Key takeaways:
-Understand the current market for enterprise AI products.
-Learn about the framework for designing enterprise AI products.
-Be able to capture requirements for complicated AI use cases
5. How to build enterprise AI products
Ying Fei, Product Lead of AI & Industry Solutions @ Google Cloud
6. Speaker profile
● 2017 - present Google Cloud
Product Lead for AI & Industry Solutions
● 2013 - 2017
Tech Lead for Google Payments/Ads
7. Disclaimer
The deck is created to summarize and share my personal learnings. It does not
represent Google or any past employers.
8. Enterprises will be transformed by AI in ~15 years
● Enterprise AI grew over 270% over the past 4 years1
● Enterprise AI market will reach $53B by 20262
● Contribution of digital workers grew by 50% in two years3
9. Technology is ready. AI becomes more and more intelligent
Google’s AlphaGo AI wins three-match series against world’s best player.1
10. But ...many traditional enterprises are not ready
● Legacy work environment
No IT infrastructure set up, costs a lot to upgrade, etc
● Labor intensive work processes
Lots of dependencies to manage to improve automation
● Industry ecosystem is not ready
Downstream and upstream processing are blocking digitization
11. Focusing only on model training requirements is not enough
for designing successful AI products
Only a small fraction of ML systems in real world is composed of ML codes1
12. Key elements for designing enterprise AI products
Business
Workflow
User Persona
Production
Environment
Data
Availability
Data
Annotation
Model
Training
Model
Evaluation
Deployment
& Scale up
Requirements
Product
Design
Partner Ecosystem Direct to Customers
Go to
Market
13. Learn about business workflow
Questions to figure out
● How does this workflow generate business value?
● What are the different steps?
● What are the preconditions and dependencies for
each step?
● What are the challenges in each step?
14. Look into each user persona
Questions to figure out
● What are the roles involved for completing the
workflow?
● What are their responsibilities?
● How will AI products change their work?
15. Investigate production environment
Questions to figure out
● What are the devices being used by customers?
● Will they be willing to purchase new ones in order to
adopt AI solutions?
● Is there any limitation for legacy devices?
16. Understand data availability
Questions to figure out
● What is the data that could be collected?
● Will the customers be able to create labeled
data?
● Is there any limitation for data sample?
17. Output of requirements analysis
Requirements summary
● User persona
Shopper, shipping requester, approver, buyer, supplier
● Problem statement
Improve the efficiency of order fulfillment process by automating information collection process to reduce
manual input
● Production environment
Server, network bandwidth, legacy logistic system
Sample use case: logistics management AI
Logistic management workflow
Shopper Requester Approver Buyer Supplier
-Added catalogs,
Forms
- Enter org code,
shipping location
-review and confirm
shopping doc and
shipping location
-Approve the request
-confirm org code
-Review price and
items
-Select and send to
supplier
-Fulfill the order and
ship the products
25. Get ready for the launch!
Business
Workflow
User Persona
Production
Environment
Data
Availability
Data
Annotation
Model
Training
Model
Evaluation
Deployment
& Scale up
Requirements
Product
Design
Partner Ecosystem Direct to Customers
Go to
Market