Product Management

Managing AI Products

By Stratrix Staff Writer

Boston, MA | Updated 30 Mar, 2021

Managing AI Products

Managing AI Products (or artificial intelligence product management) is somewhat different than traditional product management and more akin to in-depth science product management or R&D product management – at least in the current state of evolution of the discipline.

Managing artificial intelligence products is the art and craft of shepherding an embryonic idea focusing on leveraging AI, Machine Learning, and other Cognitive technologies to become viable and valuable products and platforms and manage the lifecycle after that.

Managing AI Products

Why is product management of artificial intelligence products and platforms different than the traditional B2B Product Management or B2C or B2B2C product management? Well, for one given the nascency of these technologies, there is going to be a lot of hits or misses and pivots. Solving consumer and business and industrial use cases is fraught with scaling lab ideas, and hence it poses a different set of challenges. Unlike in other instances in which one is using standard tools and programming paradigms to automate or transform a business process, the level of cognition and machine comprehension make product management of artificial intelligence platforms a lot more dependent on the “technical” folks (expert programmers, data scientists, linguists, neurologists, mathematicians and the like.)

Perspectives on Managing AI Products:

If you are involved in managing artificial intelligence product development, here are seven factors to consider as a part of your concept to completion journey.

Chicken or the Egg?

Do you have a use case in search of an AI solution or have an AI technology concept in search of an industry use case? This is a foundational product opportunity assessment question that a product owner managing AI products must ask and seek to find out.

The chicken and egg scenario is familiar with emergent technologies and concepts. Sometimes there is a specific use case for which there may be an AI/cognitive solution.  For example, the driverless car is a use case many technology giants are trying to solve.  This use case is well known, but how to actualize it is challenging and may involve many AI and cognitive technology constructs and may even necessitate further fundamental research.  Likely, all the current state technologies may not be sufficient, and the project may have to wait until something new, which no one is thinking about today, comes around in the future.

On the corollary, one may have a concept in Artificial intelligence and a technology breakthrough and is looking at industry uses to apply the same.  For example, if someone has made improvements to machine learning in pattern recognition, there are a lot of potential industry use cases.  For product managers, figuring out the applicability, sizing of the opportunity, and technology/market/solution fit is a challenging task.

Product or a Framework or a Platform?

Another challenge for a product manager in artificial intelligence product development is the extent to which one needs to productize the technology constructs.  We all know productization with some configuration options helps foster reuse and easier deployments at multiple clients. On the other hand, if the concept is in its nascency and the future path, scope, and use case range is unclear deeper productization may limit the extensibility and evolvability of the solution.

For example, your product framework has advanced machine learning chops in gleaning insights and meaning from unstructured data. Do you take the leap of faith and build out a product framework for creating cliff notes from scientific journals? Or perhaps a summary of vacation destinations based on millions of reviews?  A leap of faith into one arena may be fraught with future regrets as the use case may not work well in mass scale, or the business model may not provide the market value your technology underpinnings deserve.  So, in this particular machine learning technology example, a loosely coupled platform components that can be customized to different use cases and business scenarios may be the right approach.

Let’s take a slightly different example. Your neural network algorithms have consistently out-performed the stock market indices. In this case, the more productization and domain features you include in the product, the better.

XAI (or Explainable AI): 

Hitherto, the conventional wisdom of product management is to obfuscate the intellect so that no one can reverse engineer and steal it (or, of course, obtain patent protection).  However, in the case of artificial intelligence products and platforms, there is an innate need for openness, transparency, and explainability so that the AI does not become a Blackbox.  This explainability is a pre-requisite in the B2B space as the corporate customers are keen on knowing the inner workings.

So, as a product manager of AI products, one has to allow for tractability and traceability in every step so that it is easy to demonstrate why the machine has done what it has done and how it has done it.

A Reliance on data sets and knowledge graphs:

Managing AI Products - knowledge graphsThe AI algorithms and results are as good as the data that they rely on to generate the inferences and deductions.  In most industries and corporate settings, the data is somewhat limited, and also the data interrelationships at a conceptual level are not mapped out.  The lack of reliable data sets and rudimentary knowledge graphs (or complete absence of them) is a known industry-wide obstacle to higher efficacy in results.

For example, if an algorithm that is predicting next best action in a financial advisory space in the wealth management sector relies on only one transactional data limited to the firm, and does not take into account allied external data and the interrelationships and cross-impact thereof, the results will be slightly skewed given the constricted perspectives.

Guarding against bias:

A continuation of the previous point is the need to guard against the implicit and explicit bias that one may introduce into AI products and platforms.  The introduction of such bias in AI systems may precipitate through reinforcement learning and self-learning – all due to the fact the seed assumptions have biases, however small they may be.

For example, in an autonomous driver situation, not accounting for the fact that a part of the world drives on the left-hand side will result in a bias that can be deadly in such circumstances.

Or in the case of the financial services industry, not accounting for Black Swan events and training an algorithm on bull market data will instill bias and yield disastrous consequences.

Minimally intelligent product:

The move toward agile product development, the lean startup model, and the desire for continuous integration is generally excellent practices. Still, in the case of AI products and platforms, a traditional MVP (Minimum Viable Product) may not be advisable.  Without thinking through a comprehensive and coherent set of use cases and scenarios with a reasonably representative data set, the solution may come up far short of expectations and may not scale for real-life deployments.

One way to mitigate the risk of a barebones product lacking real brains to conceptualize a MIP (Minimum Intelligent Product), which can be a standalone and robust representation of core capabilities – albeit in a limited and often controlled environment.

Dealing with “Doctors”:  

Perhaps similar to a traditional deep science field and R&D space, the number of Ph.D.s or other experts galore in AI firms. A product manager often is not a deep subject matter expert. And the deep SMEs have trouble articulating the technology concepts into a coherent product construct. Furthermore, evaluating the readiness of a science experiment to translate to a commercially viable product is more of an art than science.

Product managers have to collaborate, coordinate, and coax the right information of the well-meaning experts. The latter typically do not have the discipline necessary to take a concept to market in a way that is replicable, scalable, and provable.

The delicate dance that a product manager must do to manage AI products shall be intricate and may make a ballerina proud.

Of course, managing artificial intelligence product management is not impossible. It just takes a non-traditional approach by a savvy product manager.  The product manager must be an “impresario to orchestrate a workshop of wizards.”

As you can see, managing AI products is different and requires a product manager who is up to the challenge. How does your firm product manage AI platforms? Are we missing anything in our approach to product management of AI products and services?