Veo Analytics: Coach Assist Q&A with Alexander Harvey-Falkboo

Coach Assist is a new feature in Veo Analytics that instantly generates talking points from your matches. Using advanced AI technology combined with your Veo match data, it delivers five key analyses: Match Summary, Game Plan Analysis, Scoring Efficiency Analysis, Trends and Patterns Analysis, and Training Drill Suggestions.

The feature works seamlessly within the Analytics Studio, where coaches can now see a Match Summary for each video as they browse. Whether you're just beginning your video analysis or not sure where to start, Coach Assist reveals points of interest and even suggests clips to share with your team – saving valuable time in match preparation and team feedback sessions.

We sat down with Alexander Harvey-Falkboo, Product Manager at Veo, to learn about the inspiration behind Coach Assist, the technical journey from concept to reality, and how this AI feature is transforming the way coaches interact with match data.

1. What inspired the development of Coach Assist?

I think this was a case of 'inspiration driven by friction of opposing forces'.

One of the challenges we face in our Analytics product is that the interface is full of information. Some people feel it's too much, or at the very least, they don't use everything available. At the same time, we regularly get customer requests for more. More tables, more charts, more metrics.

We became less convinced that people wanted more metrics, and more convinced that they wanted more answers, faster. What if you could open a video and receive an answer to the question: What are our key areas for improvement in this match? Stuff like that. We started discussing ideas like 'Smart Insights' that would somehow trigger in the UI when certain metrics crossed certain thresholds and other tactics like that but it was very complex, even just as a thought exercise. We couldn't really see how we could make it a reality.

Then in late 2022, ChatGPT was released. And it wasn't very good! But I remember reading something from Notion (B2B software), who were very fast to view this technology as a game changer for software and had strong opinions about where it could go. And just in general seeing how the world reacted to ChatGPT. From then on it became very clear to me that we had to pursue this path, that this is how a lot of software products would behave. (We were so confident that - in Jan 2023 - we shared our intention to work on Coach Assist.)

Fast forward 2 years and we are ready to build something that can act as a validator of theories, a catalyst for discussion, an idea generator and timesaver.

2. How does the technology behind Coach Assist work, and what technical challenges did you solve?

The Coach Assist version that you see today, that we call an 'Alpha' version, is actually pretty simple once you break it down. When you use it, you're basically starting a little back-and-forth with OpenAI's technology:

1. You trigger it – that's the moment you ask Coach Assist to help.
2. It sends a prompt – behind the scenes, Coach Assist has some predefined questions (we call them prompts) plus a bit of Veo match data.
3. AI generates a response – those prompts go to the OpenAI API, which returns a text response that tries to match your intent and the context provided.
4. You see the result – finally, that answer shows up right inside the Veo Editor, ready for you to review or use.

For developing the Alpha, the technical challenges were quite light in my opinion: making sure we don't spend too much every time someone uses the feature, making sure that we send enough data context so that an answer can be useful but not any data that is inappropriate, making sure people don't have to wait too long to get an answer, and so on. And I think we solved those things quite confidently. I think we knew we could build it, and were more concerned with 'would it actually provide good value?'

3. How did your team balance making AI insights both functionally effective and tailored to coaches' needs?

One key step for us was to start really small, and really basic. We started by talking to a handful of customers and actually just emailing them various outputs from various LLMs, based on Veo match data that we were manually inputting. Many thanks to Aaron, Daniel, Ian and Matthew for reading through some heavy emails! And we identified some things that we thought would be useful. Then we increased the number of people testing the product, refined, refined and kept refining (the interface and the 'prompts') until we got to the feature you see today. We also monitor feedback given on the responses and adjust accordingly. To give you an idea of how useful it has been to customers so far, the thumbs up/down ratio is significantly positive, 80% or 90% depending on which response you look at.

This kind of feedback is critical for a very simple reason: my team has no idea if a response is accurate or useful enough because we did not play in or watch the match in question! That's also been one of the trickier parts of building with this technology. That, and the fact that LLM responses are non-deterministic: you can pass the same match data and info, and ask the same question 3 times, and get slightly different answers every time.

But I want to be really clear that we are at the beginning of a path with a lot of potential. This is a very early version of Coach Assist with some early version flaws, that we only made widely available because the feedback was so positive. Now that we have that out and can get even more customer feedback, we can turn our attention to building a feature with more depth and more reliability, capable of providing even more value.

4. How do you envision Coach Assist contributing to the overall Veo Analytics experience?

A lot of the coaches using Veo are ambitious people who love the game and their team but have severe time constraints. They're not full-time professionals - often they're parents, volunteers, or semi-pro coaches juggling work, family, and football. For them, every minute spent on coaching tasks (like video analysis) has to deliver maximum impact because they are passionate about achieving success.

Coach Assist is going to help them squeeze as much value out of those minutes as possible. It combines sports knowledge with information retrieval and summarization. It quickly connects your Veo match data, allowing you to uncover opportunities and consume large amounts of game data, fast.

We have a very good idea on what the next Coach Assist capabilities will be, and will be announcing those more widely very soon.

Want to try Coach Assist for yourself? Learn more about the feature here.