Product Management in the Age of AI
(First published on https://www.linkedin.com/pulse/product-management-age-ai-avinash-singh-pundhir)
“Artificial Intelligence threat to humanity”, “Amazon Alexa is coming”, “How AI is powering retail”, “Nvidia’s earnings rise with booming AI business” and “Wells Fargo special group to focus on AI”.
These are few headlines that I have gathered in last 24 hours about Artificial Intelligence (AI). We are rapidly moving in an era where AI will transform most if not all industries. It is evident that AI will soon take its rightful place in the club of General Purpose Technologies (GPT) such as steam engine, electricity, computers and internet. Economic historian Gavin Wright defines GPT as:
“ Deep new ideas and techniques that have the potential for important impacts on many sectors of the economy.”
For a technology to be termed as a GPT it, should spread across many industries and result in significant productivity gains. Prominent examples of GPTs includes steam engine and electricity. These technologies have revolutionized many industries such as transportation, manufacturing, textile, logistics etc. In addition, over the course of last two centuries these technologies have increased productivity, generated substantial economic wealth and transformed societies.
GPTs and Productivity Gains:
An exploration of productivity gains achieved by the implementation of GPTs displays an interesting trend. When electricity was first used in factories in late 19th century, there were no immediate productivity gains. But, significant productivity gains were reported only a few decades later. This lag can be attributed to the fact that post implementation industries took a while before they adjusted their workflows to truly harness the potential of electricity. At present, we see the same trend for AI. Though AI has made a significant progress in the last decade. However, the practical applications that result in significant productivity gains are not visible.
Challenges in AI-Driven Development:
AI plays a direct part in the evolution of at least 10 different emerging technologies in the Gartner’s hype cycle of emerging technologies, 2016. Google, Amazon, Microsoft, IBM, Apple and Netflix etc. have shown hundreds of possible applications. However, if we look beyond this initial euphoria, we can identify numerous half-baked products. Either these products do not integrate AI in their solutions at all or the AI applications they offer often fail to engage users.
One core contributing reason that hinders AI based development in Information Technology (IT) sector is the established development practices and team structures. In IT industry product manager acts as a link between users and engineering teams. IT products are designed to fulfil explicit and specific user requirements. In most cases, users are highly certain about the solutions that they are looking for. Product managers rely on requirement gathering approaches such as interviews, observations and secondary research to gather requirements. Later, these requirements act as requirement specifications for downstream development teams. During, project execution the primary focus usually is to minimize failure or restrict any deviation from the requirements. Processes are designed to ensure a strict adherence towards quality to minimize rework. This approach is effective when users have specific requirements and expectations from a system and want to follow a well-defined workflow. Though, this approach is not effective to execute and effective AI-driven development.
Embedding AI based abilities in the products pose many interesting challenges for product managers. Such as:
- Limited market data: Most AI implementations at present are first of its kind with not enough market research to demonstrate value. In many such scenarios, product managers have to deal with a lot of uncertainties and ambiguities.
- AI Products can only be as good as Underlying Data: Effectiveness of AI solutions is based on the quality of data that is fed to the AI engine during the training phase. In many cases, existing products either do not capture or store data that can produce high quality of input data for AI development.
- Failure Avoidance: AI solutions require a good amount of experimentation and tweaking before they can be effective. This requires multiple rounds of experimentation and continuous learning from previous failures. Traditional, development cycle followed in technology industry discourages failure. This failure avoidance reduces the power of technology teams to experiment and iterate. This puts product managers in a difficult spot. Where it is very difficult to justify this new experimentation, failure and learning based development model.
- Lack of Infrastructure to Gather Feedback: AI deployment and effective evolution of AI solution requires a lot of feedback and post implementation user engagement data. Retrofitting existing products to support user engagement data capture requires a lot of resources and investment. Due to this, we see many half-baked AI features in products that do not evolve post first implementation based on user engagement and feedback.
Finally, traditional domain, competition and feature based market research in many cases does not offer detail insights about AI features. AI product managers must explore possibilities, synergies and opportunities across industries. They should constantly look out for opportunities to embed small and easy to deploy AI features in their products. They should explore for new ways of gathering user feedback and user engagement data to identify the effectiveness of their implementations.
In the next post, I will discuss some approaches that AI product managers can use to support effective AI driven development.
Please share your thoughts, suggestions and ideas in the comments.
