Working at a global accelerator, I came across many product ideas across various industries like Fintech, MarTech, Cyber Security, Advanced Manufacturing, Health Care, and Construction. Our company was sector agnostic, and the only mandate was to focus on technology-driven startups, which means that the startups need technology as their point of difference to set them apart from competitors. While we did not initially focus on AI, it was interesting to find that AI became a common theme amongst our startups.
AI is becoming ubiquitous amongst startups these days. CB Insights reported that investments into AI in 2019 almost tripled from three years ago, a growth of 22% per year. Startups almost have to include AI into their pitch decks to get a second look by venture capital companies. Venture capital themselves are using AI to analyse potential investments, and Gartner predicted that by 2025, more than 75% of venture capital would use AI as the first screen for early-stage investment decisions.
AI is sector agnostic and is likely to make it into every industry. Just like digitising was the buzzword in the last 10 years, AI Transformation would be a common term in every corporate meeting and taxi conversation. According to McKinsey (Figure 1), many industries have become highly digitised. The level of digitisation has increased significantly due to COVID, and many industries have completed their digital transformation. As more and more processes become digitised, so too the volume of data collected. Corporates will inevitably face the question of what to do with this data. The ones that survive will use this data to generate valuable insights, automate key processes, and create better customer experiences.
Figure 1 — Level of digitisation by industries in 2015 by McKinsey
A few questions would arise: Who should manage this process? Could the traditional Product Managers manage AI solutions? Where do we start?
AI Product Manager is not a well-understood role in Australia. But in California (I guess most of the searches are from Silicon Valley), it is a growing area of interest.
Figure 2 — Google search for AI Product Manager in California
The question remains, do you need an AI Product Manager?
Certainly, there are many differences between a traditional Product Manager and an AI Product Manager, like the differences in processes, productivity tools, team mentality, and costs. I have listed out four reasons below why any company that is seriously considering AI solutions would consider AI Product Managers that are dedicated to AI products.
AI Opportunities
Contrary to popular beliefs, AI can’t do everything, at least not yet. It takes a certain level of understanding of AI and how it works to determine the problems that AI could solve. While AI can solve specific problems, others require human-in-the-loop or are still better to be solved entirely by humans.
With that understanding, an AI Product Manager can look across the entire organisation and talk to customers to determine automation opportunities or improve customer experiences.
At face value, this might sound quite simple until you realise that AI is growing rapidly. Moore’s Law observes that processor speed doubles every two years, but AI computational power doubles around every six months, or approximately four times faster than Moore’s Law. Stanford University found that an AI-based image classification that took 13 days and $2323 to train in 2017 fell to just seconds and cost slightly over $12 in September 2018.
A company would need someone to constantly monitor the rapidly changing AI capabilities and decide if an AI solution that wasn’t possible six months ago could be a solution to meet an internal or external problem.
AI products could also be tailored for individual users to provide a personalised customer experience. Traditional Product Managers typically work with a user persona and define a single product that would be most suitable to meet the requirements of a group of users. This fundamentally changes the concept of products, and a Product Manager will need to understand how AI works to develop appropriate solutions, e.g., Google ads recommending relevant advertisement after analysing your user profile and search history (that is why all that sports ads come up after you searches for running shoes)
Experiment vs Build
AI solutions are inherently unpredictable and could require multiple rapid experiments before it is ready for deployment. Traditionally, Product Managers spends a lot of time defining the requirements for a solution and getting the Engineering team to build the product. An AI Product Manager would need to identify the problem that AI could solve and work together with the Data Science team to search for a solution.
People have different (sometimes unrealistic) expectations for AI products. Traditionally, it is expected that products “just work” as the processes are defined business logic. However, searching for an AI solution is an interactive process, and with each iteration, you might get better or approaches a baseline expectation.
For example, a factory may want to implement a computer vision solution. The owner may expect the model to detect defects with 100% accuracy. But there are many limiting factors making this expectation unrealistic (e.g., the lighting condition, the quality of the camera, definition of a defect). The AI Product Manager would need to quickly establish baseline expectations — human-level performance or other similar models — and work interactively to come close to or exceed that baseline.
One problem with AI experimentation is that a company cannot just expect the Data Science team to develop solutions that improve a company’s growth. A Data Science team is good at interactively improving machine learning metrics, but there is no guarantee that this model would improve business metrics. AI Product Managers need to balance critical business metrics and tie that back to machine learning metrics for the Data Science team to improve.
Figure 3 — Business metrics to machine learning metrics
Compliance and Ethics
AI unleashes an entirely new set of regulatory and ethical questions, and the current Regulations and Standards are just playing catchup. For example, machines are not good at deciding whether a vehicle should crash into a cat or a person. AI Product Managers need to monitor relevant regulations and the environmental impact of AI decisions to ensure that the company does not breach any rules or code.
Another problem is that many deep learning algorithms are black boxes and could be a problem. For example, I know a financial institution that refused to implement an AI model over an excel spreadsheet because the excel spreadsheet was more straightforward and the decision-making process was more transparent. The AI community as a whole is improving on making models more transparent, and the AI Product Manager should be aware of these commercial challenges and come up with a solution.
Buy/Build/Rent/Invest Decisions
Finally, an AI Product Manager is in an excellent position to advise on the buy/build/rent/invest decisions for AI capabilities. Not only are they more familiar with the hidden costs for building a specific AI solution, but they are also more aware of off-the-shelf solutions and how they could be used to reduce the cost and time to market of an AI solution. Furthermore, an AI Product Manager will better understand the AI solution that the company is buying or investing in.
Conclusion
AI Product Managers’ role is not currently well understood, especially in Australia, but their need is expected to grow dramatically as AI becomes ubiquitous. Any company considering incorporating AI into their strategy should seriously consider an AI Product Manager who can devote his or her time to monitor the changing AI landscape and manage AI product solutions efficiently.
