AI is the Golden Age for Product Managers
For the past year I’ve been meeting with corporate executives and business leaders to talk about growing revenue with AI. Some of the most common questions concern talent acquisition and organizational change:
- “How do we find and acquire talent?”
- “Do our people have the skills necessary to launch new products?”
Good programmers and data scientists can learn these skills from a growing number of great educational resources.
The biggest job change — and one for which there is still little training — is the Product Manager (PM).
In this post I limit PM responsibility to product development activities — most PMs have many other jobs.
Training data is the new spec in AI
37 Signals’s (Basecamp) mantra of interface first has guided product teams for a decade. Mobile and consumer apps have created a huge demand for product managers with a design background.
UX takes a back seat in AI and the PM’s primary responsibility changes as a result.
Training data is the new spec in AI
All practical AI products use supervised learning to train algorithms that generate outputs from a set of data inputs.
What outputs? What inputs? Exactly — that’s what the PM needs to define. Machine learning engineers can only optimize for their training data.
In AI the product manager owns the training data
Many educational examples use free data from sources like ImageNet. This data is fine for initial algorithm training but achieving useful results requires domain-specific training data.
Since generating training data requires expensive human labor PMs look for ways to get the most of out of it. Every potential data source requires analysis.
Is this data complete? Do we have feature gaps? Are the labels accurate? Will it improve our coverage and lower error rate? Will our progress be CPU-bound when we add more data?
Answers can mean the difference between success and failure.
AI PMs need statistics and data science skills.
PMs can discover business opportunity
Most money-making AI products are currently based on research problems like image or speech recognition. Companies like Baidu and Google have limitless need for fundamental breakthroughs.
AI applications will increasing move from solving fundamental problems to business-specific problems; PM’s will play the key role in discovering these opportunities.
AI innovation is a creative process of hypothesizing solutions and testing them with data. Customers usually can’t provide a list of features or desired outcomes.
Unfortunately there is no easy answer to , “what can AI do for my business?”
Identifying opportunities starts by hypothesizing desired outputs and predictive input data.
Can AI reduce our customer response time by 40% by using bots to screen inquiries?
Can we double sales by dynamically serving the right content based on prospect behavior?
Can we use weather patterns to reduce supplier lead time by 10%?
PMs are in a unique position to ask these questions and hypothesize the answers.
Best practices for PMs are still emerging
We’re in the early days of AI and don’t have best practices for building complex systems. Even simple topics like technical debt are quite complex (pdf).
It will be an exciting decade for product companies and perhaps a gilded age for technical PMs.
