Third Bridge provides integrated research solutions for the world’s top investors and business leaders. Our expanding suites of products can help you drive change, putting crucial insights, value chain maps and experts at your fingertips.
The following transcript is generated from Third Bridge Forum, the biggest archive of high-quality expert interviews in the world, providing investors with critical knowledge to make smarter and faster decisions.
Submit your details to gain access to our premium content. If you would like to discuss our coverage of companies and sectors and how we can help you move faster and with greater conviction, please contact us today.
Specialist
Former C-Suite at Humanising Autonomy
Agenda
- Best-positioned companies to leverage AI for cost optimisation and driving revenue, including expected product roll-out timelines
- Investment criteria for AI start-ups – how to recognise a durable advantage beyond simple ChatGPT screens
- Pace of development of GPT and LLM (large language models) technologies and expected timeline for next-generation roll-outs
- How companies have and will continue to infuse AI into existing product portfolios
Questions
1.
Could you identify 2-3 key trends that you think investors should be paying attention to when it comes to AI’s usage, development and trajectory?
2.
I know you’ve spent a lot of time on companies’ AI product strategy. Where do you expect most of the AI spending to be directed over the next couple of years?
3.
When you talk about building an agent or something such as that, what does that development look like? Is it hiring a player such as Palantir or C3.ai to come in, take an LLM [large language model] and build on top of that? Is it using existing products and vendors and seeing them infuse their products with AI? What does that spending look like? Who’s going to feel the benefits of building agents if that were to happen?
4.
When you talk about the small start-ups that are focusing on niche tasks, are they going to come in and sell a SaaS [software-as-a-service] product, or are they coming in and selling a service, building something out on an internal network?
5.
What do you see as the business model in the medium-to-longer term? Is it going to continue to be more customised solutions, or as we have today with the SaaS leaders, will we have an out-of-the-box solution that might require some implementation and upkeep, but could be applied to companies across industries and the tech landscape, or any landscape?
6.
It seems as if the key use case that companies are looking for in the short term is using AI and generative AI to optimise employees’ workflows. We’ve seen some reports, one by Goldman Sachs, that say AI could take over 60-70% of people’s working hours. Can you help us understand what types of jobs that’s applicable to and how realistic that 60-70% figure is, in whatever time frame, of using SaaS tools and services to replace huge chunks of what people are doing?
7.
I’d like to push into the data side, because you’re not the first person who has talked about how important having access to data is and what you can do with it. I read an article that said, “As we move from more general-purpose LLMs such as ChatGPT to enterprise-specific engines – so models trained on enterprise-specific data – companies need to get their data in order.” Could you help us understand where this specific enterprise data that they will train models on comes from? How far out do you think we are from seeing LLMs plugging into internal systems and generating answers based on that specific data set?
8.
You said that anyone who has access to large amounts of data is going to have a material advantage. Do you mean the AI applications or a company such as Zoom saying, “We have access to all this customer data,” What type of data are you talking about that is going to give people an advantage?
9.
Is there a specific type of company that is going to hold the data advantage, or is it more just the larger you are, the more data you have access to? How might we bucket people into, “Really well-suited to use AI. Doesn’t have a lot of data”? How can we be thinking about that?
10.
Is what you mentioned around the Zoom example allowed? Is Salesforce allowed to say, “We’re going to train our model based on all this information”? Are there rules out there? Are we waiting for regulation? What’s the deal there?
11.
I mentioned that 60-70% of work is going to be able to be automated. Is the idea that I have access to all my systems, and instead of having to pull reports and do different analysis, I can look up, “How many customers do we have that have two kids, live in this region and have logged into that system?” Is the idea that you can basically query anything going on in your organisation and it’ll immediately spit out an answer or generate a report? Is that the internal use case that you’re anticipating?
12.
Do the LLMs that plug into internal systems exist today? Have we seen the development of these over the last eight months, or are we still waiting for an effective commercialised version of this concept?
13.
You mentioned Sama Labs and what it does. I feel as if that is sort of what we’ve been waiting for from a commercialisation perspective, so why have I not heard more about that? Why has there not been more immediate adoption of this type of solution that is seemingly going to eliminate reporting and disparate data sources? Why do you think we aren’t hearing more about this?
14.
You were saying that people are going to want to build their own internal agents. Is that going to be on the back of an open-source LLM? What might that development look like? How big does a company have to be to justify building its own vs signing up for a SaaS model from Sama Labs, for example?
15.
Understanding the open-source nature of a lot of the foundational models, which vendors stand to benefit from that trend? I would assume every corporation is going in and building the agent, fine-tuning it, giving it access, and then creating this query bot, but which vendor or systems integrator is going to be generating revenue as this process and transformation goes on?
16.
In what timeline do you expect an acceleration in implementation of the internal LLMs trained on internal data?
17.
What will the upkeep of the internal models look like? Will it be a SaaS company who helps as part of a package, system integrators continuing to come in and tweak it, internal team making changes or something else? How frequently will adjustments be required?
18.
You mentioned spend on AI and how that’s starting to trend upwards. Obviously, every company is talking about AI strategy and what it wants to do with it. How much of the buzz around AI do you think has resulted in budget being allocated in the short term, as opposed to simply looking around, going on trial and speaking with companies, but not actually pulling the trigger yet on large-scale AI investments?
19.
When do you expect to see the massive acceleration in AI spending? I think there will be a lot of interest in this question generally, but especially for someone evaluating a Palantir or C3.ai.
20.
Have we seen a lot of investment from non-tech companies so far, or are we still waiting to get to that tipping point where they say, “It’s time to invest USD 1m in AI, see what’s out there and start putting together and investing in our strategy”?
21.
You’d think for something such as a law firm, the whole agent idea would be fascinating for it in terms of being able to access all this information once it’s digitised as well. Are there any additional product hurdles? I know the generative AI step was a big one for the industry, but what do you think might be the next challenge we’re working to get over that might be another lead into AI and another opportunity? Or is there nothing else in the short term that would lead to a significant acceleration in the industry?
22.
Generative AI obviously transformed the AI industry. Is there a next generative AI? What might be next up in the road map, if anything?
23.
Are there any existing tech companies that you would identify that could be beneficiaries of the development of agents or internal query models? We talked about potential start-ups related to interfaces, and we discussed systems integrators, Palantir and so on, that could actually develop the additional step on top of the foundational models, but is there anyone out there right now that you’d say, “There’s a tailwind for this reason”?
24.
A few months back, a leader at IBM said that over the next 5-10 years the company could potentially replace up to 8,000 back-office jobs, utilising AI. If we assume a USD 100,000 average salary there, that would be almost USD 1bn in cost savings related to personnel. As people start to realise these cost optimisations and there’s the ability to spend 60-70% less time on the remedial tasks we talked about, how much of that cost optimisation do you think will be directed towards AI and software spending vs just retaining employees, having them do different things, or pocketing that profit?
25.
How do you expect budgets to trend for software and specifically AI-related software? Do you think we could see 10% of the USD 1bn allocated to software that previously wasn’t there, or is it just going to become the expectation that if you’re a solution, you’re offering AI and it’s part of it, and, “We’re not willing to pay you more”? What are your expectations there?
26.
What do you think gross margins will look like for the LLM-powered applications, and how and why might they differ from traditional SaaS margins?
27.
I saw a report that a gen-AI-powered query has 5x more compute power than a Google search right now. How do you expect that dynamic to trend as we see more development and as compute capabilities improve? Do you think it will become cheaper to run these queries, or is it just always going to be lower-margin than what a typical SaaS is? Is it always just going to be expensive to run these models?
28.
The development of the LLMs is obviously a very expensive proposition, but for the applications being built on top of that and the start-ups that are developing those, do you think there’s a path to achieve profitability amid high growth in the short term? Or are these going to be the types of businesses where they’ll burn through cash in the short term, be in grow-at-all-costs mode and figure it out later?
29.
Back to the development of LLMs, you mentioned that Google said in an internal memo that it thinks open-source models are going to win out. That being said, if you were advising a company, would you advise against developing its own LLMs and instead rely on public-source foundational models? How long do you think it will take for those to outperform LLMs from Google and OpenAI? It seems OpenAI put a lot of money into handling bias, filtering out hateful speech and things such as that. Is that something that’s going to be able to be accomplished within open-source models?
30.
If you were talking to an investor and they said, “I’m investing in a verticalised LLM for the healthcare industry”, would you say, “I wouldn’t advise doing that”? Or are there certain circumstances where there could be a really large return on investing in an LLM right now?
31.
When you’re advising people to invest in agents, where are you directing their spend? Is it the system integrators and small start-ups? Where do you advise them to spend money?
32.
If I were a VC investor, how could I get involved in benefiting from the process of agents being implemented, or is there not really a way?
33.
You were mentioning how we’re seeing different AI companies popping up. A specialist in a previous Forum Interview a few months ago said it feels like a bit of a gold rush – having a million AI companies with half-baked ideas. With a VC or PE audience in mind, what specific criteria are you using right now to select AI start-ups and projects to invest in?
34.
A specialist in a previous Forum Interview said it’s probably going to take a couple of years to flush out the winners and losers. He said he’d be cautious about investing in companies right now. Do you agree with that sentiment, and given the valuations, would you be cautious in taking shots right now on companies? Or is this a situation where even if 90% of these companies fail, we’re going to see some massive unicorns and future public companies developed right now? Would you be investing? Would that be your primary area of investment right now?
35.
I’d like to discuss the hallucination idea. I know there’s been a lot of conversation around it. I cover chatbots, and one of the problems with implementing generative AI is you don’t want to give people an incorrect answer if you work in healthcare or financial services. You can’t really afford to be giving incorrect answers, so is this idea of being able to plug into specific data sets that you can control going to fix the hallucination problem for commercialising a gen-AI-fuelled tool? What are your thoughts on the development and advancement through this hallucination challenge that’s plagued these platforms over the last year?
36.
When the agents start coming to market, who do you think will be the first to roll out what we’re discussing? Within six months, do you think we’ll see the large tech companies saying, “We’ve got to do all our reporting. We built out an agent, we used a foundational model”? They obviously have the technical capabilities internally to develop it. When do you think we’ll start hearing about it, and when is it going to become common practice to have one?
37.
We were talking about ancillary companies or industries that could potentially feel AI tailwinds. You mentioned GPUs [graphics processing units] and talked about the idea of picks and shovels. One of the obvious ones we said was Nvidia with chips. Is there any other public or large private company that you think will play a role in the powering, implementing or anything else related to AI adoption that you’d call out? What do you think investors should be aware of right now?
38.
Is there anything we haven’t covered or any points you’d like to highlight around enterprise software and AI?
