AI Product Development Lifecycle (AI PDLC)
Unlike increasing adoption of AI by businesses, there is not a standardized Product Life Cycle or framework for applying AI solutions. Therefore, if you are a PM considering to apply AI solutions in/on your organization/product, I’ll go over a general approach/PDLC in this post that I found helpful.
As a dead business cannot revive with AI solutions, an AI Product Manager also cannot lead an AI Product Development Life Cycle without core Product Management skillset such as having proper industry specific domain knowledge, AI solution understanding as well as having proper interactions with customers to recognize their pain points and keeping them engaged, establishing business partnerships and communications, helping sales and marketing on creating stories and get feedback (market opportunity), knowing company’s and product’s vision and staying align with them.
Recently, I attend a class called “AI Product Management” in Stanford university taught by Adnan Boz, Lead AI Product Manager at E-bay and here’s a quick takeaway of my learnings around AI Product Development Lifecycle:
Design sprint:
Each AI solution requires Infrastructure, Framework, Algorithm and Model to be defined. During design sprint in AI PDLC, team(PM, Marketing, Sales, …) looks at the problem from customer perspective and follow the below-mentioned steps:
- Break down the problem and pick a specific piece to focus on → this defines the core business value proposition or “sound business strategy” (where you don’t need to apply AI just yet)
- Wireframe or so do speak sketch ideas/designs on napkin!
- Turn ideas into testable hypothesis.
- Come up with high fidelity prototype.
- Test the prototype in real world and validate the idea.
As you would guess by end of Design Sprint (5 days), you’d have a viable prototype and all the ambiguities associated with the original idea would be resolved.
If you are a PM, it’s better to make sure you are leading the Design Sprint with discipline and following a robust framework to have efficient outcome.
Business Requirements and Data Analysis:
In this step, first, you’d gather information about business values, objectives, data stories and find out if there are any flaws need to be addressed. Then, you’d study to understand organizational priorities and trade-offs, risks involved, understand impact of the AI solution(aka orthogonalization) and its performance. Finally, you’d organize and prioritize information and define KPIs(eg: S.M.A.R.T criteria), metrics (in 3 categories business, product and platform levels) and the optimization objectives.
AI solutions and experimentation
Depends on the type of solution you’d be considering, research phase would greatly vary. However, research follows a general lifecycle along wit a rapid experimentation which require Product Managers and Data Scientists contribution. Here are two simple schematic:
PS: Don’t ever tell technical folks what to do(that would be end of relationship with them). Give them the problem, they’d take care of the solution.
Build and Release
To fulfill build and release, goals, requirements and limitations must be clearly defined along with roadmap, risk management(follow a framework to go over identifying, assessing, controlling and monitoring the risks), release plan and test plan(QA).
Evaluate
Once Data Scientists (DSC) provide the solution and models, you as a PM would run some empirical experiments against your KPIs to evaluate the proposed models. There are a lot of experimentation approaches such as A/B testing, multi-funnel testing, multivariate , etc.
It worths mentioning that these steps are not applied in a sequential manner. There are absolutely lots of overlaps, back and forth and iteration that depends on the AI solution would vary. In general, Even though there are some key differences between AI PDLC and general PDLC, I believe they all follow the same core values.
Finally, what I mentioned in this post was just a tip of the iceberg compared to what we covered in the class; however, I would say the main message was that AI is nothing but experiment. As a PM, you should understand that AI would empower your business ;however, it doesn’t change the core value of your business; therefore, don’t expect to revive a dead business by applying AI! I’d like to end this post by a quote from Phil Frost emphasizes on importance of sound business strategy in digital marketing domain:
“Success with digital marketing is the result of sound strategy, not perfectly executed tactics”
