From their own preso: "AIM moat is the automated orchestration of complex workflows ... into AI engines frame by frame"

My initial impression that transcoding was at all part of the moat sounded silly to me, but I get the encoder is necessary to enable AIM product suite.

This reads to me like the the LEXI product is essentially a pipeline into using cloud platforms services (ie. amazon/google/microsoft transcription services). This sounds great for the cloud platforms but the pipeline itself isn't much of moat. Essentially, a product only useful for legacy broadcasters or smaller organisations, that is those without a mature tech stack. What benefit do big tech get from a partnership with AIM? Apart from the revenue, perhaps to help provide more training data. In a similar way to Appen, I can't see this continuing long term and long term the value add from AIM will shrink as the off the shelf cloud models improve.

The product still makes sense for the legacy broadcasters and other organisations (like government) with low tech maturity, which is still a great addressable market. But if my understanding above is correct, it will may be a low margin for AIM since they have to pay for the cloud services transcription models. How low margin is still unknown, it may be that this kinds of AI infrastructure becomes commoditised then the workflow itself should still be able to achieve good margins.

This characterisation of the workflow pipeline being the moat also misses an opportunity to describe the fine tuning of the models, that I'm sure they are doing. I assume they use the video and metadata to provide context and custom vocabulary with a ensemble setup where they weight different models depending on the topic. Now if they save and build this out for a large number of topics (different live events), this can allow them to deliver a product better than that off the shelf (AWS transcribe, for example). Even more powerful is if you are saving this training data - which AIM is not doing themselves but perhaps AIM products allow the customers to do so.

My other concern is if customers would accept lower quality in future as off the shelf models improve, it reduces AIM's moat to competitors. I see they define < 98% accuracy as bad, and that tracks my intuition. On youtube for live events, live captioning can be enabled (for free), and I think that's a single shot model using the audio stream only which achieves pretty good results, but would probably rank as 'bad' using the < 98 % as bad metric.

My broader concern with the broadcast segment is that global legacy broadcasting market share is declining and will continue to do so. Further, streamers like Netflix with high tech maturity and no need for AIM will move move into live events, eating not just the non-linear lunch but also the linear lunch of legacy broadcasters. If this happens, it will only happen slowly, and even after consolidation perhaps the public broadcasters will always be around. Essentially, AIM needs to help prolong dinosaur broadcasts life, a tough ask since they are competing against global hegemons like netflix, amazon and google, who are much higher capitalised and more mature with established distribution networks. Diversification of customers is important to see to reduce these risks.

In terms of another services other than live transcription that AIM can provide I am sceptical until I see uptake. In particular any non-live captioning would be low margin. Things like LEXI brew I am sceptical about use-cases but I like how they have not spent their own resource but partnered with someone else. LEXI voice does seem like it is solving a use-case for broadcast customers so I will be watching this one closely. It is key that additional products like voice do catch on for this investment to work.

In the short term, growing customers in other segments and cannibalising existing customers to grow LEXI still seems like a way forward for moderate growth.

Held.