The business leader’s guide to AI

Transcript: the business leader’s guide to AI

Hello. Hi. I’m Helen Craven, and thank you for the introduction Holly. I’m a data scientist at Peak. And alongside Ed, Blaine and Matt, we’re delighted to welcome you to the business leader’s guide to AI.

Now when navigating the complex world of artificial intelligence, business leaders can often be overwhelmed by buzzwords, competing categories and quite a lot of hype, which makes it difficult to decide whether they should invest their efforts, where they should invest their efforts, how to achieve success, perhaps even how to find success.

Over the next 40 minutes or so, we’ll have a down to earth conversation on AI today, decoding a complex market with some simple, actionable insights. We’re also gonna have some time for questions at the end. So you have the QR code on your lanyards, scan away, ask any questions you have for the panel throughout, and we’ll grab them up on stage for you at the end of the session.

Without further ado, let’s meet our experts. Ed, could you kick us off with some introductions, please? Yeah. Ed Thompson, I’m a CTO and co-founder of Matillion. We’re across the, well now over on the edge of Spinningfields. And we are a data integration company. And work closely with our colleagues over at Snowflake.

Thank you. Blaine. My name is Blaine Carper. I lead our marketing intelligence team at Snowflake. I started on the front lines as an SDR. I know there are a lot of SDRs here today. And since, in the last six years, moved over to London, and now acting as essentially a data diplomat translating all of the data points into just a few important parts for our VP of marketing over here.

And I’m Matt. I’m the CTO and co-founder at Fuzzy Labs. We’re a Manchester-based company who specialize in machine learning operations, that is MLOps. So what we do is help businesses to productionize their AI to scale it, to improve their processes, things along those lines.

Thank you so much, all of you, for being here today. And as I said, today’s panel is going to be all about diving beneath the hype into the business value that can be generated by AI. But it’s undeniable the AI hype this year has been real. So if we may indulge ourselves for just a moment, putting aside value completely.

What would you say is the coolest AI tool that you’ve seen emerge in the past year that’s really hyped you up? Ed, why don’t you kick us off with that one? So in the past year, you’re kind of including the release of, GPT three and GPT four in that. So then, and they kinda kick started this kinda huge wave of innovation.

So they’re pretty cool, but I guess what’s been fascinating to see is the pace of innovation, kind of as always with tech standing on the shoulders of giants, building on those, on those examples. There’s two that kind of spring to mind that I’ve only seen the last kind of month. They’re both open source projects. And they both kind of do the same thing, but it’s an interesting new idea. One is called Chatdev, and one is called, Autogen.

And they both take this concept of rather than having a single conversation with the LLM. What about if you take a group of them? And then make them work together and give them different roles and different tasks and make them think in different ways. And it’s a little bit terrifying, and I don’t like to share it too much with my sort of engineering organization, but you can essentially build yourself a little engineering organization out of chatbots, and then give them a task to do and watch as they interact to, build some software, which is both fascinating and terrifying.

But, but but pretty cool, I think. Client team work is always cool. Yeah. And the good news is for people employed in the tech industry like myself is, they don’t really work.

But the idea is fantastic.

And maybe they will fairly soon, and that’ll be really interesting. Yeah. For sure, Blaine. How about yourself? Really cool AI?

I think this is a really tough question to answer because there are so many amazing innovations out there, but I’ve been particularly interested in what’s helping happening in the healthcare space, especially there’s a really cool company called Kyron, and they have this product called Mia that is an AI assisted, breast cancer screening tool, supports radiologists and essentially giving them a second opinion. And those kinds of innovations that are having this really material impact on human life, I mean, they’re incredible. And it really makes you think what’s gonna happen next? What is possible?

So those are very cool and inspiring to me. Sure. How about yourself, Matt? Yeah. I think, as Ed says, if we’re having the window of the last year, then that does include the release of chat GPT and while it sounds kind of boring, that would be my starting point to answer this question.

And the reason it’s that I have not come across a technology that has captivated so many people in so many walks of life, you know, when my physiotherapist is telling me how they use chat GPT to improve their marketing.

I’m like, wow, like, how did you even come across this? I thought this was just something for us nerds that the reality is it’s everybody, everybody in every walk of life in every business, in every place in the world that can, is using this to do something interesting.

I think where the innovation, where I see the innovation is all kinds of copilots. You have GitHub copilot for instance, which is gonna make people like me redundant because it will write my code for me. Apparently, you have things that are assisting medical professionals like like Blaine mentions as well. I don’t have a particular technology as an answer here, but co pilots broadly seems to be the the exciting thing. Yeah, for sure.

Yeah, I’m referred to our earlier, Co:Driver by Peak.

Okay. So let’s drive into that usefulness a little bit more then.

Blame working for one of the world’s biggest data companies. I’m sure you have eye firmly on AI use cases. What AI opportunities would you recommend to businesses to get started driving value today?

The boring ones, fundamentally, the things that take up your time that you really should not be spending doing, whether that is troubleshooting code, or if you’re an SDR, writing sequences that are super tailored, and you can use chat GPT to help you with that. I am not a data scientist.

I make errors in my code all the time, but I can use chat GPT in order to figure out what’s going wrong and not have to wait for the US to come online. Well, that’s a really boring use case. It is something that helps me drive time to value faster.

And once you’ve started implementing those really mundane use case, is for AI, and it frees up your time, and you can start to think about, okay, well, what’s the next thing? What’s a little bit more complex?

Because now you understand one of the possibilities, and you can start to explore further.

And honestly, that’s that’s where I’d start. With the boring stuff. Wonderful. Nice. How about yourself? Where have you seen AI driving the most value in the past year?

Yeah. I think I I kinda layer on what Blaine says. It’s it’s, it’s about driving productivity.

And it’s interesting. We kind of people that are at the forefront of technology, and then suddenly become very kind of small, small, small, so conservative about, certain aspects. And, we’ve found that, we’ve actually had to lean in reasonably hard to bring co pilots to the engineering team, bring chat GPT to the marketing team, and there is kind of an element of leading people towards, what’s going to help them to be more productive, gradually changing their working practices. It doesn’t just magically happen.

Seems to be, it was a lot easier to, get my kids to use chattyPT to help them do their homework, then whether that’s right or wrong.

But, but but but more difficult to people to get people to, utilize AI on their day jobs, but, that’s where we’re seeing real world applications being land constrained to the business and working.

Would you add to a similar or a different vein there, Matt?

In terms of where AI is driving value, I guess the people always go to automation as as we’ve heard. I think there’s maybe a broader way to look at it as well, which is just the degree of economic stimulation that how many new businesses or new business ideas even if they don’t pan out, how much like how much are we generating ideas and businesses and concepts using this tooling that weren’t happening a year ago? Or two years ago, that that seems to be significant.

Yeah. Thank you for your thoughts there. So I’ll dive into our office big question. Although those were certainly quite big, I’d say.

Buy in for leaders and budget to buy, these are big stepping stones at the start of any AI journey.

Blamed for businesses that are starting looking to take their first step into AI, can you give some advice about how they could secure that buy in and budget for their business?

When I think about buying an AI tool, I also think about it as buying any other enterprise tool. Fundamentally, you’re gonna go through the same procurement process, the same security vetting process, but the caveat is there are a lot more things that you have have in place in order to be successful in implementing that AI tool and things that you have to look out for to ensure that it’s the right thing for you to implement I mean, the one that is top of mind for me immediately is how secure is this product. How does it fit into my ecosystem?

How much do I have to move my data around in order to train this model? And bearing all of those questions in mind as you’re evaluating, you know, is this the right tool for us, is it going to make us more productive?

That’s where you have to make a really strong use case for your your budget owners.

When it comes to getting the buy in though, we’ve talked a lot about productivity use cases, and that I liked your point about, you know, kind of thinking more about the ideation. What is the possible as well?

Bringing together that list of use cases with with material outputs, even if you can’t measure them yet, when I think about, you know, AI KPIs, it’s quite difficult. The tech has not been around long enough for us to be able to say. This is what it’s going to achieve with this amount of ROI, but having the material list of use cases, understanding how it fits into your ecosystem, knowing it’s secure, that’s how you’re going to get, you know, your budget and buy in, very similar to any other tool, but just with a little higher level of risk.

Yeah. Thank you for your thoughts. That’s some really some really useful advice, I think. Matt, your customers are further along in their AI maturity journey.

What’s important to consider in terms of next steps So yeah, typically it is the case that our customer is a little bit further along in in the AI journey.

Is also often the case that what they do as a business depends on the AI. They have a product that’s built around the AI or something of that nature, or a new business opportunity around the AI, at least.

Where I would go there is questions like can what you have scale can what you have scale in terms of just the technology, but also in terms of the the people building the technology. I that your team’s ability to collaborate on iterating the model, iterating the data, and so forth.

And then you can also look at, especially if it’s a generative kind of AI, like a large language model, do you have the right safeguarding in place? Do you have the confidence that what your generative AI is going to do when the public interacts with it is not going to come back, and bring harm to you as a business or cause harm to the world. So there’s a lot of details to to start to think about as you go from AI concept in those very early days to products and and product productionization.

We often talk with our customers about the first week versus the first six months of any AI project or product.

And right now, especially with Chat GPT and Open AI, it’s very easy to have a very impressive demo in the first week or two. All you really need is the OpenAI API But everything else, all of those things I’ve mentioned, that that’s harder and that takes a longer term view Yeah. Thank you.

And I suppose the the foundation start is always gonna be your data. Right? Absolutely. Everything is always gonna be case no matter how smart the the AI is.

Yeah. Yeah. So let’s, yeah, dive into that a little bit more, to go with me here. I’m a business leader, and I’m worried about my data, and its readiness for AI.

What advice could you give me about getting my data ready for its first steps into the AI world?

Yeah. So, I mean, it is It is crucial. And I think the thing that’s changed with the generative models, so there’s always been a problem ever since AI, from, you know, simple, ML style use cases right through to the new drain engine models. It’s it’s always been a challenge to get and prepare the data, for feeding, those kind of models, any kind of model, really.

That challenge just got harder with generative models. And the reason it got harder with generative models is because they really, more suited to the unstructured data in your organization And generally, the unstructured data is kinda more messy than the structured data. Like, hopefully, if you got a good system, it’s it’s it’s it’s collecting and generating and creating structured data with some structure.

But very often the unstructured data is is is more difficult.

So there’s kind of a there’s there’s kind of an opportunity which, everybody will need to take because the data is valuable.

But there’s also the extra challenge of preparing, preparing that. And then finally, it’s the cost. So, using AI to do productivity enhancements is relatively affordable, particularly if you put it against cost of a person writing some marketing copy versus the cost of AI of writing copy’s massive orders of magnitude difference.

However, once you start getting to training an AI or fine tuning a large language model with your own business data, the cost starts to go up quite quickly, because the underlying platforms, require a lot of power, a lot of technology, a lot of compute power.

And so more care needs to be taken. But companies are gonna need to take the opportunity because it, what comes out the other side can be incredibly valuable. So there’s a bit of caution now, I guess, but, or come back to good data going in. Good training data going in, good prepared training data, and keeping, finding a way to keep the human in the loop to keep that quality high. If we if we can achieve all of those things, then you can, you can start to feed AI models that really work and are tailored to the business process in your business, and that I think is gonna be really exciting for a lot of businesses. For sure, and, music to the ears of a data scientist, if I do say so myself.

So Matt, do I need all of my data to be in a great place to get started, or can I just fix the data that I need for a specific use case?

Find this question interesting because there’s a couple of almost contradictory paths I can take with it. My broad answer is yeah, the the data needs to be in a good place in my opinion before you can really get value out of it from AI. I’ve found what ed the distinction, Edrew, about, a structured versus unstructured data was interesting. I wonder if what we’re seeing here is a consequence of because it was so hard to get insights from unstructured data before suddenly the ability to do it with generative models makes it the the difference that velocity makes it look like it’s much easier now, but it’s actually just we’re catching up to where we might have wanted to be some time ago.

In any case, my general view is yes, the data needs to be ready, but the counterpoint to that is I think large language models make it easier to get more out of less data quickly.

That doesn’t mean you’ve solved the whole problem, but it means that we’ve maybe a small amount of data about a very specialized topic. If what you want to do is the ability to query that data and ask questions about it or remix it into some other format you can get something a demonstration up and running pretty rapidly to prove a business case to justify funding. To build on top of. So I guess there’s that that flip side as well. Yeah. Thank you.

We’re heading into our last question of my notes here. So if you have any questions yourself, ahead and enter those into the app, and I’ll get to those shortly.

Okay, so final question, and let’s take a bit of a glimpse into the future if we may.

We’re firmly in the AI era. And unsurprisingly, that means there’s a lot of conversation around what the future of jobs or work is going to look like.

If we allow ourselves to dream, imagine, in the next five, ten further into the future, there’s a lot of questions we might ask about how our day jobs might look.

So let’s say we’re five years into the future, and we’re writing or perhaps on a panel on the business leaders guides the AI, guide to the AI?

Ai.

What’s one thing that you would expect to appear in that guide in five years’ time. So it it might be Vai because it might just be one, ruling us all.

So I think business leader’s guide I think the the thing that slightly slightly concerns me about the direction that kind of the the tech industry is taking right now is we’re seeing is the emergence of lots of point solutions. So there’s lots of kinda me too, yes, my product has AI.

Built in. And I think what we actually want to see is what companies will actually want to do is apply AI to their business processes. Those business processes don’t necessarily live in a single system. They span multiple systems.

And even if they do live in a single system, the context data that you wanna bring together to make decisions, ask and answer questions, to work the tools is, is multiple systems. So I think there’s gonna be an element of, creating AI solutions that are able to span multiple systems that are able to make decisions across the business process, not just on, you know, in a particular product in a particular way.

And then five years is quite a long way out. So could be quite we could be quite a long way, by that point, but, I think that it’s it’s gonna be very important to get there to keep the human in the loop.

Where we are now, I don’t think we, want to be unleashing large language models directly on people. I don’t think that’s gonna end well, and I kind of give the example of, you know, chat bots. Everybody hates talking to a chatbot, even if it’s a really good chatbot powered by the best large language model, still think you’re gonna end up hating talking to a chatbot at this point.

What I can’t tell you because I’m not a very good futurologist is whether we will be there or thereabouts in one, two, three, for five years time, feels like we probably will be. Okay. So if that’s five years, ten years, blame, the business leader’s guide to AI.

I’ll echo what Edts said that ten years is, like, a lifetime away in terms of technological development, but I think in ten years time, we’ll be at a point where AI is ubiquitous to Ed’s point in your business process you’ll have to have an overarching AI strategy that speaks across all of your different systems. And today, we, at least it’s not like we constantly pushing that you have to have a data strategy in order to be successful. But ten years time, that’s not even a question. If you don’t have a data strategy, your company just won’t exist, because you should be focused on how are you implementing AI in order to move your business forward.

It’s it’s very important. I think at that point as well to highlight how are you using AI to differentiate your business?

Because there are so many companies right now that say they’re doing the same thing with AI, but as we move across the next decade, you’re gonna have to get really, really specific with the value that you provide by AI. Otherwise, you’re not gonna survive.

Big words, ten years’ time. Let’s, dream further.

Twenty years.

I love I love this. I get the hardest question.

So to answer this, I I’m just gonna want us to cast our minds, those who can cast their minds back twenty years to begin with.

No one had heard the words Facebook or Twitter.

I’m trying to remember whether I had was still on dial up. I’m pretty sure I was on ADSL, but it was a new thing and this idea of being permanently connected to the internet blew my mind.

It was probably only a few years prior to that that I had to convince my mom that, no, you can actually use the phone at the same time. I promise. It’s fine.

So imagine thinking about chat GPT twenty years ago, that felt like sci fi. It still feels like sci fi.

You know, some some other things happened over those twenty years as well. One of my favorite phrases that sums up the growth of the software industry came from Mark Andreessen, who of of Andreessen Horovitz, the Los Angeles venture capital firm, He said software is eating the world and what he meant by that was this idea that it’s not just that software is for us nerds cast back to what I said earlier. It’s for absolutely everything, organizations and industries that you can’t imagine benefiting from software will ultimately be software based like your local shop, the post office your government services, everything. That’s what he meant by software is eating the world. That was the opportunity he saw for a massive growth in the software industry to conquer every single other industry.

I think we’re probably looking at the same with AI that it’s going to conquer every industry. I said earlier that what amazed me about chat EPT is how people from every single walk of life seems to have touched it and have an opinion on it and have been amazed by it. Well, then every industry, imagine every industry is using AI and perhaps generative AI to do whatever they do now in a more efficient way in ways we probably can’t imagine. So I don’t have any concrete predictions for twenty years because I don’t want to be wrong and I definitely will all be. But just to paint that broad picture, AI will have eaten the world. Hopefully not in the take over the world way that, was suggested by it earlier, though.

I look forward us in your back here in two thousand and forty three for a bit of a fact check.

Okay, let’s turn some from the audience. Thank you so much to those that have submitted them.

I’m gonna start with this one, just open to the panel.

How do businesses find the balance between speed and customization when they’re taking a step in their AI journey?

I put us. I was literally scratching my head and I put my hand up, but you look it at me anyway. So are we talking about the customization of AI for their business? I’m gonna guess we are.

So I think, right now businesses will need to be relatively selective about the data particularly if they wanna use generative models.

Before you do anything with AI, you need to make sure that the generative model is the right for your use case because we’ll one of the things I’ve heard, is a lot of there’s so much hype around them. Everyone’s like generative model will fix everything and then they come with like a very structured data use case and it’s like, well, the gentry model is not the right use case to apply to that particular problem to get the best result.

But where generative models do apply, the second problem right now, which I imagine is a problem that will, very quickly go away over time, is They’re very computationally expensive. So that cost has to be absorbed somewhere.

So that means you can’t really run enormous amounts of data through generative models right now, and get insights like that. That’d be nice to do. But probably very prohibitively expensive, depends on what, depends on the insides.

So you you you need to be quite selective, to start with. It’s like to figure out, okay, where’s my really insightful data?

Where’s the stuff that’s gonna make the difference? Example, in my business is, kind of support case. You know, we spend a lot of money on support.

And those support questions are detailed and complex.

And, being able to do that, any efficiency you can put in there pays back, really, obviously and really easily. And that’s a great AI use case for us that we’re tackling. But, every business has something like that.

Thank you.

Okay, this one’s got a couple of up votes. So I’ll come to this one next, and it’s probably gonna be our last one today, I think.

Perhaps a common worry, that the data, businesses’ data is valuable, and a company might not want to hand over their data to get the best. The example here at sales price forecast, but whatever use case, for their own products. How do you see API services, perhaps other tools developing so that the function can be called without handing over the data, can people, as opposed to summarizing people use AI use cases without giving away their valuable data, I’ve got two takes on this that perhaps are a little bit left field. One of them is that actually you want your own data sovereignty. So the way we work with our customers is actually we we build you something for it for you to own and manage and run out of open source components, including ideally open source foundation models, that gives you autonomy, it gives you sovereignty, it gives you control of where that data goes, it gives you visibility over it. So that’s one answer.

A sidetrack that it’s more of a curiosity about how people interact with a APIs in a privacy preserving way and there is this notion that’s called differential privacy far too technical for today, but it’s this currently an academic idea, but I can see ideas like this where there are essentially ways to use APIs where your disclosing an encrypted or obfuscated version of your data so that you’re not actually disclosing the real data. That’s probably something that’s gonna become more commonplace in the future, two, two different points of view on that. Interesting. Thank you. Is that a worry that you hear Blaine?

At Snowflake concerns people handing over their data in order to get used from AI. Oh, absolutely. I mean, your data is yours. It’s terrifying to expose that to external parties. And That’s why it’s so important to have the ability to have all of your data just in one place where you do have complete visibility into it, but then also the option to have some kind of data sharing component as well, where you can mask just how much data that you want to share and don’t share what you don’t want to share. So having those kinds of capabilities are key in order to have the ability to enrich your data with the data that you need while still having that privacy pres preserving component as well.

Thank you so much, and thank you to you all so much for your time today.

I hope you’ve all found the entire conversation.

Interesting, and have a great rest of your day at AltitudeX.