And we’re off…
Twin Path Ventures Mid Term Report
This week we at Twin Path Ventures will be writing our 5th pre-seed cheque, since our launch in June this year, into UK based AI-first startups. With our 6th deal signed and ready to go and with hopefully at least one more close to being completed, we are ahead of schedule with just over 50% of our 2023-24 capital committed.
To achieve this pace of investment has required us to source, meet and review almost 150 startups and subsequently carry out initial Due Diligence (we now warn startups that our 55 question-strong Due Diligence questionnaire is not for the mere curious, or those whose AI capabilities are not world class, or for startups building AI capabilities and functionality not now but somewhere in their future stack) on over 80 startups. Subsequently, if and when we believe an AI first startup is a fit for our first cheque AI specialist fund, we deploy our specialist technical and commercial advisors from the UK's leading research institutes, universities or leading AI startups to help us assess and select startups that have the know-how, capabilities (including data), demonstrable "market pull" and GTM pathway to build a best in class AI powered product that can disrupt an industry. We make no apology for the intensity and depth of our Due Diligence process (unusual we believe for a pre-seed fund) and it is definitely frustrating for us and those startups who after going through our process only to find that we fail to get conviction on (which like most funds is the majority). Nevertheless this activity and data has I believe given us good insight on the good and the bad of the present UK AI startup scene.
In the good category, we have found:
Deal-flow is strong especially from technical founders who have founded AI start-ups before, or had previously worked in senior AI roles in the UK startups, scale-ups and corporates. We have also found a good hunting ground in universities but only when Tech Transfer teams do not try to take unreasonable amounts of equity (we have seen universities who insist on 30-50% stakes in spin-outs they are not intending to support, finance or commit to going forward).
Other VC's, especially those based in London, have been supportive by either introducing us to AI first startups or, at the bequest of founders, carving out allocations for us in over-subscribed rounds. I appreciate that fellow investors do value getting our opinion and insights on the startups they source AI capabilities and the feasibility of their tech/product development roadmap. Thank you to the likes of Concept Ventures and Seedcamp and, so far, the other 7 funds we have co-invested with. Also a big thank-you for the bulk of UK and Europe’s VC's who are still waiting on cutting edge AI startups to achieve significant market traction (especially first time founders even when their research and publication records are outstanding) before making a commitment to invest. We may not have gotten into rounds of pre-seed startups founded by the likes of ex-Deep Mind researchers or from the UK's leading AI university if most VC's were not bravely waiting on the side to see how the present AI revolution will work out. I am still haunted by the great startups that I failed to invest in from my previous funds or as an angel (I am probably going to be haunted by at least 2 startups we at Twin Path passed on or did not fight hard enough to get in) but it seems ...
That we are building conviction in the investability of this present wave of AI startups. As I write Open AI has just released ChatGPT Enterprise that will address some of the security and customisation issues that enterprise needed addressing before they let their staff use generative AI and as a consequence the "thin skinned" startups whose offer was simply to enable enterprises to use the pre-trained Large Language Models are now probably in big trouble. This "platform dependency" on foundation models built and subsidised by big tech is making many investors nervous of backing generative AI startups but the space is incredibly fast moving and for every development that sinks a "fleet of startups'' others arise that create blue oceans waiting to be explored. For instance the same Open AI is now allowing its LLM's to be "fine-tuned" which will give new commercial opportunities for those with unique and exclusive access to albeit smaller databases (especially if very domain specific) to use LLM's to build something powerful and unique for specialist markets and defined use cases. Whilst news last week that Phind has managed to fine tune "open sourced" Code LLAMA to match the same performance as GPT 4 for a fraction of the cost, means the gap between Open Source and Closed LLM's is closing and therefore mitigating somewhat the platform dependency threat. In the end investors can either sit on the sideline whilst the most significant technology since the birth of the internet explodes waiting for things to be clearer and growth trajectories for startups to return to linear (oh the good old SaaS days) or they can dive straight in to the murky frothy water and trust themselves to swim. We are doing the latter. In these torrents there are, we believe, some things ( liferafts if you stick to the metaphor) to grab on to and here are ours ( the type of AI first startup we are looking to hear from):
Large Language Model Augmented Autonomous Agents (LAA's). Yes AI Assistants or Co-Pilots are all the rage but to get them to truly be self-learning requires a team of world class NLP, RL and Knowledge Graph experts, a clear loss/ reward use case, the goodwill of a community of super-users and a very smart market entry plan that quite quickly allow a team to gain the first hand experimental data needed if the agents are to be trained to generalise. So in short LLM augmented autonomous agents will be super hard to pull off and again it will come down to the talent and "nerve" of exceptional people. In our experience to date these are founders that can be found in the UK or can be attracted to locate to the UK. When it comes to AI, London is presently a very big draw. If this is you please get in touch.
Domain specific "work flow" automations built on top of LLM foundations but with specific task related processes requirements that LLM's cannot do on their own. Startups able to fine-tuned LLM's with domain specific data and integrate the results so that their product is demonstrably able to carry out tasks that LLM's like GPT or Claude cannot or carry out tasks to a performance level that is unmatched by humans or machines. If the domain is big, the task you are automating is non-trivial to the end user and you have the ambition to build a full stack service and not just a “bit better” software then we are super keen to hear from you.
Synthetic Data Generation startups whether they are using ABM, GAN or LLM techniques to generate new data sets that can be used to build uniquely capable models that can for instance find "needles in haystacks" or can be used by AI teams in startups and corporates to build more robust and accurate models are super interesting. If this is you please get in touch.
Startups building Computer Vision or small model NLP classifiers especially those in heavily regulated sectors like health that can be trained on relatively small, hard to get labelled datasets to produce highly accurate non-trivial results. Again super interesting, if this is you do get in touch.
Startups blending and structuring multi-modal data in order to apply AI prediction models to solve big problems in rapidly expanding markets led by fantastic teams with strong sales and marketing capabilities. Don't we all love these startups and if this is you then please get in touch.
So in short lots of opportunities for first cheque funds like Twin Path to invest in potentially great AI first startups. Finally there are some things that are definitely headwinds that AI startups need to overcome. They include:
The lack of Series A focused funds presently investing in the UK means that we are seeing a lot of startups that are out raising "bridging" rounds because despite good progress have not acquired sufficient traction or Product-Market Fit data to attract the very scarce lead for a late seed/ series A round. We are a first cheque fund with around a 1/3rd of our cash from SEIS investors so we at Twin Path will struggle to participate in these bridge rounds but it seems that most startup raising bridge rounds between £1-3M are struggling to find leads.
Pre-seed valuations that are just far too high in relation to the valuations that late seed/ Series A investors are prepared to pay, if and when the startup goes on, over the next 12-18 months, to achieve data points signalling product-market fit. When in today’s capital constrained market we see good applied AI teams raising bridge rounds at flat or down valuations on their seed or pre-seed rounds, despite reaching traction north of £500K ARR, it is still amazing to us that pre-seed, pre-product, pre-revenue applied AI startups regularly demanding valuations that are much higher. We will and have invested in the Advance Subscription Agreement rounds of amazing technical teams building breakthrough technology with wide and significant applications at potential high valuations (somewhat mitigated by the floor and discount price baked into the ASA) but the conviction hurdle for pre-seed investors in these cases is much, much higher. So my advice to most AI first early stage startups is to price their rounds accordingly to present market conditions. For if you want investors to back your team with a small cheque when you have little more than a strong hypothesis, (in order for all of us to find out if you can build and sell something that backs up that hypothesis), then for god’s sake give your pre-seed investors a valuation that will be 2-3x smaller than what they can presently expect to get on the next round if everything does work out swimmingly.
Overall we are happy with our mid-term performance but fully aware that the work only starts now. We are still hunting for at least 6 more teams to back with a good size seed cheque over the next 6 months, we will be launching in late September our regular Twin Path AI Start-up breakfast events and our quarterly evening network drinks and we are in the process of building out our added value post- investment support activities. One of the ironies of investing in early stage startups is that all the challenges, troubles and sleepless nights startup founders face, will become yours once you have invested. Bring it on is what me and my colleagues say.
Go to www.twinpath.vc or email john@twinpath.vc if you want to follow up on anything reported in the blog and especially if you are an AI-First startup looking for your first investor cheque.
