Scrum methodology

- [Narrator] So let's look at how the teams are going to work together to create this minimum viable product. In the AI projects that I've been involved in, the minimum viable product is well defined and small, and so it takes between six to 10 weeks. I found that this isn't so short that you can't get enough work done, but not so long that the team are starting to get tired. The project is broken down into sprints and these could be weekly or fortnightly. Now for a product and project new to the organization, I think a weekly sprinT makes more sense. A weekly sprint is short enough for the product owner and the data scientists to accommodate changes in business requirements, and it's long enough to get a good chunk of work done. At the start of the sprint, the product owner provides the priorities and the features and this goes onto the product backlog. This is just a list of all the things that need to be done within the project. The sprint backlog are items taken from the product backlog that the team will work on for the sprint. The scrum team work together with the scrum master to plan out the sprint and break it out into smaller tasks. And they decide what work will get completed for the sprint with a review at the end of the week. Daily scrums are check-ins for the team members to seek help, talk about success, and highlight issues and blockers within the project. At the end of each sprint, it's a good idea to have a review or demo to the product owner to showcase what has been done. You don't always have to have this at the end of a sprint, perhaps you might have a demo when you've reached a significant milestone. Now sometimes what works really well is to not only demo to the product owner or the scrum master but include the management team and the platform owner as well. You want to introduce AI to the organization, right? Then show them how you get there step-by-step. The more people who get involved and see this work, the greater the acceptance of AI into the organization. So if the scrum team have spoken to the platform owner several times during during this process, the platform owner starts to understand how they're solving these problems. You see it builds acceptance of the final product, because the platform owner has been a part of it the entire way. The team retrospective at the end of the sprint is to understand what went well and what didn't, and to provide this feedback to the rest of the scrum team. This way the team is continuously improving. The feedback from this retrospective helps provide some context for the product backlog for the next sprint. It's important that the product owner and the management team understands that when you're looking to build an AI minimum viable product over six to 10 weeks, the focus in the first couple of weeks is around accessing the data, cleaning it, and getting it into the right format and data exploration. It's only later in the project that AI models are created, tested, and experimented with to improve their accuracy. The last part of the sprint is typically dedicated to moving the AI model to work in the current production environment, and then monitoring the model in production for any degradation. Now based on my experience, a rough rule of thumb is 40% of the time is spent on accessing, analyzing, cleaning, and data exploration. 20% of the time is developing the AI models. And finally, 40% is spent not only deploying the models, but monitoring them in production, and retraining and redeveloping models where necessary. The monitoring of the AI model doesn't end at the end of the six to 10 weeks, instead the data science team needs to keep monitoring the performance of these models as working on real-world data can degrade models. Given that these models are in production, they can have a real impact on the business. So any model degradation needs to be dealt with quickly and effectively.