AI/ML Product Management
80% of the ML Projects Fail.
https://hbr.org/2023/11/keep-your-ai-projects-on-track
https://www.kdnuggets.com/survey-machine-learning-projects-still-routinely-fail-to-deploy#:~:text=43%25%20say%20that%2080%25%20or,that%20their%20models%20usually%20deploy.&text=Key%3A,that's%20already%20been%20successfully%20deployed
Why?
Problem Framing
AI investment ROI
Human Resource
Compute Resource
AI Project expectations and inherent dissimilarities to software development
Is there a customer/user problem?
End-user pain points
Operational improvements
2. Is there any business value in solving this problem?
Classify the business value as Low, medium or high
AI can either automate or Augment. Automate when risk is low and augment when risk is high. Risk is cost of being wrong. Medical diagnosis - high risk and therefore augment. Tech content writing - low risk and therefore, automate.
AI can automate, presonalize, predict
Dont use AI/ML because
Everybody is doing it
Leadership said so
It’s a cool thing to do and we will have bragging rights
3. Can we solve this problem using ML/AI
Every problem can be solved by using heuristics (Non-ML) or Predictive AI or Gen AI.
For predictive AI or Gen AI solution, we have below questions to consider:
Classify as Easy, Hard, Impossible
AI can predict, classify or generate text or image. However, these capabilities are evolving fast.
B. Do we have the data to train AI in case of predictive AI?
4. What’s the ROI to solve this problem using AI?
High business value problem which can be easily solved with AI are the low hanging fruit
High business value problem which will hard to solve with AI is still worth pursuing
Consider the risk of using AI
Environment change - frequent environment change required re-ingeering of ML Model. Adapting to changes in environment for ML can be costly and difficult
Ethics - AI has no ethics or sense of wrong or right
Accountability - who is responsible for decisions taken by AI? Developers?, service provider?
Consider the following questions when comparing a non-ML approach to an ML one:
Quality. How much better do you think an ML solution can be? If you think an ML solution might be only a small improvement, that might indicate the current solution is the best one.
Cost and maintenance. How expensive is the ML solution in both the short- and long-term? In some cases, it costs significantly more in terms of compute resources and time to implement ML. Consider the following questions:
Can the ML solution justify the increase in cost? Note that small improvements in large systems can easily justify the cost and maintenance of implementing an ML solution.
How much maintenance will the solution require? In many cases, ML implementations need dedicated long-term maintenance.
Does your product have the resources to support training or hiring people with ML expertise?
Predictions vs. actions: There's no value in predicting something if you can't turn the prediction into an action that helps users. That is, your product should take action from the model's output.
AI Project Management is costly because
Need Wider set of skills/team
Have higher technical risk
Data needs and quality
Model limitations
Far more challenging to plan and estimate,
are not linear but experimentative in development
Probabilistic and not deterministic
Further work and experiments may not necessarily improve the model or accuracy
Harder to show progress
Variance of model outputs
Require more ongoing support
You frame a problem in ML terms by completing the following tasks:
Define the ideal outcome and the model's goal.
Example: Predict when an item will be delivered to the customer-> Predict a data range by which customer will receive an item
Identify the model's output.
Example: Min-Mac dates for delivery of an item
Define success metrics.
Example: Accuracy of the predicted date range when compared to the actual delivery date.
AI Projects:
Need Wider set of skills/team
Have higher technical risk
Data needs and quality
Model limitations
Far more challenging to plan and estimate,
are not linear but experimentative in development
Probabilistic and not deterministic
Further work and experiments may not necessarily improve the model or accuracy
Harder to show progress
Variance of model outputs
Require more ongoing support
Team Structure
Business Sponsor - Business strategy, goals and alignment because of high technical risk and project risk
Proagram Manager who leads the below team members:
AI Product Manager
Data Engineer
Data Scientist
ML Engineer
============================
In the realm of AI and machine learning, a staggering 80% of projects fail to achieve their objectives . For product managers, this underscores the importance of thoughtful AI product design and a solid business case. Here’s how to navigate this complex landscape effectively.
1. Assess the Need for AI/ML
Start by asking whether there is a genuine customer problem or operational improvement that AI can solve. Identify end-user pain points and classify the business value as low, medium, or high. Remember that AI can either automate or augment processes, depending on the risk involved. For example, in high-risk areas like medical diagnosis, AI should augment human decision-making, whereas in low-risk areas like tech content writing, it can automate tasks.
Avoid Using AI Just Because:
Everyone else is doing it.
Leadership demands it.
It seems trendy or offers bragging rights.
2. Evaluate the AI/ML Feasibility
Not every problem needs AI. Consider whether a heuristic or non-ML solution might suffice. For problems where AI is appropriate, assess the data availability and classify the problem's complexity as easy, hard, or impossible.
3. ROI Consideration
Evaluate the return on investment (ROI) for AI projects. High-value problems that AI can solve easily are ideal. Harder-to-solve problems with high business value might still be worth pursuing, but consider the cost of adapting to environmental changes, ethical concerns, and accountability.
4. Defining AI/ML Solutions
Frame the problem in ML terms:
Define the Ideal Outcome: What is the ultimate goal of the AI model? For example, predicting delivery dates.
Model Output: What specific results should the model produce?
Success Metrics: How will success be measured? This might include accuracy, precision, or user satisfaction.
5. Consideration of Costs and Maintenance
AI projects require significant resources and expertise. Assess the costs and long-term maintenance requirements, ensuring that your team has the capability to support the project.
6. Predictions vs. Actions
Ensure that AI predictions lead to actionable outcomes. There’s little value in a prediction if it cannot be translated into a useful action that benefits users.
7. Managing Complexity
AI projects demand a broad set of skills and involve higher technical risks. They are non-linear, experimental, and require ongoing support, making planning and progress tracking more challenging.
8. Team Structure
Assemble a diverse team:
Business Sponsor: Aligns AI efforts with business strategy and goals.
Program Manager: Oversees the project and manages the team.
AI Product Manager: Leads product development and strategy.
Data Engineer, Data Scientist, ML Engineer: Provide technical expertise and implement AI solutions.
AI product design and business case preparation require careful consideration of problem framing, feasibility, and ROI. By addressing these factors, product managers can improve the likelihood of success in AI projects, ensuring they deliver real value to the business and its customers.