This is an audio transcript of the Tech Tonic podcast episode: ‘Making money from AI — Searching for a ‘killer app’’
Madhumita Murgia
This is Tech Tonic from the Financial Times. I’m Madhumita Murgia, and one thing I’ve been wondering about a lot lately in my role as the FT’s AI editor is this: loads of money is pouring into AI, but how is AI actually being used by companies? What difference is it making? Is all the fuss and money well worth it? Which is why we sent our producer, Josh Gabert-Doyon, to a call centre in London.
[AUDIO CLIP FROM CALL CENTRE]
Madhumita Murgia
At the UK’s largest utility provider, Octopus Energy, employees are doing their best to manage customer complaints and queries.
[AUDIO CLIP FROM CALL CENTRE]
It’s a big open room with a purple carpet and clusters of desks. Joe Richardson is the operations director.
Joe Richardson
Business team over there. So they look after our business energy customers. This is Will. He’s one of our experienced energy specialists. You see him on the phone with a customer. Looks like he’s updating some billing there.
Madhumita Murgia
Our producer Josh visited Octopus’s London call centre because this is one of the places where generative AI is actually being used in real life. The workers here use automatic transcriptions, AI-generated summaries of customer calls and an AI assistant that automatically drafts responses to customer queries.
Joe Richardson
So if a customer emails in, when you go to open up their email, there will be a pre-populated draft reply. We don’t send any out automatically. You can see that reply. You can edit it and then you can hit send.
Madhumita Murgia
Instead of having to figure out what went wrong with a bill or the answer to a complicated tariff question, the customer service reps at Octopus have that all figured out for them by AI.
A few years ago, during the height of the European gas crisis, Octopus took on millions of new customers from failed energy suppliers. They found themselves swamped by customer inquiries so they turned to AI, specifically a version of ChatGPT, which helps them draft an email response.
Joe Richardson
It takes away a lot of the less interesting parts of the role. If you’re someone who works in a customer-facing role like this, you love helping people but actually, like, typing out emails is not often the most fun bit. I’ve asked some of the team, I said, do you feel it like, lacks empathy though, if you’re sending it from AI? And sometimes they say, actually it’s got more empathy. He’s read that email and it said, oh, I’m very sorry to hear about your wife’s illness. It adds these touches because it’s got that time to it, which you as a busy person working in an operations team just might not have the time to do.
Madhumita Murgia
Almost half of the emails sent by the customer service team are now generated using AI. But is that actually saving the company money? Can they now, for example, cut the number of call centre workers that they need to employ?
Joe Richardson
We’re not really anticipating any kind of broad shifts in kind of moving people around different departments. But I think the nature of what people will do over time will change. But what we have noticed in 2024 is our response times have got better, our account health has got better. We’ve also noticed that it speeds up the time from a new starter to be able to get quite proficient.
Josh Gabert-Doyon
Is AI making the business more profitable?
Joe Richardson
I mean, there’s so many moving parts that I don’t think I could quantify that. But lots of things change all the time. It’s quite difficult to isolate.
Madhumita Murgia
This is something that’s been bothering me while covering AI over the past few months. How much is AI helping companies increase revenue or become more efficient? Is the technology here to stay and is it going to justify the huge amount of money that’s been bet on it by investors?
Silicon Valley has been going through an AI boom. Since ChatGPT, investors and Big Tech companies have been pouring billions of dollars into the technology, turning AI companies into some of the most highly valued in the world. But lately, investors have started to wonder when they might start seeing a return on that investment. Companies like OpenAI are burning through cash to build bigger and bigger models, while the Chinese start-up DeepSeek has proven it can be done more cheaply. The worry is that without a killer app for AI or real use cases, that return on investment might never come. It’s a worry that’s created a sharp divide in the financial community, one I wanted to hear more about.
Jim Covello
I’m Jim Covello from Goldman Sachs. I’ve been at Goldman for 25 years.
Madhumita Murgia
Analyst Jim Covello has been called Wall Street’s biggest AI sceptic. He works at Goldman Sachs with his old friend George Lee, who’s much more bullish on AI.
George Lee
George Lee. I’m the co-head of the Goldman Sachs Global Institute. I have Jim beat by a few years. I’m a 30-year veteran of the firm.
Madhumita Murgia
George and Jim are Wall Street veterans. But they disagree over whether AI will ever make any money. Covello says the problem is a lack of any obvious application.
Jim Covello
You have a technology that isn’t quite ready for prime time today that’s getting more expensive. And therefore, what I’m looking for is really what applications are people going to start using and paying significant dollars for that justifies the cost of the investment.
Madhumita Murgia
George, if you could comment on, you know, why are you excited by the upside of AI? What makes you bullish about this?
George Lee
OK. First of all, I would say it’s a bet on the fact that this technology will be so generally useful that there will be use cases and demand that will emerge that we can hardly imagine today, that will soak up all this capacity and will create the predicate and platform for profitable business models. People in the legal profession, in the engineering and software area, in consulting, in media, in material science, there are sophisticated, high-paying job impact of this technology and there is adoption there.
Madhumita Murgia
Jim, do you want to come in?
Jim Covello
Sure. So far, what we’ve experienced is while they’re useful in those things, the ability to monetise any of that is limited, right? It’s a relatively small increase in efficiency. It doesn’t really allow you to replace a significant number of employees; certainly doesn’t allow you to replace any high-value employees.
Madhumita Murgia
How do you think this compares to other types of technology transitions that you’ve witnessed over the last, you know, few decades?
Jim Covello
If you start off by thinking about the history of technology transitions, the ones that have worked really well, the common element of those things in my mind are there were applications that were immediately identified that could take use of the great technology platforms that were being developed. E-commerce allowed Amazon to sell books more cheaply than Barnes & Noble from day one, so there was economic value and benefit to the user. Here, this is very expensive technology that in some cases is replacing cheap labour. If you look on some of the other technology transitions that we’ve seen over the last 30 years that maybe were less successful — think blockchain, crypto, virtual reality, metaverse, 5G — one of the common elements that they shared is the technology platforms we developed and then we tried to find solutions that could utilise the technology. And that’s more where I think AI fits.
Madhumita Murgia
George, what do you think the DeepSeek developments mean for this profitability question?
George Lee
The DeepSeek models are garnering enormous attention in this ecosystem. I think they signal what always exists in technology waves like this, which is the potential. I wouldn’t say there’s evidence yet, but the potential for discontinuous innovation that really does lower costs. And if one believes what they read about the DeepSeek models, they’ve achieved near-frontier capability at an order of magnitude less capital investment in data and computation than other models. And the good news is, if that’s true, I think it addresses many of Jim’s frontal concerns about the technology.
Jim Covello
Yeah, I think George says that really, really well. And this is something else we’ve wound up talking a lot about . . . the more costs come down with more capability, the better, right? That enables applications. It lowers the bar for applications that could generate profits for shareholders.
[MUSIC PLAYING]
Madhumita Murgia
So lower costs mean AI could become more profitable. But generally I agree with Jim. AI looks like a solution in search of a problem. There are no obvious moneymaking applications yet for a technology that has been incredibly expensive to develop. But at the same time, this is general-purpose technology. It might be hard to imagine where applications will emerge in the future.
The more I’ve spoken with users of AI, the more curious I’ve become about how it might change our workplaces and how that will affect a company’s bottom line.
AI-generated voice clip
You are unique, and so am I, as well as all our colleagues, partners and customers . . .
Madhumita Murgia
That’s an AI voice used in a staff training video at the clothing ecommerce company Bestseller. It was made by an AI company called Synthesia.
AI-generated voice clip
We all have a unique personality that makes us behave differently.
Madhumita Murgia
The voice you’re hearing is AI-generated and accompanied by a realistic-looking human, delivering their lines straight to the camera.
AI-generated voice clip
. . . together as a team. To recruit or sell clothes, it helps to have a clear understanding of your own and others’ personality profile.
Madhumita Murgia
London-based Synthesia uses AI to make instructional videos like this for businesses to use with employees or customers. Victor Riparbelli is the start-up’s chief executive.
Victor Riparbelli
I mean, imagine you open a PowerPoint, you select an avatar and you kind of position it in the . . . in your video. And then in the speaker note, you just type the script of what you want the avatar to say. And then you have a visual aid so we can add in, you know, your images and put some text on screen, etc. It really is that simple.
Madhumita Murgia
Basically, Synthesia’s selling point is that making a video using its AI is a lot faster and cheaper than hiring a professional video producer. It also means you can make a lot more videos a lot more quickly.
Victor Riparbelli
So if you want to personalise videos to a wide variety of customers or employees, you can also do that. And what that means is that you could, for example, create, you know, 50,000 different videos for 50,000 different customers.
Madhumita Murgia
Riparbelli said he didn’t start out with a plan to make training videos, but that’s what their AI model ended up being useful for.
Victor Riparbelli
The first thing we tried to do was actually to build technology for video professionals, marketing agencies, video production companies, Hollywood studios, etc. What we realised was that the quality threshold for all the kind of video professionals was incredibly high, and it wasn’t really that . . . it was kind of a bias and not a painkiller. They weren’t really that interested and they thought it was fun to play around with. But ultimately, at the end of the day, they would still go back to doing things the way they’d always done it. And so what we realised was that there are billions of people in the world who really, really wanted to make video, but they’re not making video today because they didn’t know how to use a camera, they didn’t know how to edit it after the fact. They couldn’t get the budget from their boss, etc.
Madhumita Murgia
Synthesia is now valued at over $2bn. Still, like many other AI start-ups, Synthesia hasn’t actually made a profit yet.
A potential candidate for the killer app for AI that many people are really excited about is coding. I’ve been following AI coding companies closely, in part because Silicon Valley sees programming as exactly the kind of thing Gen AI should be good at, a task that involves writing a lot of repetitive, boring code. One of the start-ups that has emerged as a frontrunner in the world of AI coding is a company called Replit.
Amjad Masad
You describe the app you want to build. We have an agent that builds the app for you.
Madhumita Murgia
This is Amjad Masad, Replit’s founder.
Amjad Masad
And then you give it feedback and then iterate on the app. For the first time, non-developers are able to make a software that was only possible with code.
Madhumita Murgia
Venture capital investors have been all over Replit. At their last funding round in late 2023, Replit was valued at over $1bn. But does that actually translate into profitability?
Amjad Masad
We’re not profitable. I think if we really wanted to, it’s possible. But I think that given our cash position, given how much growth we are experiencing and how much growth we can experience, it makes sense to continue investing.
Madhumita Murgia
So Replit may not be profitable either, but Masad says that the customers who are using their AI coding tools are finding it a way to cut down on costs, particularly the costs spent on software-as-a-service products used by businesses, also known as SaaS. Take Sears, the famous American department store.
Amjad Masad
Sears is a 100-year-old company. And their systems were very old. They hired this really tech-forward team. And, you know, typically companies are using hundreds of SaaS tools to manage their business. But this company and this team went all in on Replit and started building AI agents instead of traditional SaaS tools. So like one tool they built is for their workers that everyday go around and take jobs for maintenance at different houses to optimise their routes and to make more money. And who’s using these tools? It’s not engineers, it’s the operations team.
Madhumita Murgia
Masad is even using Replit’s AI coding agent in his own daily life.
Amjad Masad
I use it for a lot of quantified self things. Like, you know, I was working with a sleep doctor to optimise my sleep and he gave me a sleep diary, gave me a PDF and I’m like, I’m not going to print out a PDF and like write it down. I’m probably going to forget. So I gave it to Replit Agent. I just gave it the PDF and told it make it an app and it made it an app. And now every morning I just log my sleep and it can chart things for me. I kind of can send it to my doctor. And I do that quite often with every sort of aspect of my life.
Madhumita Murgia
Masad says Replit’s tool allows users to work faster and to be more productive. So has he quantified how much more they’re getting done?
Amjad Masad
It’s hard to put a number on it, but I would say perhaps anywhere from two to five X faster. And the newer you are to programming, the more impact it has on you. I think the more experienced you are, the less impact it has on you.
Madhumita Murgia
Replit’s technology is impressive, but it still leaves me wondering about the valuation some AI companies are getting. After all, productivity hacks don’t make an industrial revolution. We’re not talking here about technology that utterly transforms the way we live, in the way electricity or the combustion engine did. In the best-case scenario, artificial intelligence might give workers more freedom. The boring work gets done by the bots and humans get more time to be creative. But there is another issue which we haven’t yet addressed. Everything depends on workers actually adopting AI in their daily work lives in the first place. More on that after the break.
[MUSIC PLAYING]
As the FT’s AI editor, I’ve spent the last year trying to figure out how AI is going to make money. Investors have been throwing tons of cash at the technology, but up until now it’s been impossible to tell exactly which applications will make a return on that investment. And much will depend on which applications workers and companies will actually want to use. So let’s head back to the London offices of Octopus Energy. You heard our producer Josh there at the start of the show. Joe Richardson, head of operations, gave Josh a demo of their generative AI tool for customer service emails.
Joe Richardson
Here’s an email. So it says, hello there. Can you let me know what my balance is and when my next direct debit will come out, so when my next payment will be made? I’m also thinking of getting an EV. What are the rates for Intelligent Octopus Go — that’s our EV tariff — and how do I get an EV charger? Fairly typical kind of questions that a customer might ask. You can see here the pre-populated reply. It says, hi, Joseph. Thanks for reaching out. It talks about your EV options and how to get an EV charger. And so if you want to use that, you can click use and then you can go in and edit it. And one thing I’ve noticed is that it’s got the proper name, whereas it’s been signed off here, so I can go in and change that. It also hasn’t included my name. So I would change that. It’s included something that wasn’t asked directly in that one, but was probably asked further back up the thread.
Josh Gabert-Doyon
Sorry, can you just explain that line that you just deleted and what it was?
Joe Richardson
Yes. So in a previous email, I’d asked about the compatibility of different cars. And so I actually, it hadn’t just looked at the last email. It replied to some unreplied questions from earlier.
Josh Gabert-Doyon
But that line about compatibility with Tesla model Y, that wasn’t a question that was asked in the email. So that seems like it’s coming out of nowhere, right? I mean, just can you read out the line?
Joe Richardson
Yeah, so it’s compatibility. Yes. A Tesla model Y is compatible with Intelligent Octopus Go. Now, actually, if I go back further back up the thread, it was a question asked last week that didn’t get an answer. The AI has found an unanswered question earlier on and replied to that. So in a way, I mean, you could leave that in and maybe say, hey, I see you asked earlier, but in a way the AI has done something that might be missed by a human there.
Madhumita Murgia
The email-writing robot got a bit muddled replying to a question further up the email chain by mistake. This seems like a pretty common experience with generative AI. It might do some useful work, but it still needs pretty close supervision. That means the nature of people’s jobs could change dramatically.
Claudia Harris
Companies underestimate how much people’s routines are disrupted by AI. And by routine, I mean how they spend 9 to 5, and the challenge for people in unlearning the routines that they’ve established over years and decades.
Madhumita Murgia
This is Claudia Harris, the co-founder of a tech-training service called Makers, which works with businesses on AI adoption. She says two years into the Gen AI boom, not that many companies are getting employees to use AI. It’s just not really catching on.
Claudia Harris
We often come across companies that buy an enterprise AI solution and are surprised the uptake across their teams is 20 per cent or less. People are given access to a new AI tool and they just don’t use it.
Madhumita Murgia
The question is: why? Well, Harris says managers and executives don’t realise how significant the shift is for employees. It’s a big change. What’s more, there’s instinctive resistance, this idea that AI will take our jobs. But Harris says that fear may be misplaced. Take the example of a spreadsheet.
Claudia Harris
When you had Excel introduced into the workplace, you might have imagined that the number of accountants would go down. But actually just the expectations of what accountancy can do went up and you needed more accountants. And I think the same is going to be true with things like software. Now that we have the ability to do more, more quickly in software engineering with AI, then what we want software to do in our world and in our lives is going to increase.
Madhumita Murgia
It’s too early to say how jobs might change with AI, but the tech industry is still hunting for the killer app. Some people are sceptical they’ll ever find it.
Sarah Myers West
The business model for this sector is in a place where it’s profoundly speculative.
Madhumita Murgia
Sarah Myers West is managing director of the AI Now Institute and a former adviser on AI to the US Federal Trade Commission.
Sarah Myers West
There’s just so much sort of testing and experimentation that’s coming from different companies. It’s the promise of cost savings rather than demonstrated cost savings.
Madhumita Murgia
West’s argument is that at this stage a lot of the big tech platforms are simply making sure that if AI becomes wildly profitable, they will be at the forefront of it.
Sarah Myers West
Both Google and Meta’s CEOs — so Sundar Pichai and Mark Zuckerberg — have said in interviews over the last year that part of why they’re making these tremendous bets is that they see the cost of underinvesting in AI as much more significant to them than overinvesting. So the vision really is one where they want to make sure that they’re able to lock people into their ecosystems for the next decade or two. And if they underinvest, they worry that they’re going to lose out, which really is a race to the bottom in terms of competition in the sector.
Madhumita Murgia
I mean, do you think there’s going to be an explosion of innovation among start-ups building using LLMs, or do you feel more sceptical of that claim and think that it’s all just going to accrue back to those making the models themselves?
Sarah Myers West
Oh, I mean, I would love to see an explosion of innovation, but I am feeling really sceptical about whether it’s going to come from start-ups in this sector. I think certainly when you look to the types of use cases that we’re seeing, they’re not the kind of like disruptive or creative societal changes that one would hope to see with, given the amount of hype around this technology. A lot of it is just, well, we’re going to automate practices that people feel have gotten a little bit boring in their lives.
[MUSIC PLAYING]
Madhumita Murgia
What’s clear is that for now, there’s a sizeable gap between the ambitions for AI and the reality of its applications. So where does this leave the giants of Silicon Valley, particularly when Chinese companies like DeepSeek are at their heels? Have the likes of Google, Meta and OpenAI got their AI business model all wrong?
Voice clips from the next Tech Tonic episode
OpenAI is estimated burnt through about $5bn last year, which is almost as much as it raised.
They are essentially betting on these models reaching something like human-level intelligence and beyond within the next couple of years.
Madhumita Murgia
That’s next time on Tech Tonic.
[MUSIC PLAYING]
This season of Tech Tonic is presented by me, Madhumita Murgia. Our senior producer is Edwin Lane and our producer is Josh Gabert-Doyon. Our exec producer is Manuela Saragosa. Breen Turner, Sam Giovinco and Joe Salcedo are our sound engineers. Original scoring by Metaphor Music. Our global head of audio is Cheryl Brumley.
