Notes taken from reading the blog Moving Towards Managing AI Products.
The blog gives a brilliant talk on what mindset a product manager should have, as well as skill set.
1. Track how the market is using AI technology
Need to learn more than 160 use cases for AI across a variety of industries.
Be aware that only 12% has progressed beyond the experimental stage.
AI industry review from McKinsey Global institute, Gartner and CB Insights AI research:
https://www.cbinsights.com/research/artificial-intelligence/
https://www.gartner.com/technology/research/artificial-intelligence/
2. Follow the trends in deep learning research
An AI product manager's true competitive advantage comes NOT from expertise in the algorithms themselves, but his / her ability to shorten the time-to-market of products and services that apply those algorithms.
Materials to read:
- AI scientists from big time companies believe in having their researchers share breakthroughs
- The summarizing of the most recent research in AI
3. Cut through the AI hype -- focus on practical use cases
Be critical thinking to separate the hype from real- world capabilities and have insights into the practical use cases of AI.
Read the MGI report Artificial Intelligence the Next Digital Frontier, and AI practical use cases
4. Be obsessed with customer-centric data
Customer obsession requires going beyond features & benefits, and understanding the meaning for customer's jobs, their purpose, motivations and the conscious choices they make.
The two aspects of Data Obsession:
- being a champion of digitization while quantifying problems that customers care about;
- being able to build comprehensive datasets for building quality AI models;
Important: Urgent to Master:
Having an understanding of handling the data flows including data ingestion, data processing pipelines (Extract-Transform-Load) and data visualization tools helps in setting up the stage for building AI solutions that create customer value.
5. Build a usable product with a simple model before exploring complicated AI models
Don't be over obsessed with the complexity of AI models.
Additional to read: https://brianpagan.net/2015/lean-startup-mvp-how-to-make-meaningful-products/ https://twitter.com/jopas/status/515301088660959233
6. Iteratively build use cases where AI directly impacts metric
Use methodology that offers fast validated learning loops
Use hypothesize-design-test-learn
Not use build-measure-learn
The use cases should ultimately focus on creating significant value to the end user while improving earning, and should be tied to a small set of few metrics that matter to the customer.
In pragmatic implementation, avoid optimizing an end-to-end AI model for multiple objectives, but each optimized for specific metric.
7. Build Breadth-first(Data/Pipeline/Model) instead of Depth-first(AI Model)
An AI product manager should have familiarity with the tools and techniques used to create an end-to-end product that leverages AI. This provides them the ability to influence:
- AI Engineers and Data Scientists to utilize the right level of sophistication in their models, while still ensuring the ability to add complexity;
- Data Engineers to build robust systems and scale them appropriately;
- The entire team to leverage the appropriate cloud compute services and virtualization arthitectures.
The understanding requires a high-level awareness of the API ecosystem:
- Data ingestion tools: Kafka
- Data processing systems: Spark
- NoSQL DBMS: Cassandra
8. Ensure your product fails gracefully
Later
9. Insist on AI model explainability
Black box models lack interpretability that poses that involves liability in areas of law, medicine and safety.
Read the paper that summarizes two approaches to explaining the predictions of deep learning models.
Ensure there are no biases in the models.
Recommendations from SAP design center to eliminate the bias.
10. Establish clean communication with your teams -- know the fundamentals and language of data and ML
Read the Data Science Hierarchy of Needs pyramid.