Selecting the Best Methodology for Your AI Project
Let's put Agile and Waterfall to the test.
Project managers don’t choose which methodology they will use for a project on a whim.
They choose ones they feel comfortable running a team with and have succeeded.
When tackling an AI project, you probably don’t have previous experience running that type of project.
If you have run software projects, you are probably familiar with scrum, Agile, lean, waterfall, Kanban, and extreme programming.
But there are so many choices - where do you start?
Let’s approach each methodology and see which is the best for our AI project.
We’re not in Kansas anymore
When I produced video games, the engineers were given a design spec and told to make this functionality in the game or application.
Engineers already had an idea of what they needed to build which is the opposite of an AI project based on data. In a data project, it’s like wandering into a library and determining what cool things you can choose from the data - you’re going right to the source.
There isn’t any app that’s acting like a librarian. You need to determine how AI will access and use the data.
Remember all the aspects of data that come with data project development:
Data collection
Data prep
Data security, privacy, and governance
Who owns the data?
Data quality
Let’s keep in mind as we explore these methodologies that when you undertake a data-based project, you need a data-driven solution and not a code-driven solution.
Don’t go chasing waterfalls
Waterfall is the granddaddy of project management methodologies, and it’s the sworn enemy of Agile project methodology.
The process is all-inclusive, so you start with project requirements and work through them until you reach the end. Waterfall is suitable for large projects with many set stages and rules - an example could be a construction project or building a spaceship.
One problem with waterfall is that you need to keep going forward in the process. You can’t go backward and redo the last step. This sucks on software projects since you can’t iterate or be flexible.
It’s also time-intensive, so if you have 18-24 months to complete all the stages, the waterfall methodology may work. But with AI and data projects you want them to fail fast, so you can deliver the next version very quickly.
Plus, testing is a nightmare - you can’t test anything until the last stage is complete which isn’t very flexible. Bugs could pile up, and the functionality of the software could be impacted.
Based on this, waterfall is probably not the best choice for our data-hungry AI project.
Agile is sitting in the back saying “Pick me! Pick me!”
Since we have discarded waterfall as a methodology, the logical choice is to turn to Agile for AI projects.
Agile has five popular methodologies – below are very brief descriptions.
Scrum: Scrum involves taking the work and breaking it into sprints lasting 2-4 weeks. The team meets periodically to review and adjust their progress.
Kanban: Kanban is centered around a Kanban board that has lanes. Tasks are typically moved in the lanes with cards for each stage of the process.
Lean: Lean’s primary focus is rapidly delivering the product/value to the customer. To provide this value – it eliminates waste and accentuates continuous team improvement.
XP (Extreme Programming): The core of XP is rapid iteration, feedback, and team collaboration. Fast feedback loops and solid code can be achieved through aspects of pair programming, automated testing, and continuous integration.
Crystal: Crystal is customizing the team and project development process. Experimentation, communication, and flexibility are used to find solutions.
Agile methodology provides a framework that allows project managers to:
Plan sprints
Define milestones and deliverables
Recap with retrospectives
However, with data there are a few gotchas in the Agile process. Data projects are more complex to define, which leads to the following problems:
Completion dates are hard to determine
Data projects aren’t similar to software development projects
Task estimation is difficult
Project scope can fluctuate
KPIs with data don’t easily translate to ROI
Agile is a viable option, but we need to adapt it to our AI data project to succeed.
What are these strange things?
There are three methodologies that use data at their core and have Agile integrated or will need to have Agile “upgrades”.
These are:
CRISP-DM (Cross Industry Standard for Data Mining) - This methodology is over twenty years old and has a framework of how a model may be tested, but not specifically for AI. It hasn’t been built specifically for Agile, so modifications would need to be made for the project. Even the website says “…CRISP-DM indirectly advocates agile principles…”
TDSP (Team Data Science Process) - This agile and iterative approach is the result of a joint collaboration between IBM and Microsoft. Organizations may be leery of using a vendor-centric approach, so this may not be a viable option.
CPMAI (Cognitive Project Management for AI) - This approach is vendor-neutral and has Agile methodology built into the core of the program.
Each of these is an iterative approach better suited for AI project development.
For example, if you play tennis, you can buy a racquet or a finely tuned professional-grade racquet. It’s just going to perform better, and it will make you a better player.
Each of these has a typical “path” that can be used for an AI project since the core of any AI project is data.
Business Understanding
Defines the business case, goals, and requirements
Data Understanding
Defines the data source for the project, methods of collecting data, determining data quality, and finding out what data is missing for the project
Data Preparation
Defines how the data is prepared for the project
Modeling
Defines the way the model will be built and the techniques
Evaluation
Defines how the model will be tested
Deployment
Defines how the monitor will be deployed, acts as expected and monitored
There may be a learning curve initially, but after you’ve used some of these data methodologies, you’ll see that these are more adaptable.
Final thoughts
Data is the key to any AI project, and the methodology choice should consider this.
This type of project has its risks, but choosing the way to execute the project shouldn’t be one of them. Any one of the data methodologies has extensive documentation and use cases. Your data scientists and analysts will thank you since they may have used these processes in projects before.
Who knows? Those data peeps can teach an old project manager some new tricks!
AI-Driven Tools for PMs
WaxWing - AI-powered project management tool.
Sanebox - An AI-powered email tool that brings sanity back to your inbox.
guidde - The generative AI platform for business that helps your team create video documentation 11x faster.
AI News PMs Can Use
Update on the Recall preview feature for Copilot+ PCs
Delays, Implementation Issues, and Unrealized Benefits Challenge Generative AI Initiatives in 2024
Cool ChatGPT Prompt for PMs
Team collaboration
Suggest collaboration tools for remote project teams.Additional Prompt: Recommend collaboration and project management tools for a virtual team spread across different time zones.
Become a Certified AI Project Manager!
CPMAI (Cognitive Project Management for AI) is the leading certification for professionals looking to master AI project management, providing a structured framework for managing data and AI projects, and ensuring your success in the rapidly evolving AI industry.
❤️ Get 20% Off CPMAI Certification - Enroll by June 30, 2024❤️
Elevate your career with Cognilytica’s CPMAI certification at 20% off using code AIPM20. Learn effective data and AI project management, gain insights from industry experts, and explore key areas like data governance and management, AI fundamentals, generative AI, and AI ethics. Engage with interactive modules and real-world case studies.
“I highly recommend Cognilytica's CPMAI Training & Certification - it's been transformative to my career. The quality is a 12 out of 10, and worth every dollar!” - George Fountain, Senior Project Manager at BAH and CPMAI Certified
Don’t miss this exclusive opportunity! Visit courses.cognilytica.com/cpmai and use code AIPM20 to save 20%. Offer ends June 30, 2024.
I recently joined The Write 4/28 Challenge with this crazy Australian I follow online - Tim Denning.
If you have been wanting to consistently post online or learn to get better at writing, this is the course for you, so I thought I’d pass the opportunity along.
It’d be a kick to do this together.