BCCT Gen-AI: A Presentation by Harry Seip, Partner at McKinsey & Company

In June 2023 we hosted an event ‘GenAI Panel: Creating with Machines’, organised by the British Chamber of Commerce Thailand (BCCT). The event featured a series of presentations followed by a panel discussion with guest speakers Harry Seip, Partner at McKinsey & Company; Tee Vachiramon, CEO at Sertis; Akarat Ngandee, Head of New Business at infobip; and Four Kositanont, UX Design Lead at Appsynth. The evening was hosted and moderated by Gareth Davies, Partner & CTO at DDX and Chairman of the British Chamber of Thailand.
This article is a summary of the first presentation of the evening by Harry Seip, Partner at McKinsey & Company, in which he shared his fascinating experiences and insights on the business of Generative AI. 


Harry Seip leads McKinsey’s work on digital and analytics in Thailand. Harry is an expert at building core technology platforms and bringing innovative digital experiences to clients, he is trusted by organizations across industries and geographies.

Harry has a deep understanding of digital-transformation journeys and has helped leading financial-services, retail, and telecommunications clients on a range of issues related to customer-journey design, core-system replacement, and agile capability building.

Beyond serving clients, Harry plays an active role in Thailand’s burgeoning start-up ecosystem. A foodie at heart, he founded a Thai food-and-beverage business in the Netherlands prior to joining McKinsey.


Below is a transcript of Harry’s presentation. 


It’s really an honor and pleasure to be invited here for the British Chamber of Commerce and talk about the business potential of generative AI. I’m a partner at McKinsey Company in Bangkok and I lead our digital analytics and AI work in Thailand and across the region, across different sectors like banking, oil and gas, and consumer. This topic is also a personal passion for me. Before I joined McKinsey, my master’s degree was in artificial intelligence and I did my graduate thesis with the professor who is now one of the leads at Google DeepMind on reinforcement learning. 

Back then, we thought a lot of the use cases that I’ll show you today were 50 years out. So, we’ve come a long way since then. And what I wanted to do in the next 10, 15 minutes or so is just talk about three things. First, what’s happening in generative AI. Many of you have heard about it. Some of you probably dabbled in it. There are probably also some experts in the room. But what do we see? And what does it mean for business? What’s the value at stake and which use cases are driving that?

And then finally, I want to share some reflections of how you or how your business can capture the value from that. And of course, as a disclaimer, at McKinsey & Company, we work a lot with Fortune 500 companies. So, I’ve made an effort to also highlight how to deal with this as an SME or, if you’re not a Fortune 500, how to deal with that. The fact that we have this event today already shows that it has captivated the mind share of business leaders. And specifically GenAI, which allows us to generate new types of content and in my view, we are just at the beginning of that. There’s a little bit of hype. There’s definitely a lot of media attention.

phase of AI

What’s more important is there’re a lot of VC dollars going into GenAI. So, if you set up a startup last year, you should have put Web3 in your name. This year, it’s GenAI and Microsoft investing 10 billion in OpenAI and it is also really a big research focus. And that has been happening across years, of course, but with these new foundation models or LLMs, it has really exploded. We’ll see that not every use case that we see today will be successful. And probably the use cases that will be successful, we don’t know yet.

But in my view, looking at some of the megatrends, AI is a big one. So, yes, there’s hype now, but we are still at the beginning of that. And it will move faster and faster. This is not a technical presentation, and my co-panelists would do a much better job of that. But I did want to share some concepts and some of the terminologies that you will hear, like foundation models or large language models. 


How GenAI Works

Basically, what GenAI is doing is taking in very large-scale data sets, like printing out the whole Internet or all of Wikipedia, then ingesting that into what we call transformer models, which are a form of neural networks with parameters, and then training it to do something. The function that it wants to optimize is something that’s similar to, for instance, if I give you this part of the sentence, tell me what’s the next word. Or if I give you these inputs, statistically, what would be the next highest correlated output? And that’s what it’s basically doing so it’s not intelligent or able to think for itself, but it does that on very large data sets. 

And then we can get these foundational models that you can ask questions that can go beyond what was in the data set. So, it’s not reproducing, it’s generating, and you can fine-tune those models for specific domains. What’s important to know is that this part, training such a model, takes a lot of computational power and data. So that’s only for the larger companies that can do it. But once you have trained and you have a foundational model available, that is much more mobile, much smaller, just to do the inference. And we expect that soon it will be localized and you can have it on your mobile phone.

One thing that is really special about Generative AI is just the sheer scale of adoption.

It’s off the chart. In how many days it took to get 100 million registered users, ChatGPT blows away any of the historical popular apps that we’ve seen. It took them about two months to get there. Probably many of you have a login to ChatGPT. Compared to Facebook, that took more than four years. So, the scale and the speed at which people are experimenting with this is off the chart.

What GenAI Means for Companies

So now we come to what does it mean and what does it mean for me as a company?

What’s the value at stake? So, at McKinsey Global Institute, MGI, we try to quantify what the value at stake is of applied AI, which is admittedly a bit larger than GenAI.

But we come to a number, if you look at all the use cases, of 10 to 15 trillion US dollars contribution of GDP. To put that in perspective, China is 17 trillion US dollars of GDP.

So we believe that AI and GenAI is an opportunity at the same scale as China.

And what we do see is that three quarters or so of that value that we estimate is in four families of use cases. Customer operations: this includes chatbots and call centers. Marketing and sales: like copywriting, pictures, sending campaigns. Software engineering:  code development and R&D, where we see initial results more in the life sciences, but can be generalized to any R&D, product development. And the types of impacts across these use cases are three things that we look at.

One is automation. You don’t have to do it anymore, and a lot of the questions around ‘will I lose my job’ is coming from that. The second is acceleration, so doing things faster, from writing an email, writing a report, writing a presentation. And then augmentation, which I personally believe is the most important one. Imagine that each of you, if you’re a designer or you’re an accountant, you essentially get a virtual assistant that is there all day long that actually can understand your questions and help you to achieve your tasks.

So, your productivity increases by many folds through that. I think you’ve all heard that the latest version, ChatGPT4, actually passes the bar in the US. And not only passes the bar, but it also actually scores in the top five percentile. So, it’s better than human lawyers on the exam they take, and we see this being replicated also across medical exams and elsewhere. What’s interesting is it doesn’t only do the multiple-choice parts but actually can write the essays and understand that.

The other big part that we see is in code development. GitHub Copilot, maybe Appsynth is already using that, and other tools that just help you to, with a few prompts in English, allow you to build an app or a web page or at least the functions, without having to code it yourself. We estimate that’s a 55 percent increase in productivity. GenAI can also do creative work, or what we consider as creative work, right? This picture won the first prize, an art prize. And, of course, you can imagine that it raises a lot of ethical, moral, and other discussions around what is then real, what can be used. And I’ll talk a bit about the risks.

And then finally we saw here a video, that Bruce Willis did for a Russian company without doing anything else other than his okay to it. So, they deepfaked with AI his face, his voice. He starred in ads but has never been in Russia and had nothing to do with it. 

deepfake AI examples

Considering the Risks

I do want to leave you with a message that we believe that AI is a game changer and we’re only at the start. But, of course, there are a lot of risks. And when we look at it, there are eight key risk categories. I won’t go through all of them. One of the common narratives out there is that AI will become sentient and kill us all. That’s not a risk at present. It’s debated a lot, but it’s not something that we see as a big one, especially if you understand how this current crop of AI actually works.

But I think for most of us, this point around IP infringement and privacy concerns is a real risk if you’re a company. Because technically everything you type in gets absorbed and they own a copy of that. The other concern from a society view is around impaired fairness and consumer protection from misinformation. AI can have biases, which are the inherent biases of the training sets that gets into it.  

AI IP-Infringement

And this misinformation gives bad actors a tool to, on a completely new level, sort this code. So you can make a fake ad that is indistinguishable from other types. And here you see an example of IP. So stable diffusion clearly looked at that original image from Getty Images because it can even almost reproduce the watermark, right? And they got sued for that.

And here we see one part that you see a lot in the press is hallucination. Because GenAI doesn’t know what is right or wrong. It just reproduces the next word in the sentence. So, you can get explicitly wrong answers, and more fine-tuning is needed to allow it not to say that 3 plus 4 is equal to 8. 

AI accuracy-reliability

Capturing Value for Your Business with GenAI

And I wanted to actually ask ChatGPT about the last reflection. How can we capture value from GenAI? And this is what comes out. And what you see is not to go to all the 10 points. But the speed at which it can generate and answer such a question. And actually, definitely at first sight, it does say a lot of the same things that we found after spending months of team research on it. So it is around marketing and sales, around automation and things. Hence it is actually quite impressive.

I encourage every one of you to dabble, to open an account, to play with it and try it out.

The same with the image generation. It’s actually quite impressive what they already can do, and these are only the consumer-grade versions that are there. I think right now ChatGPT 4 is the benchmark, but Google especially is working on overcoming that. And this is the final slide. This is how we think about when we get the question. And we get this question a lot. How should I as a company think about capturing the value and scaling the value from GenAI? 

We look at these six dimensions. And specifically in Thailand, it’s A, B and F that are very important for our Thai clients. So, A is actually what’s the roadmap, where do we invest? Also, what’s the business case around that? Because ultimately it will take a lot of money to build, right? And if it’s not clear how you get the ROI, beyond the quick wins that are clear, it’s very hard to get that alignment. So, what we see a lot is that some of the businesses are bullish, but they don’t understand the limitations. And technology says we can do it, but it will take five years. Hence, aligning is a lot of the work that we do. 

Talent, which is very scarce. And specifically in a Thai context, there are certain roles that we need where I would know the six or seven people in Thailand that could do that role so it’s very hard to get that. And there are only a few Thai conglomerates that are able to get that skill and density of talent. That’s definitely the battlefield if you look at Thai companies. Then comes the adoption and change management, because a lot of these things look cool — but if your front-line staff doesn’t use it, then it actually doesn’t capture that value. And for SMEs specifically, I think it’s a lot around the time, space and limited budget that you have.

And how are you actually able to adopt it? A lot of the smaller companies are already busy with doing business as usual. To think about GenAI is something that could be a pipe dream. But I think the unlock here is not to be a generator or a specialist in GenAI, but a good user or translator, as we call it. So, getting smart on how to use the tools and rely, but not over rely on third party providers; that’s the ability to actually take what you need and use it to improve your business instead of trying to reinvent the wheel or trying to go toe to toe with global hyperscalers.


We’d like to express our gratitude to Harry Seip and our other esteemed speakers for joining the event and sharing their valuable experiences and insights. Harry can be reached through the McKinsey & Company website.

Share the Article:

We use cookies and similar technologies to understand how you interact with our website and improve your browsing experience. By using this website, you agree to our use of cookies. Learn more.