Generative AI – Preparing for the Long-Term Impact


Ritu Jyoti’s research focuses on the state of enterprise AI efforts and global market trends for the rapidly evolving AI and Machine Learning (ML) innovations and ecosystem. Generative AI is not a fleeting trend but a powerful force that will reshape industries for years to come. This session will review practical applications and guidance for CIOs and CTOs as they take on an overarching leadership role in reshaping their organizations at this pivotal time. It will review business, organization, and technology architecture transformation to help organizations tap into emerging business models and achieve sustainable competitive advantage.

Register Now


[This transcript was auto-generated.]
Good afternoon, everyone. This is Ritu Jyoti, I'm group VP of AI and automation at IDC. It's my pleasure to be here with you all today and talk about this very important topic. You know, we all know the technology is a source of advantage. But what we need to focus on is that how we can actually unlock business success with it. So without further ado, let me get started. So here's the agenda that I'm going to talk about, I'm going to kick off the session that where we are from an industry transition perspective, get you some insights about the voice of the customer, big picture market trends, and then give you some tangible guidance and as to how you can harness the business transformation. So there's no doubt in my mind that we all have entered the era of AI everywhere. You know, our current era was primarily fueled by cloud and mobility, and low cost semiconductors, cloud made compute elastic and plentiful, and mobility made compute ubiquitous, the internet made the cost of distribution of apps and data almost zero. And now, we are kind of, you know, expanding into the era of AI everywhere. As we all know, AI is nothing new. We have been using it forget about the definition, and which was kind of formulated in 1950. We've been using the narrow form of AI for the last decade or so. And with the announcement of generated AI, enhancements and innovations last year, we are seeing an exploding, but broad set of use cases. So what are we hearing, you know, we at IDC, we do tons of surveys and customer interviews, and we see that generated VI is gaining a lot of attention. You know, as part of the future of enterprise resiliency survey, when we conducted a survey in March, about 39% of the respondents worldwide told us that they were doing nothing. And it was quite alarming. But, you know, I speak spend time with the end users and many of them told me that this will change dramatically in the next couple of months. And And absolutely, they were right, because when we conducted the survey again in July, it has shrunk from 39% to 22%. And the number of people percentage of respondents who have actually started early exploration has changed from 37% to 49%. In July, what stood out for me from the survey was also that the large enterprises who have more than 10,000 employees, they're planning to spend an average of 23 million on generative AI over the next year. And IDC is projecting that Jenny, I will add nearly 10 trillion to global GDP over the next 10 years. So as I said, we have been conducting these surveys in a very regular fashion. And then we conducted the survey in March, the predominant response from the respondents were like, knowledge management applications is going to be holding the most promise. And then we ran that again in July, the area that got the biggest jump was conversational applications, this is going to be holding the most promise for most of the organizations. And the answer is very simple, because it's all about how you interact with the machines and conversational AI applications are bridging that gap between human and machine interface and go to completely transform that marketing applications code generation applications, they will continue to get a lot of attention. But what was interesting to notice that between March and July, the marketing applications, they became a little bit less important strategically, then how the conversational applications and the design applications have jumped out. Broadly speaking, when we asked them that Where are what are the business areas regenerative AI will have the most impact in the next 18 months, software development and design and product development and design. They are the top two across all the regions, although there's a little bit of a nuance invest in Europe, and Asia pack, because they are kind of prioritizing and focusing a lot more in, say customer engagement. I think AP is the same as product development and design. But you know, software development in design is a little bit less a priority, and product development and design versus the supply chain in Western Europe. So as I said software development and design is really, really kind of getting a lot of attention. But it's not about just the cogeneration. It's about helping the customers and the software developers make use of these tools to refactor the code to help them with automation with help them with documentation with help them with vulnerability testing. And you know, you know what universally we always kind of talk about that. This is a great tool for the software developers. But the more you think about it, this is actually going to be significant.
In the era of infrastructures code, because it will help the IT ops engineers as well. For example, you have a network engineer who actually wants to organize all the log files by the network name, and delete log files that are more than 30 days old, or 90 days old. And we can see it in English and the code is created behind the scenes generate the automation has a huge amount of potential. And some of the use cases that is getting a lot of attention right now. The first example here that comes to my mind is the automated analysis and summarization. As we all know, generative AI has huge potential in this area. So there's a company called viable that has worked with, say, open AI, TPD, 3.5, and four. And what they have done is that they have actually aggregated the insights from emails or NPS surveys or channel and different kinds of feedback via email, and some unstructured content. And what we're hearing is that the accuracy obtained as part of this exercise is actually leaps and bounds better than the traditional methods. So this is going to be huge game changer for customer operations, product management, customer service, people who are marketing professionals who are trying to get insights and more timely insights about what customers are thinking about that we will have heard about, you know how it can help marketing folks create automated blogs and emails. But what is very interesting is how it can help accelerate the salespeople in doing their job, especially creating automated sales proposals sitting right in the Outlook, or in the team's instance, and getting insights from the different CRM systems. As I mentioned, at the start of the presentation, conversational applications will hold a lot of promise across different organizations, because it's not just going to be used in the customer service. But it can be used for HR applications, it could be used for accounting professionals getting insights from the data. But here on the slide, I'm talking about the example where, you know, search disk, the capabilities or knowledge discovery can actually boost the conversations, whether it is via virtual agents, or through the live agents. While there's so much excitement about generative AI, there's some significant amount of inhibition and fields as well. And this is very common with, you know, standard predictive AI technologies as well, we've been hearing about cost being a big inhibitor, and please don't kind of misinterpret it any form and fashion talent, data. And responsibly AI governance is actually very, very crucial. But having said that, the reason I have put two, three and four, ahead of the number five, you know, challenge here is because these are very new, they are nascent challenges that are being addressed by innovations in technology. And they are very specific to all the advancements into degenerative AI. And there are some solutions being developed. But it's a rapidly changing environment. And we'll talk about it in the next few slides. And from the survey results as well, if you can see data security and privacy tops the list in terms of the concerns of the barriers to adoption of generative AI, what stood out for me is also that, you know, finding a partner is also becoming extremely difficult. And he has some advice on that as to how it's critical for the end users to partner with trusted providers of technology. And that brings him to the point that the most of the end users, they are saying that they will be partnering the most strategic technology partner that they will be actually working with to co create and CO innovate is not just the public cloud provider, but also IT consulting partners who could handhold them because there's a lack of talent there, they really need to get some kind of, you know, jumpstart in terms of what they're trying to achieve, whether it is the ROI of the solution, whether it is the cost study, whether it is ramping up their skill set. So you can see that, you know, there's a lot of interest in North America, the highest word got to the degenerative AI to providers, because there's a lot of experimentation and evaluation of the use cases going on, in addition to kind of working with the public cloud providers. So quickly jumping into where are we in terms of the market trends, if you kind of simplify the entire value chain, how right from, you know, kind of creation of the technology to delivering it to the end users, we see that you know, today, the majority of the value is being captured at the infrastructure layer. But the real value in the long term is not going to be just by the traditional foundation models, but it's going to be optimization of the foundation models for domain specific use cases. And there's a huge gap as to how
to industrialize it, how do you scale it. And for that we see the AI platforms and the AI applications will have the highest amount of opportunity for the new entrants. And in this cycle, we're also seeing that there's a huge amount of tax evolving. If you think about it, like in the past, people were kind of talking about mostly the structured data. And then they are kind of now looking into an generated vi view holds a huge amount of potential for the unstructured data. And for that, you need a ETL process for your unstructured data, like the PDF, the dark, the PowerPoint, and is doing some phenomenal work in actually, how do they eliminate redundancies? How do they bring similarities and consistencies and how the vector embeddings are created. And there's a lot of interest in making use of data frameworks and orchestration frameworks like luncheon in line days, but also the best databases. Apart from the traditional recommender engine use cases, they've been very popularly used for the retrieval augmented generation, and for the build of the generator applications, because they provide a quick and easy way of organizations to make sure that they can crowd the data with their enterprise context. Here's a quick overview of all the NEC new players that did not exist without degenerative AI and the transformer model introduction in 2017. And see from the bottom to the top, you see that, you know, the breadth of model providers and builders, we just captured a few of them some of the most interesting ones are cool hair, anthropic ai 21 Labs, or, you know,, apart from open AI that we all have been talking about, but tap nine and the other players also kind of, you know, providing some interesting solutions. And when you think about the application layer, writing assistant, I was just talking about how marketing players and salespeople can get a huge amount of help. So some interesting players in this market, definitely we all know is Jasper comes to my mind, it comes to my mind. But an interesting player that is going to be great game changing is also the adapt, which is the action based transformers. So there's a lot happening in this space, all the Evolve evolution or the advancements in the technology as to how they can bring in better efficiency, better accuracy, and better scale will be a huge amount of focus. And similarly, from a knowledge discovery perspective, you see all the traditional players coming into this market, but also the existing players, they are kind of not sitting idling, they are doing a lot more innovations. And this is a busy slide, but I wanted to share it with you all that where are we seeing in terms of the software supply trends, and whether it is the hyper scalars like AWS, Microsoft, or Google, or whether it is IBM, so the world or even the enterprise application providers like SAP and Oracle, they all kind of embedding generative AI technologies, either by themselves, or they're working in partnership with the ecosystem and augmenting the solution. big trend that we're seeing here is the evolution of the copilot, or the duets from Google, or CO pilots from Microsoft, which becomes a digital assistant from the end users to expand an interface with the machines in a very, very conversational language. The other area that you're seeing is that people are trying to make sure that they're not reinventing the wheel. And they're partnering with the leading generator via technology providers to provide the best solution and accelerate. For example, Oracle has partnered with co-head SAP has partnered with Microsoft ServiceNow has partnered with Nvidia, you see a lot of cool collaboration between the technology suppliers to accelerate the time to the market. And also, we are seeing a breadth of technology suppliers investment into the emerging and the innovative startups. So as I said at the start of the conversation of the presentation today, that it's all about harnessing the business transformation in the most effective way. And for that, we are kind of talking about three levels of transformation that every organization has to go through. So the very first area of transformation is actually business transformation. So I spend a lot of time talking to the end users, and we do surveys, and then Worsley, people talk about lots of focus on expanding labor productivity. But the real issue is that that is important. But is that going to give you the most competitive advantage in the long term? I speaking to an end user who had mentioned to me that they had identified 1000 use cases for productivity gains. So my recommendation is that focus on what your corporate strategy is align it with your AI strategy, and figure out what you're dealing with and how you can kind of make the maximum ROI. So for example, if you think about this, you're hurting with
Say, my margins issue, then you should focus on process centric AI. If you are dealing with some kind of a disruption in your industry, then you should look for product and service innovation. Similarly, when you're thinking about the business models, my recommendation would be that think about it as what exactly can you change in terms of how you what is your value proposition? Who are your partners that you're going to kind of partner with to create your value? How do you deliver the value and then how you capture value. So whether you're going to change your cost structure, how you're going to kind of realize your revenue, whether you're going to use an aggregation business model, you could be or the affiliation business model, are you going to combine the platform business model with the aggregation or affiliation, for example, you know,
Microsoft operating 365, copilot can become the next operating system. Because of it plugins and connects with all your solutions. In the case of some specific solutions, like generative AI can be used to generate synthetic data. And organizations can create revenue from that because they can use it to kind of improve the accuracy of their models. And that can help them increase revenue. But they can also sell that data. Generator vi can also aggregate market data to test concepts, ideas and models. For example, Stitch Fix, which uses algorithms to suggest type choices to its customers has experimented with daily to visualize products based on customer experiences, preferences regarding color, fabric and style. And monetizing generative AI extends beyond pricing considerations and organizations position and the ecosystem can guide the commercial partnerships they have to make, for example, with relevant platforms such as document management system to access a broader market, and enhance the value proposition. An interesting example that comes to my mind is that how Mondelez, India tapped with three, three, it's got aI platform, text video, powered by JDI, offering to record an avatar of an Indian actor, and then used it to create personalized ads for local stores. So my point is that always focus on what you could do to tap into the business value and the business equation that you're trying to go after. Here are some very interesting case studies. Carmax actually partnered with open AI, to aggregate all the customer reviews and completely transform customer experience as to how they buy a used car. General Motors is not just limited to using it for creating their product design. But they're also transforming the whole development tool chain. And one of the best interesting example is that how they can actually help the improve the driver experience, suppose you're driving and you had a flat tire by using English conversational language, you can communicate with the system. This is this is like kind of, you know, using across the whole ecosystem. Similarly, I talked about Morgan Stanley, how they actually trained their GPD 3.5 with the adult management data, and have powered their financial advisors to provide accurate guidance to their subscribers. The last example that I want to talk about is that it's not going to be in silos, you know, there's a lot of interest in generated bi. But in many cases, it's the combination of generative AI and predictive AI that is going to give the real value. So be on the lookout of convergence of different types of technologies. And customer proposal is a great example. So generative AI can be used to create the first draft of the proposal. And the predictive AI can be used in tandem to assess what is the feasibility of the probability of that proposal winning. And it can be used to build your AI strategy aligned with that, make sure that you are kind of focused on the right set of usage of AI. And I kind of learned about different business models that you can explore. But the most important thing that I want to kind of stress here is that you need to have a framework for prioritization of the use cases. So without further ado, let me introduce you to that this is going to be iterative approach, you know, the two dimensions that I'm looking here to start the x axis is the value chain. And the y axis is the complexity. So the number one thing that you need to see started, there are some very high risk use cases, please stay away from that. If a company has decided that they're not going to use facial recognition, use cases, please stop that and not spend your time on that. It's important for you to be agile, fail fast, but look into the value and the complexity. So even if in this chart, you see here, that there might be a use case that has significant value, which is high there, but it is also high in complexity, then you might want to kind of see if there are some data algorithms process systems and know how that kind of helps you cluster those use cases. So that is the clustering approach that I'm talking about. That might be helpful for you to kind of make the
decision whether you want to bundle that, you know use case which has high complexity with the others, because the complexity relatively becomes much, much less. So again, this is an iterative approach go through this process, but don't spend hundreds and millions of dollars. And then kind of learn that, you know, this is not the right investment. Remember those data warehousing days, where we went on analysis paralysis, focus on being agile, focus on right prioritization of use cases focus on failing fast, doing it in bits and pieces, and then kind of learning your lessons and cutting correcting your course back, because some horizontal domain use case ideas, I'm not going to go through that. But the point that I'm trying to say that every organization is going to have a different set of use cases, you need to kind of identify your use cases, and then prioritize it with the framework that are recommended, or some variation of the framework that you that works out for you. I do want to stress that this is not just about the horizontal domain use cases, there's a significant amount of opportunities across the industries here and I'm trying to highlight a few of them like in the healthcare, financial services, manufacturing, generative design has been used in the industry in the manufacturing sector for a long time.
General Motors made use of it to actually do the design of their tire pressure, and reduce the weight of the vehicle. Financial Services, I talked about the Morgan Stanley example, in healthcare, this is going to completely transform the electronic healthcare record record and the workflow systems and important use cases drug discovery, but the area where it will be the biggest transformation will happen this personalized education. And the other industries that I'm not talking about here today is media and entertainment, agriculture has to be on the lookout for some very interesting use cases for that. So what is the guidance, you know, basically, in order for you to actually tap into the successful unlocking of the business transformation, assemble agility device steering committee, so that you can make well rounded business decisions. For the organizational transformation, be proactive in your change management, think about, you know, creating a cross functional platform team, which will include both LB and IT team. And it will be responsible for approving the generated via models, and platform services, staff with the right skills, it will consist of the technical team as well as the business team. And interesting anecdote here is that a lot of work organizations that I speak to, they're hiring prompt engineers, but they're also kind of looking into can they be upskilled can their existing software developers and software engineers, we have skilled, I cannot stress enough that how responsible AI governance is starts from the top, the C suite has very important critical roles and responsibilities. For without that it can lead to unintended negative consequences. And, for example, the CEO of an organization needs to have very, very clear governance charter and an organizational accountability, that, you know, if something goes wrong, what is your remediation steps, and also steps that you can take right from the get go to avoid any kind of unintended negative consequences. And from a CIO CTO perspective, they have an overarching leadership to manage the business reputation, it's not just the responsibility of the marketing people are the legal and compliance people. And they have to be in charge of taking the internal policies and risk management, as well as making sure that they're working with the right partners with the right guardrails and the right set of transparency. So essentially, we are talking about that you have to create a conducive environment for employees to not just survive, but thrive in an AI era. And the last area is the how do you you know, most of you have an existing technology architecture? And how do you kind of, you know, plug into your existing technology architecture to make the best use of it. So here's a framework that you could use to kind of decide whether you're training and tuning. In the interest of time, I'm going to kind of go through it as fast as I can, on the right hand side, unless you have a big strategic advantage, you should not be looking into, you know, training your own model. And if you have a huge strategic advantage and have availability of talent and budget and all then do examine designing a proprietary model, and then you can defer examiner doing it in partnership, in case if you do not have in house talent and budget on the left hand side is that you can improve the contextual accuracy by using fine tuning an existing model. And the two options are there if you have high amount of internal talent and data. And you can imagine using the open source, but be careful that sometimes open source can be very, very tempting to start with and think that it is economical, but it lacks the tooling. You're still looking for the Red Hat for a generative AI open source and in that case, you might be better served by you know, kind of just starting with
Open Source for prototyping and experimenting. And majority of the organizations will actually just buy it from provider, whether it is from Microsoft or SAP or Google or Oracle or any OC or Samba know what any of those companies, and you focus on the service agreement for those. So how do you place bets on the overlapping providers that you're buying software, make sure that you're getting the right set of assistance, prebuilt industry API's robust security, but I cannot emphasize enough on the pricing aspects, make sure that you're able to negotiate the best pricing because the costs are spread very, very fast. From a hardware side, make sure that you are kind of demanding a performance optimization, and full AI stack, tight integration with the cloud and then up system. But also flexibility in your pricing and professional services will be very, very valuable to get you started to get you accelerate your ability to accelerate your time to value and make sure that they have
the ability to kind of provide your ROI models, expand your
industry insights, help your teams kind of ramp up. And also, if they have existing relationships, that your business that would be really, really helpful for you to quickly ramp up. The next area that people are looking into in the short term is that how can they reimagine some of the technology functions, software development, is giving them huge amount of insights they can get, you know, 35 to 45%, faster development of code. But it's not just about development of code, it can be the documentation, it could be the translation, it could be testing. The most interesting example is that of the technical debt, I've been speaking to some customers who told me that they have used it to, you know, first optimize their legacy COBOL or Fortran code, and then kind of translate them into, you know, Python. And that is huge, because it's a rare commodity to have COBOL developers now. And lastly, in the field of IT operations, this is going to be huge in terms of kind of, you know, helping organizations be self healing, self configurable, self reliable, and improve the observability and resolutions responsibly AI governance, we all have been talking about the traditional governance, but generative AI challenges have been some toxicity, hallucination challenges and copyright. And the idea is that how to kind of build an architecture that not only taps into all of these challenges, but brings the enterprise level readiness, right. So if you're using different types of data sources, could you use retrieval augmented generation to provide the right context management and caching? Do you kind of, you know, look into the right, you know, customers need choice and different types of models. So make sure that you have
access to the model hubs, which gives you access to proprietary models, as well as open source models, you can create a prompt library that can be fed into quicker access and don't reinvent the wheel, a lot of work has been done in the ML lab space. But there's some rapid set of innovation happening in terms of LLM ops, in terms of model validation model. You know, orchestration, so look for, you know, tools and techniques that kind of converge and it's not siloed. And lastly, I'd like to say that data architecture will play a very important role. Can you make use of generative AI to create synthetic data to augment your data gaps? Can you make sure that you make use of data architecture for automation of your data pipelines, and bring different types of data sources, but be cognizant that it's the quality of the data that is important, and not really the volume of the data? So lastly, you know, reimagine your technology functions and establish a composable architecture. No company is going to focus just on one technology provider or one particular deployment scenario. So please converge and make the best use of it. So in closing, this is how you're going to prepare for the long term impact, look for business transformation, look for organizational transformation, and reimagine your technology functions to establish composable architecture. So my closing word to you is that start harnessing the power of generative AI. As I said IDC is predicting the journey I will almost contribute to $10 trillion over the next 10 years. So it's your opportunity to act on act responsibly and effectively now. Thank you so much. Enjoy the rest of the day.