Beyond the Hype: Embracing the Poetry and Plumbing of AI

According to Prashant Kelker, Chief Strategy Officer of ISG, generative AI promises revolution, but companies must grapple with practical complexities and tradeoffs.

Tangled mess of plumbing

It’s been less than two years since Generative AI became publicly known, and for many, it is already the most revolutionary general-purpose technology since the beginning of the Industrial Revolution.  

Generative AI has shown significant improvements in areas that were believed not to be impacted by technology yet, especially around creativity. In recent experiments, software developers increased up to 40% productivity in tasks when using Gen AI, and some brand and marketing agencies have shared a 70% increase in productivity, as at least two examples among hundreds that populate the news every day and that’s just the beginning of the impact.  

However, according to Prashant Kelker, an Outthinker Networks member and Chief Strategy Officer of ISG, companies are still in the phase of testing the use cases, and many organizations are far from understanding the investment, governance, and managerial implications of scaling these technologies. ISG is a global technology research and advisory firm that supports hundreds of companies in making decisions about which technology products and services they should acquire or develop.

“No one's talking about the plumbing of AI. What happens when this actually works?”

I had a long conversation with Kelker about the current stage of technology adoption by companies, what comes next for AI, regulation, business ecosystems, and the intertwined relationship between technology and strategy. Kelker leverages his own experience and the knowledge his company provides to give a more realistic picture of today and the future. 

Below are the main takeaways from our conversation. 

Prashant Kelker Outthinker Networks

Nail and Scale – the Plumbing of AI

Kelker believes that many organizations are taking the right first steps. They are looking for ideas and use cases. They have hundreds of ideas and experiments, which is a good thing. However, it doesn’t mean that all are prepared for what happens if an experiment works. If an experiment succeeds and has the potential to add value at scale in an enterprise, organizations have to consider if they have the required data and information architecture.  

“If they’re considering it, they’re likely not approaching it in a structured manner,” warns Kelker. Artificial intelligence (AI) is ineffective without a solid Information Architecture (IA). Without information architecture, scaling AI is impossible. Without information architecture, you’re in the dark about your data. And without data, there is no AI.   

“When you go from nailing it to scaling it, you have to think of operating at production scale. You have to consider the cost of running something that works at 10,000 times the scale.”

Kelker says that everybody’s talking about the poetry of AI. “No one’s talking about the plumbing of AI. What happens when this actually works? How much does it cost? What’s your monthly data cloud invoice? Should you have your own GPUs?  Is it viable? These are just basic questions around the plumbing of AI which is very complex.” 

Insight

Artificial intelligence (AI) is ineffective without a solid information architecture (IA). Without information architecture, scaling AI is impossible. Without information architecture, you're in the dark about your data. And without data, there is no AI.

Disillusion with Technology

According to Kelker, the implications of scaling AI will meet a tech disillusion faced by many organizations. “10 to 15 years ago, organizations spent millions of dollars on master data management. It did not meet the promises. Next came the data warehouse – which also had trouble delivering on agile analytics. Then came the concept of data lake – millions of dollars later, the vision was still not met. Then came data lakehouse. And now they say – actually what we need is data mesh. Companies are tired of spending on the topic of data – there is tissue rejection,” shares Kelker

As AI and data are interconnected, are companies open to spending millions of dollars? “Technology platforms carry part of the blame,” says a reflective Kelker. Companies don’t want to be in a never-ending investment situation without extracting value. 

“Companies are tired of spending on the topic of data – there is tissue rejection.”

Starting with the Problem or Opportunity

Technology platforms tend to push technology to solve a problem. “It’s the equivalent of buying a hammer and starting to look for the nail,” says Kelker. What comes first, the need to go milliseconds faster or the concept of a real-time database? These questions are not asked. With all companies feeling the pressure of becoming technology companies, they easily fall into many traps. Firms should think more strategically and ask what problem they want to solve or what opportunity they want to explore. 

Outthinker Networks NYC Roundtable

Centralization vs. Edge Polarity

Cloud computing has made highly sophisticated solutions easily accessible to anyone. Accessing a new piece of software requires only a credit card, with values well below the discretionary power of individuals. In many companies, nobody waits for the CIO or the IT department to make a decision. They just do it on their own. “So, there are wildflowers blooming across and it’s getting more and more difficult to bring this all together, says Kelker.  

Should organizations centralize technology, or should they keep it on the edges of the enterprise where it is actually useful, where opportunities are caught, and where innovation can happen? Technology is moving faster, and nobody wants to miss the train. Five years ago, the average number of software used by a company was 54 on average.  Today it is 268 and growing. “Everybody’s building, buying best-of-breed solutions, but the best-of-breed solutions only work if you have data,” explains Kelker.

But data is also not sitting in one place. Engineering data is with the engineering department or an external engineering system. Marketing data is with the marketing department or probably sitting in Salesforce. There are too many silos – not necessarily organizational ones. And if that is not solved AI will never scale. 

“Everybody's building, buying best-of-breed solutions, but the best-of-breed solutions only work if you have data.”

According to Kelker, that’s where the paradox comes in. Centralizing gives you more optimization, less cost, and more efficiency. But you kill the culture of innovation, or “intrapreneurship,” by allowing people to create new ideas and new businesses that sometimes require agility and quickly test software that’s required in everything today. 

 “The only way to solve this is to provide guardrails to nail – so that experiments are allowed – and to scale. Playbooks for the two phases,” reflects Kelker. “Suppose you find that an experiment delivers business value. For example, the experiment delivered 10 percent more car sales in Malaysia. Is that the solution for Hong Kong? Is that the solution for the US and Canada? Is it different? What does scaling mean? Are there nuances? Is there something in common? Does each geography need a separate thing? Does the same thing apply to trucks?  So, scaling it is a very different playbook.”  Defining “nail” and “scale” playbooks requires guardrails, and then everybody is allowed to experiment and have a way of scaling

Insight

The only way to solve this is to provide guardrails to nail – so that experiments are allowed – and to scale. You need playbooks for the two phases.

The Role of Strategy Leaders in Deciding on Technology

Kelker notes that many companies still treat technology with an “old world mentality.” According to him, in the old world, companies defined their strategy and then how would they implement it using technology.  

“The new world is not so simple,” Kelker says. “The new world is about which technology is available, and what does that mean for the company’s strategy?”

In the old world, technology was the solution; in the new world, technology could be the strategy itself. Technology actually influences strategy so much that it is no longer just an enabler. It directly impacts or even shapes strategy.    

“In the old world, companies defined their strategy and then how would they implement it using technology... The new world is about which technology is available, and what does that mean for the company's strategy."

Leverage the technology opportunity within business ecosystems—when a client’s experiences are the outcome of the interaction and dependency among many different players. Companies need to constantly adapt their systems to accommodate the fluid reality of the ecosystem. This means that companies have less control over their own technology implementation choices. 

According to Kelker, “Value chains are no longer sequential – one step after the other. Everything is asynchronous now. What comes first? Is it offer to order or idea to offer? We are beyond going beautifully from left to right.” 

In ecosystems, things are asynchronous and modular. “Chaos happens because you don’t know what comes first. Imagine each part of that value chain is exposed to the rest of the world with an API. Now, joining the dots, you have about 7,000 combinations. A unique combination of joining the dots is a new business model and it explodes,” Kelker explains. 

Kelker gives one example of how data monetization became an opportunity to ignite business ecosystems. As companies join the AI world, they need data, and someone well-positioned can sell data and make very high margins through simple APIs. He shares, “Many can just monetize their data without making the leap to AI. Now technology heavily influences strategy.”

New world vs old world IT

The Old and New World of IT

A major challenge for many Information Technology areas and organizations is the constant need to adapt systems to the new possibilities emerging technologies bring. It is a never-ending quest. When you’re committed to commercial off-the-shelf solutions, and a promising disruption arises. To cope with that, Kelker recommends changing the way organizations think about technology. 

IT used to think of systems in terms of systems of record and systems of engagement. Systems of record were the most commonly known acronyms – ERP, CRM, PLM, APM, ALM, etc., and systems of engagement related to marketing, social media, apps, client experience, etc.  

“It’s no longer enough to think in terms of these two layers because a third layer emerged in relevance between the other two: the systems of intelligence. These encapsulate the use of AI and insights? What systems will help me forecast? What systems will help me create scenarios? What systems will help me predict? What systems will help me judge?” 

Kelker reinforces that systems of record and engagement will still be important since systems of intelligence still need data from these systems. He remarks that even the three-level systems cannot be the best way to look at the systems.

Insight

IT used to think of systems in terms of systems of record and systems of engagement... It’s no longer enough to think in terms of these two layers because a third layer emerged in relevance between the other two: the systems of intelligence.

He suggests the example of what’s happening in retail and CPG, where they categorize into two levels: business solution and platform. “They don’t say ERP. They say this is our sales and after-sales business solution. And inside that sales/after-sales business solution, there might be a little proprietary code, a little AI, a little LLM; 10 percent of it must be brand new, and 20 percent of it is a 30-year-old legacy system. They look at it as an end-to-end solution. Common platforms emerge when they use software like Oracle to implement end to end value chains. Then you say, okay, there’s my business solution team. These are my platform teams. And then there’ll be some tools and data and intelligence.” Structuring companies this way allows businesses to keep one foot in today, in what’s predictable and reliable, and one foot in tomorrow, where we are still adapting to new realities.  

The Limits of SaaS

There is an illusion present with business leaders that because of the cloud and SaaS (software as a Service), technology is much easier, cheaper, and accessible. But that’s not necessarily true. Kelker says, “If you look at the world of AI, many people associate it with generative AI. Gen AI is in the experimentation phase, and most of the solutions are associated with individual productivity gains.” Other applications are cost-efficiency gains. The best hint for what can go wrong in Gen AI is to go and look at where grand old-fashioned AI (GOFAI) was.

Enterprise architecture was only about buying technology. Now, it is about effectively combining platforms such as ERPs and SaaS and/or building custom solutions. 

 “We are entering a world in the application space where many organizations are moving away from SaaS,” Kelker adds. Banking and financial services, for example, have been using AI for a while and have very low SaaS usage. “If you switch off the server, there’s no bank, right? Their experience and intelligence are actually software. Software is their core business”, shares Kelker.

The pendulum swung too much to the side of SaaS and is quickly returning to the middle, closer to proprietary systems. It doesn’t mean SaaS will disappear. “But we have reached the limit of what SaaS can do,” Kelker explains. “Even in the case of LLM, we are seeing companies like Apple betting on on-the-edge AI capabilities. And then you need good enterprise architecture skills to bring this together.”  

One reason is that many organizations don’t realize how expensive it can be to use the cloud. In the case of banks, besides the subscription costs, the exchange of data is massive. Network and egress costs are highly expensive. It’s more viable for banks to build their own edge data centers, not including many potential cybersecurity benefits. 

“There is an illusion present with business leaders that because of the cloud and SaaS (software as a Service), technology is much easier, cheaper, and accessible. But that’s not necessarily true."

Kelker shares a recent experience of a client of ISG, as an example: “Let’s talk about autonomous driving. So when a car goes across a test track, it is six petabytes of data that needs to enter a data center within 20 minutes. Now, what they do is they take that data and extrapolate it 300, 000 times to train the machine learning algorithm for autonomous driving. Forget cloud computing for this. The data egress speeds and costs with cloud are neither feasible nor viable.” 

According to Kelker, excluding the AI hype, the growth of data centers for hyperscalers has stalled, and they are primarily taking the market from each other. ‘Edge’ data centers, on the other side are growing. 

“So, if you start looking at the capital markets, financial services management, those are the places that have been using AI and ML for five or six years. And if you see how they act and how they have structured, you’ll get very good hints on an extrapolation for the rest of the world.” 

Partnering to co-create solutions

Partnering to Co-create Solutions

Companies don’t need to reinvent the wheel all the time. If there is a need for an after-sales solution, companies should find out the champions of developing these solutions and at least be informed, but potentially partner with them. Kelker adds, “The right partner is not only a technology partner, but they understand the domain.”  

Kelker alerts that many companies fall into the trap of doing it themselves instead of partnering with someone who brings large know-how and can co-create with them what will be faster and better. 

When talking about new technologies, the logic is the same. For sophisticated technologies, such as Generative AI for example, it is practically impossible for many organizations to go alone. The only option to get onboard is by partnering or joining ecosystems. But there are the costs associated with that. “For example, a retail company that chooses a model trains it with its own data and generates output. Who owns the IP for that generated output? Is it them? Is it me? What happens if I find out later that the model was trained with copyrighted data. Now who’s liable? Is it me?  Is it the company I bought the model from?” alerts Kelker. 

Insight

If there is a need for an after-sales solution, companies should find out the champions of developing these solutions and at least be informed, but potentially partner with them.

Extrapolating to the Future

Kelker believes that companies should exert more pressure on regulation. According to him, when things are regulated, there is less fear and risk, a lot of capital is attracted, and everything moves faster.  

Lack of regulation benefits only a few. Kelker shares, “Without the right regulation, it will be too much of a big platform play. There will not be enough space for startups to grow. Just look at MLops (Machine Learning Operations) companies. They were all acquired by the big ones, and now it’s not more attractive to join the market since valuations have fallen substantially since it’s difficult to play with the big ones.”

Regarding AI, he recognizes the technology’s transformative potential but is always alert that organizations are still looking to the poetry side without understanding the plumbing side. 

He says that companies need to consider many open questions to understand the implications of AI for the near future, in addition to the technical ones: “Who owns the intellectual property of derived works? Who owns liability? What is the limit of liability? Does the world need a new risk management for AI? Do you need an insurance product as we have for Cyber-risk?” Organizations will have to answer these and many other questions when contracting with partners and ecosystems.  

“It’s exciting and scary at the same time, but no one wants to miss the next chapter,” concludes Kelker. 

About Outthinker Networks

Outthinker Networks brings together two executive peer networks – the Outthinker Strategy Network and the Outthinker Innovators Network – to help senior strategy and innovation leaders solve their most pressing challenges and keep their organizations ahead of the pace of disruption. 

Are you a strategy, innovation, or transformation executive interested in joining one of our networks?

Authors

Claudio Garcia
Claudio GarciaOutthinker Networks President
Read More

Table of Contents