How to build a roadmap in Machine Learning projects using Agile methodology in 5 simple steps?
I was recently agile coaching a team who’s aiming to improve our lives with Machine Learning solutions.
Here are the steps for a roadmap:
- Define your features.
When defining the features of machine learning projects, you need to keep it focused on the business benefits. The product exists to bring business value. These are the information you must provide :
- A comprehensive title that has a meaning from the end-user perspective.
- Description from the user perspective? What does it bring to the user? Here you can include benefits, feature description, mockups, the business need that this feature will resolve, how the users will be using the functionality. You may also include the steps to enable this functionality. Working with Agile methodology, you want to provide value early and to allow the investors to start chasing in early in the development process. The big challenge for Machine Learning projects is to talk in simple and meaningful words for the investors.
- Technical description of what you have in mind to implement. I often see the situation that everything is evident during the planning, and it becomes complex when developing it. For this situation, I have a solution: write a summary of the technical solution you have in mind. If writing is not your strength, draw it on the whiteboard or record the conversation you have with your team. You’ll appreciate all this information later on.
2. Measure the work. Uncertainty and poor predictability is a reality for machine learning projects. But how do we measure work in Artificial Intelligence projects?
- Define the end to end workflow — from the requirement to make it available for the user. Here is an article I wrote on this subject.
- Define the end to end-user stories. The stories must provide user value and be independent. The output of the user stories in machine learning projects must be ready to be released to the users in case the Product Owner decides.
- Add a short description to capture all the information you have in mind during the roadmap planning of the machine learning project. You do not need to add acceptance criteria at this stage, as you do not know which features will be approved.
- For each story, consider the EACH step in the workflow. In most of cases, a user story will have data algorithm adjustments, preparation, and annotation, training, testing, deployment.
- Add dependencies — the elements that the Agile team needs to start working on the user story or to complete a particular user story.
- estimate the user story
3. Add the business value. Every project I run and every team I agile coach, I see this step done differently. It is not important how you do it, as long as it makes sense for your business, and is in a numeric format. That’s all we need for the road mapping exercise. For a real result of the portfolio roadmap, the business value independent of point 2 — the measurement of work. We are interested in knowing how important it is for your clients and your business this particular feature.
4. Rank the features. It is my favorite step in doing product road mapping. It’s the moment to bust all the myths. There is not a person who’s right or wrong, but we all find out where we should focus our energy first. HOW? Divide business value my work and then order descending. The higher number of this equation is what the team must focus their energy on. These are the features that will provide to the machine learning project stakeholders the maximum benefits with a minimum of effort.
5. Assign Stories to Sprints based on the feature ranking. Start to split in sprints from the stories belonging to the feature, which ranked the highest and keep on going until you loaded all the estimated stories. While doing this, exercise takes into consideration the scrum team capacity.
You might be interested in:
