I continually speak with clients about Machine Learning and Artificial Intelligence (ML/AI) projects as everyone looks to harness the promise of these powerful technologies. From dozens if not hundreds of conversations, I often find these questions steer ML/AI project leads to a successful path as they work with colleagues, management, and vendors - all important constituents to support and with whom to cooperate for a successful outcome. A little planning upfront goes a long way for improved efficiency, shorter project timelines, decreased costs, and most importantly higher and sustainable ROIs.
VISION AND FOCUS QUESTIONS
- What is the focus of the project??Can the problem definition for the project be narrower??Is the current focus too broad? (Hint: If it is hard to measure, it is too broad)?
- What?is the ideal?end state or envisioned workflow of the final solution?
- What are the measures of a successful project outcome and for the different stages of the project? Does everyone involved understand these measures?
- Does the current workflow have some dependencies that need to be considered?
- What deadlines are important to meet??These can include commercial deadlines tied to clients, corporate deadlines tied to reporting or financial planning, or project deadlines tied to priorities and projects of technology team.
- What is the target budget for this project both overall and its components??Communicate budgets with all relevant parties when applicable.
- Not all ML/AI components need to be designed in-house. Much like airplane or car manufacturing processes, subcomponents can be outsourced and the final design be the true IP of your company.
- What is the workflow you are going to improve, replace or augment?? Keeping the final (internal or customer) users in mind, what user experience are you designing for? The details around this are critical to make sure the final solution design meets the needs of the stakeholders.?
- What unforeseen variability could show up in the current workflow that may not be obvious??Those rare but often ignored events have to be part of the final solution design.
- What is your preferred cloud deployment solution?
- What internal solutions/tech stack components does the ML/AI solution need to interact with??In other words, what are the upstream and downstream processes.
- What data cleaning and data storage needs does the project require?
- Do you have enough training data??Do you have the right training data? If not, where can you get it from (there are many training data marketplaces now).
- Do you have the expertise and talent necessary to train the model?
Whatever the purpose and reasons for your projects, these questions should help you get started on the right foot. Of course, this is list is not exhaustive and your teams may have more questions worth asking before kicking off the project.
Good luck on those projects! If any of this resonated, Link with me.
Head of Asset Management at Abra | Columbia Business School.
3 个月Vikas, thanks for sharing!