoiding legal risk and reputational damage.In cases where complex AI models must be built based on images,text,or video,synthetic data can also be very helpful.It would be expensive and time consuming to get 10,000 pictures of a person with every combination of 10 angles,10 skin tone

55、s,10 lighting levels,and 10 hairstyles.A synthetic version of those same 10,000 images can be generated quickly and inexpensively(our prediction around generative AI intersects with this one).While those 10,000 images will be fake,they will be realistic enough to train a model to recognize real face

56、s much better.The prior examples illustrate why synthetic data needs to be in all strategic plans.It can save cost,protect privacy,and increase safety all while making models stronger.PREDICTION ETHICAL LEGISLATION HITS HARD For the past few years,major privacy regulations have been passed across th

57、e globe.From Europes GDPR to Californias CCPA,governments have cracked down on privacy in a big way.The next wave of legislation being drafted across the globe is tied to artificial intelligence generally,and the ethics of AI specifically.This is being driven by the rapid expansion of AI capabilitie

58、s combined with increasing unease amongst the population about the risks and implications of what is being developed and deployed.At a national or regional level,the European Union has drafted an AI Act while in the United States,the White House recently released a draft of a non-binding AI Bill of

59、Rights.There are also multiple laws(in draft form and passed bills)across Americas individual states.There are motions similar to those examples underway in other countries as well.There is no doubt that what are now proposals or drafts will soon enough become legally binding regulations.The focus o

60、f these regulations are centered squarely on implementing ethical frameworks and creating processes to certify and verify that AI and other analytical processes pass ethical muster.While focusing on the ethical application of AI and analytics is welcome,we will have to wait and see if the initial le

61、gislation that passes includes some bad policies,loopholes,or other concerning features.Some adjustments to initial laws will be inevitable.4 5 Copyright 2022 International Institute for Analytics 8 2023 Analytics Predictions and Priorities PRIORITY INCORPORATE TODAYS NEW RULES WHILE ANTICIPATING TO

62、MORROWS Most organizations have already started to focus on keeping their AI efforts ethical.However,with major legislation very close to implementation,there is no avoiding it.Since almost all the new regulations across the globe are still in draft form,there is time for the industry to exert influ

63、ence and validate and improve the final regulations.Organizations with a heavy focus on AI would be well served to get involved actively in understanding and shaping the regulations that are in development.Proactively start building features and functionality that would reasonably be expected to bec

64、ome mandatory.The more processes that continue to be built that will require major retrofitting,the more expensive,time consuming,and distracting the effort will be to comply with the new regulations.There is little downside to ensuring efforts are ethical in advance of regulations forcing the issue

65、,but there is a lot of upside to getting in front of the requirements and starting to account for them today as best you can.Focus on building flexible ethical frameworks that span applications and processes and that also enable your organization to adapt quickly to changes as they come.As with the

66、major privacy regulations of the past decade,there will certainly be some ambiguities and controversial aspects of whatever is passed.This will require updates to laws as well as litigation to clarify how the laws will be interpreted and enforced in practice.GUEST EXPERT PREDICTION:BURTCH WORKS MISS

67、ION-DRIVEN COMPANIES WILL HAVE THE COMPETITIVE EDGE WHEN RECRUITING ANALYTICS PROFESSIONALS In our November 2022 pulse survey we set out to get a deeper understanding of candidate priorities in todays hiring landscape.Questions four and five,respectively,gave us insight into extrinsic and intrinsic

68、factors that candidates value when making career decisions.What is the most important factor when considering a new career opportunity?When evaluating a new position,which elements are most desirable?Focusing on the first bullet,compensation and work-life balance are the most important factors to an

69、alytics professionals when considering a future job opportunity.Job security,ability to work 100%remotely,and generous benefits package all tied at 14%,closely followed by WFH flexibility(going into the office weekly)at 13%.It is important to note that the ability to work 100%remotely and WFH flexib

70、ility(even if it means going into the office weekly/occasionally),both play a huge role in work-life balance overall.If both options are combined,they account for 27%of respondents(surpassing compensation as the top chosen factor).To be competitive with todays candidates,it is vital for organization

71、s to have a long-term plan for flexibility and work-life balance.These have moved from“nice-to-have”perks to“must-have”core requirements for many employees and candidates.5 Copyright 2022 International Institute for Analytics 9 2023 Analytics Predictions and Priorities As weve mentioned in the past,

72、childcare/home responsibilities tend to make remote working/flexibility an especially attractive or important factor for many parents with young or school-aged children.As one might expect,having a clear career growth and development roadmap(25%)is a top priority for candidates assessing a new oppor

73、tunity.Most tellingly,personal alignment with company values or passion about a companys products was the second most popular choice at 24%.Passion for a category/product and company alignment with personal values is pivotal for analytics professionals seeking a new job.Having a mission-driven purpo

74、se and company culture that supports these initiatives is vital for both attracting and retaining talent.With the difficulty in recruiting today,especially within the data science and analytics space,focusing candidates on a mission-driven opportunity rather than on a task-focused job is an effectiv

75、e strategy.It is vital for companies to adequately convey their mission in a compelling manner to avoid difficulties with hiring when the market is competitive as candidates are increasingly prioritizing these factors when making career decisions.Strong team collaboration opportunities and the abili

76、ty to lead and manage a team were also desirable elements to many respondents.Since professionals have been working remotely,team collaboration has become a pain point due to limitations that are present when not being in a shared office Copyright 2022 International Institute for Analytics 10 2023 A

77、nalytics Predictions and Priorities IIANALYTICS.COM Copyright 2022 International Institute for Analytics.Proprietary to subscribers.IIA research is intended for IIA members only and should not be distributed without permission from IIA.All inquiries should be directed to .TOM DAVENPORT Tom Davenport

78、 is a co-founder and advisor of IIA.He is the Presidents Distinguished Professor of IT and Management at Babson College,a research fellow at the MIT Initiative on the Digital Economy,and a Visiting Professor at Oxfords Said Business School.Davenports“Competing on Analytics”article was named by Harva

79、rd Business Review as one of the ten must read articles in HBRs 100-year history.He has recently co-authored three new books called Working with AI(MIT Press),All in on AI(Harvard Business Review Press),and Advanced Introduction to AI in Healthcare(Edward Elgar).BILL FRANKS Bill Franks is the Chief

80、Analytics Officer for IIA.Franks provides perspective on trends in the analytics,AI,and big data space and helps clients understand how IIA can support their efforts to improve analytic performance.Franks is also the Director of the Center for Statistics and Analytical Research within the School of

81、Data Science and Analytics at Kennesaw State University and has authored several books on data and analytics.You can learn more at http:/www.bill-.BRIAN SAMPSEL Brian Sampsel is a vice president at IIA,where he leads the Analytics Leadership Consortium which seeks to partner with analytics leaders t

82、o share their most pressing challenges and successes,and to learn from their peers through structured,moderated discussions.Sampsel seeks to help clients through a consultative and solutions focused mindset,with experience across a variety of industries and functions,as well as a breadth of tools and techniques including business analytics,data science,cognitive services and robotic process automation.Sampsel has a Bachelors degree in Mathematics from Cedarville University and a Masters degree in Statistics from Miami University.