Viewpoint / CXO of the Future

Where to get Started with Gen AI – WINS Work: Words, Images, Numbers and Sounds

We recently had our latest CXO Insight Call: “Where to get Started with Gen AI – WINS Work: Words, Images, Numbers and Sounds,” hosting both John Sviokla, Co-Founder at GAI Insights and contributor to Harvard Business Review, and Toby Redshaw, CEO at Verus Advisory. They got a chance to lead a lively discussion around how executives can frame up short term AI objectives through a new category of work, more precise and actionable than “knowledge work” – WINS Work.

Today, executives are in the process of considering large, looming issues around AI, such as accuracy, privacy, and bias, as well as the potential impact on “knowledge workers” and even economy-wide job losses and societal risks. While this view is important, it’s difficult to translate to what it means for businesses today. A bottoms-up view is necessary to take in what’s directly ahead and identify the immediate opportunities and threats. Understanding the practical, what’s really happening in the enterprise today (not just the vendor push), is crucial to coming up with a framework of engagement.

Today, it’s starting to look like generative AI is essentially power tools for knowledge work. Natural language has unlocked this entirely new arena for people to explore. There are so many low-code ML capabilities that allow the IT organization to better assist the business. Before, you’d need to build out an enormous data pipeline and infrastructure. You’d need to have highly-paid data scientists who understand statistics and algorithms come in and write all of the code necessary to train these models. And they themselves would need enough knowledge to understand when to choose which particular machine learning technique. So much of that has been abstracted now. You can go to one of the major cloud platforms, sign up for a free account and get access.


So? Where is Gen AI being practically adopted today?

McDonald’s has kicked off “Ask Pickles” – a chatbot that will help its frontline workers get quick answers on questions around maintenance, food preparation and customer service. The bot has been trained on everything from employee training materials, to device specifications, to food preparation, and at over a hundred locations, Pickles is now also accepting drive-through orders. Since the vast majority of orders today come through the drive-through at McDonald’s (70%), this could be a huge productivity boost across their locations.

Walmart announced at CES that they’re kicking off a new gen AI-powered search experience for iOS shoppers. It will enable customers to search by specific use cases while shopping, for example, a football watch party. This way relevant, cross-category results are generated quickly as opposed to the customer having to individually search for chips, wings, drinks and a 90-inch TV.

ChatGPT is coming to vehicles, as announced at CES by a number of car manufacturers. Ask questions, get directions, use natural language instead of scrolling through endless menu screens. Volkswagen will be the first volume manufacturer to offer ChatGPT as a standard feature (kicking off in the second quarter of 2024).

Bloomberg already generated a lot of buzz when they quickly announced that they were building their own language model, BloombergGPT, a 50-billion parameter language model for finance, but that hasn’t stopped them from innovating further. They just implemented AI-powered earnings call summaries on their terminals, which use AI to help analysts with their research process. The new tool enables users to decipher complex financial information and quickly extract key insights on topics addressed by corporate management teams, such as guidance, capital allocation, hiring and labor plans, the macro environment, new products, supply chain issues, and consumer demand.

Microsoft Epic
Epic and Microsoft announced a gen AI collaboration, where they’re expanding to develop and integrate generative AI into healthcare by combining the scale and power of Azure OpenAI Service1 with Epic’s industry-leading electronic health record (EHR) software. The collaboration expands the long-standing partnership, which includes enabling organizations to run Epic environments on the Microsoft Azure cloud platform.

Ideally, with innovations like these, we can reach the point where doctors don’t just have to sit behind the computer and type everything in while talking to people. The hope is for the dialogue to be picked up, recorded, and sent to payments.

In summary, we have these huge elephant companies putting this stuff not in some “Store Eight,” which Walmart just closed down, but into actual mainline capabilities, which is very different.

Early Best Practices

PWC just came out with their global CEO survey, which found that 43% of CEOs believe that generative AI will decrease labor costs in their organizations, over 70% have some AI implemented already, and 45% believe that within ten years or fewer, their current business model will not be viable, that they will have to transform themselves.

So, everyone is dipping their toes in the water, and what companies are starting to get smart about things:

  1. First, it’s crucial to put in place the proper frameworks to enable governance upfront, instead of trying to retrofit all that. Even though this sounds like common sense, it can be very difficult in practice. Outsourcing this piece to legal means you may not like what you get – so it’s important to have a pragmatic process
  2. The models and the tools of today are changing and evolving constantly. There is no one size fits all. Everything depends on the tasks a specific team is after. Change is constant and will continue to be. So part of this foundational work must include frameworks where models can be relatively easily swapped in and out – being as flexible as you can is important, in order to avoid vendor lock-in
  3. Don’t underestimate the importance of prompt engineering. Garbage in, garbage out. It’s much like any other type of query language. Each model has slightly different ways of interpreting the prompt to gain the same output, and learning the nuances matters. For instance, Anthropic’s syntax is not the same as OpenAI’s. So spinning things up like a center of excellence to understand all this can help, but at the end of the day, engineers will need to be educated, as well as any line of business employees who are helping to build low-code applications
  4. Consider whether generative AI is even the right tool for a given use case. Some traditional AI with structured data inside a traditional database can be more effective for certain tasks.
  5. Solving use cases with traditional tools can happen first, with generative AI filling in the gaps
  6. Many companies are dipping their toe into the water today via zero-risk chatbots. Companies quickly learn that the kinds of issues they considered early on are different from the issues that occur in production. It also makes it easy to identify who on your teams is willing to take some personal risk and invest time here – you can quickly find your best people who want to learn how to do this.

What’s Happening Next? WINS Work

The noise around this topic is likely not going away anytime soon, so it’s important for executives to start considering how they want to address it. If you look at the market capitalization of the top six players in AI, as of the end of October, it’s 10.1 trillion. To put that into perspective, that’s greater than the gross domestic product of Japan and Germany combined. So these big players have a ton of cash, and this is critical to them growing their businesses, because it allows them more ongoing revenue and grants them the ability to grab hold of more of the stack. This really puts Microsoft’s investment into OpenAI in perspective. They generate just under 2 billion of EBITDA a week, and if you look at the total EBITDA of the top six AI companies in 2023, it’s 428 billion. To put that into perspective as well, the Russian government collects something like 350 billion in taxes. So I mean, this is not going away. This is central to their strategy. It’s important to keep reminding people of this and get their perspective on it. So if you look at some of these startups who have raised about $600 million to build some models, what are they going to do when they run out of money? Because that’s three days of EBITDA for Microsoft.

So where should your company start with generative AI? Everyone is saying AI will transform knowledge work, but when you think about it, a lot of people technically qualify as knowledge workers. A plumber is a knowledge worker, a heart surgeon, a lawyer, but they’re all not going to be impacted the same way.

So maybe the core concepts to consider are really the manipulation, creation, or improvement of words, images, numbers and sounds – WINS work. So yes, a lawyer would be impacted, but not a heart surgeon. GSIs are another clear example of high impact: most of the work is WINS work and most of the work is digitized. There will be massive changes in the core of those businesses, whether it’s around tax, consulting, or audit.

What we’re going to see is that over the next three to five years, there will be a shift in the relationship between capital, labor and productivity. You’re already seeing it in some offshore firms. Clients are already pushing them to lower costs because they’ve started reducing headcount on account of these new tools, offshore medical coding is a good example of this. Pharma is another winner: they think they may be able to explore ten, a hundred, a thousand, or even ten thousand times as many potential molecules using this new technology.

And then you have industries who are attempting to hold out: for example, media and education. And some industries just don’t have enough digitization: think Cisco – being operationally intensive with thin margins makes it very difficult for them to take advantage of this opportunity.

So we’re definitely starting to see some pretty fundamental edits or transformational sort of themes. It feels as though a lot of what we’re hearing is copilot is being applied to traditional back office work: software development, customer support, HR, marketing, etc. And this is sort of the new table stakes. This is how companies should do HR going forward, but it’s not fundamentally transforming the business, it’s improving productivity. It’s giving a gift to the CFO. In contrast, digitized PWC data could create the consulting of the future organization at a wholly different cost model, wholly different workflow, wholly different customer experience. That’s where the leaders are looking to get to today.

Traditionally what happens is that you’ve got three things going on. First, you’ll have some small population of existing companies adopt this aggressively and force their other competitors to join in. Usually industries aren’t just moving wholesale. Over the next three to five years, the one or two leading companies will garner 60, 70, 80, or even 90% of the profitability available in that industry. That’s the first thing you’re going to see.

The second thing will be the entrance of entirely new business models, for example, a neo law firm that has completely re-imagined their processes, but that’s 3-5 years out, it will take a little while before the market realizes this new model is viable. The most expensive price for a taxi medallion in NYC was five years after Uber was founded, when they were traded for 1.2 to 1.3 million a piece, 12 months later they traded for $34,000. So, it takes a little while for the market to believe it. But once they do, there’s massive financial disruption, not market disruption, because the taxis are still all around New York City, but nobody’s putting money in them. So that’s what happens in industries. When somebody sees the new model come in, when Amazon gets to a certain size, Macy’s gets marked to market way down. Now they stay around as a zombie for a long time and they gobble up other zombies trying to stay alive, but they’re gone and no talent goes there, no capital goes there, no good technology goes there and so forth. So they just keep getting farther behind.

So finally, companies will see the new technology and think, okay, how would I design my company from scratch with this new stuff, and change my entire business model? Any company that has access to data, modern development practices and a courageous mindset can be re-imagined and progressed. If not, others from outside that industry are going to make inroads. We’re going to see better asset utilization, better customer experience, better controls, and cheaper cost structures. There will be companies that get eviscerated and drop out of the S&P 500. From the board on down, there should be some fear that if you’re not moving fast on this, you’re going to be left behind.

The unions in Hollywood were able to get a five-year moratorium using AI to write stuff. But it’s still going to come out of Bollywood, and every single other place. Unions are going to drive themselves into the ground over this issue. What they should be lobbying for is getting the studios to grant them open source access to education, training, etc. They should be going the other way. Ideally, if you’re in a union, you want to be contracted with the companies that are going to win.


This is a situation where people who have absolutely no experience with the technology can have incredibly strong opinions about it, and that’s a really dangerous thing to have to deal with. So if companies want to be in the top quartile, boards need to be educated: What is this new gen AI wave? Assume nothing about their current understanding. Second, VPs need to be personally spending at least three hours using this stuff on something that matters to them. Doesn’t necessarily have to be the business, it could be planning a vacation, doing a personal health routine, whatever. But they need to be able to understand and use these new tools. Finally, you’re going to need a basic employee training policy – give everyone access to the tools, but also ensure that they’ve taken the training.

Training especially helps the business users: they often have no idea what to ask for. Or, what they ask for is unrealistic relative to existing tooling. In the past, they’ve never had the option of having IT applied to their business use cases. So if you’re an IT organization, waiting for people to come and tell you what they want may not be the most effective strategy. Instead, getting good at prototyping may be the answer. Leverage some of the low-code capabilities, pick a generic use case, and show the business how it works and what you’ve learned. Once you win people over, go build something in production that has actual KPIs behind it. That’s when you start to get the gears turning and get people excited about what IT is capable of. That’s when IT and the business can start to open up a partnership around business needs.

Innovation infrastructure may need an overhaul as well: this isn’t just a tech transformation, it’s a top-down leadership transformation. Organizations have lost the muscle to innovate and they need to grow that back. You don’t just need an AI champion, you need an icebreaking ship: someone needs to change company behaviors and force people to do things differently. “We may not know the distinct value yet, but we’re going to find it.” This is hard: mass innovation inside a company comes with a lot of cultural and behavioral problems, but gradual change can succeed. How can you reward people for trying things and screwing up? That sort of behavior normally gets punished at companies with monthly ops reviews on plan vs. action. So if you don’t have a culture that can broadly reward innovation, it’s going to be very, very hard to get there on AI. Many companies will be tempted to avoid building and just focus on “buying,” but it may be that commodity tools lead to commodity results, and the real outlier outcomes will come from building something yourself.

There is no success scenario 3-5 years out where you’re not good at gen AI. But it’s not just about technology, it’s about human capital management and optimization.

Risk in AI Today

There isn’t yet a clear solution today on how companies are managing risk for externally-facing AI solutions. It’s hard to know until there’s a more definitive regulatory environment. Guardrails are ultimately going to play a pretty big role. Back when the internet first came around, everyone was using it to look things up instead of going to encyclopedias on the bookshelf. But, how trustworthy is it really? What are the sources? What are the sites? Today we have Wikipedia as an assumed source of data that’s fairly reliable, but with AI, we have that added risk of “Who are you sharing your data with?” or “How is information being shared?” This is where you’ll start to see people making money off this who aren’t the gold miners, but the people who are selling shovels. For now, human-in-the-middle is the obvious choice. In five years? The world may look a lot different.

You May Also Enjoy