Investing in TrueState AI
A few thoughts on enterprise AI and backing TrueState
This past week our investment in a $1.5m round for TrueState, an enterprise AI company, was announced.
In this pre-seed round we joined Airtree and Side Stage Ventures backing the founder Will Ashford to build out a fast-deploy AI enterprise platform.
At first glance this might seem like an odd investment. Well funded, venture-backed competitors like Together AI, Anyscale AI, Databricks/MosaicML and even OpenAI all have various enterprise focussed initiatives that aim to achieve similar purpose, helping businesses build and deploy custom AI apps.
What convinced us to back Will is his unique background and experience which lends himself unique insight into this problem – spending years helping C-suite leaders ‘adopt AI’ to achieve real bottom line results. Will previously worked at McKinsey Quantum Black (their AI division) where he would consult, build and deploy AI solutions into large private companies to help them make more money and save money. It was much less about ‘cool demos’ and much more about making a real impact on the business.
Why now
Over time Will observed that the cost to build and train many of these models was plummeting, research suggests 50x faster than Moore's Law. This wasn’t necessarily employing frontier LLMs in the solutions, the costs of which to build and train are increasing as parameters increase, but across all types of Data and Machine Learning models to achieve an outcome. Over time efficiency gains are winning.
According to Will common ML scenarios which used to be expensive are now very cheap to train and run. These include sales enablement scenarios, such as making your sales team more efficient in filtering leads, and integrating ML in operations – the boring stuff, that is necessarily about fun and impressive chat interfaces.
The technical gap
The gap in this market is the technical barrier. Companies that have in-house AI and data science teams can build their own solutions but even that is often a challenge.
Will mentions, “across countless clients and portfolio companies, the story was always the same; the complexity of building high-impact AI solutions was too much for most to handle. This meant they resorted to outsourcing the initial build of their solutions to AI experts, at a substantial cost.” The ability for businesses without tech capability is event harder.
Thus the big prize is becoming a platform that allows experts, professionals in their role that are not necessarily software engineers, to adopt AI in their unique situations. This arguably is why consumer products, chatbots, are the winning use-case in GenAI revolution – suddenly, everyone can ‘use AI’.
For TrueState its around ‘workflow templates’ (see below) that allow businesses to choose from proven scenarios and then customise in their no-code builder.
The thin end of the (enterprise) wedge
Where the thin end of the wedge (aka. enterprise use-cases) kick in are really around data confidentially and privacy. For example, if you’re running highly classified medical trials, you need to run complex ML models over the live human results, and you need LLM endpoint to ‘chat through’ the results in a private cloud environment, what do you do? You’ll need to experiment with existing platforms out there – build your own deployment rails, integrate data pipeline platforms and models and host on-premise open source models or find a local secure cloud provider (but risk data breaches with this vendor). Its complex scenario for non-tech companies, such as pharmaceutical companies, but one TrueState hopes to solve over time.
To be sure, all of the LLM providers are going after these types of high-value uses cases – but locking into one vendor is not realistic. The future will (hopefully) be multi-model (see below) and enterprises demand flexibility for the tasks, costs and control. A 7B vs 70B parameter models will be useful for various reasons.
VC POV
From an VC investor perspective why did we (Galileo Ventures) invest? We love backing exceptional emerging founders who are building out new product or market categories. In reality, Will is crossing both domains but with a different approach, going after ‘non-technical’ enterprise customers versus going after internal engineering teams. While most early winners, the those raised the most venture financing, are doing the latter, we believe this market is so huge, and becoming larger, that there will be room for many approaches and many winners – the big prize is helping traditional markets access AI in an affordable and reliable way.
In some sense, the early winners are consulting firms, with many reporting multi-billion revenues from ‘AI consulting’. They’re helping large enterprises ‘navigate’ various solutions (i.e. just on-sell Microsoft OpenAI Azure) but we think this a positive signal that platforms will enter and win out over time.
Another reason we like TrueState is that we’re big believers that workflows over chat are more valuable for businesses. Businesses don’t want a chat bot, they want ways to improve or replace cumbersome workflows or create entirely new ways of doing things. Individual users make like a chat interface (more human-like interaction, easy to use etc) but businesses want much broader integration, across teams, data silos and applications that create something that is impactful.
Finally, do we believe a small little tech startup in Australia has as advantage? Yes! For one thing the distance can be an advantage – easier to build out a different go to market when you’re not surrounded by ‘same-same’ AI companies, but also our environment lends itself to different customers. Often less tech savvy, bit more conservative, but ready to adopt AI solutions that drive value. If you can win Australian enterprises, you’ll have a good chance winning other markets too.
If you’re interested in reading more, check out Will’s blog below.
