Generative AI has shaken up the product development world. So, what’s the difference between the development of AI-infused products and that of traditional ones? And how can you build something in a crowded space that still outperforms the competition?
In the first part of a series on AI product development, we’ll look at:
The hallmark of - and most likely moat for - thriving AI products
How to define defensible AI value propositions
If I had to pick a pivotal moment in the development towards AGI, I’d pick ChatGPT. It wasn’t the model that mattered. It was the usability of it.
Sam Altman - CEO, OpenAI
The Hallmark of Successful AI Products
AI and machine learning are not new, nor are LLMs popularised through ChatGPT. Creating a sustainable and monetisable AI product with a defensible moat remains a significant challenge.
A year ago, Andreessen Horowitz explored various moats—models, infrastructure, and integrations — concluding that no systemic moats exist in generative AI. Despite this, the opportunity is undeniable, with 79% of respondents in a McKinsey survey reporting some exposure to Generative AI and 22% regularly using it. How can it be leveraged?
In an interview, Sam Altman gave a hint:
If I had to pick a pivotal moment in the development towards AGI, I’d pick ChatGPT. It wasn’t the model that mattered. It was the usability of it.
In areas with commoditised technology, outstanding user experiences offer a solid path to a defensible value proposition.
We’re not just talking about delightful design here. In practice, this means leveraging domain, technical, and customer insights to deliver seamless customer experiences. It isn’t too different from what we’ve seen with the previous generation of technology development. But getting it right requires a few more nuances and refinements to the ideation approach.
Defining Defensible AI Value Propositions
Many AI products are essentially features, not products. They address a task—like image generation—but fail to connect with real user needs and workflows. This generalist approach often leads to underperformance and attracts competition.
Consider a marketer planning a new social media campaign; their workflow will involve many steps, from setting strategic goals and defining briefs to souring creative assets, typing relevant captions, targeting specific audiences, scheduling a campaign, etc. Understanding these steps in detail, identifying where AI can solve particular problems, and exposing new capabilities in a user-friendly way necessitates deep domain and technical expertise.
By focusing less on generic solutions and more on specific user needs, you not only create more relevant products but also narrow your competition. However, this needs to be balanced with the decreasing size of your total accessible market.
To define a viable AI value proposition, the need to involve technical AI experts takes on an outsized criticality.
First, during process mapping, you may already consult domain experts and draw from user insights to define which pain points are most pressing to solve and what their potential business impact might be. Additionally, the AI experts in the room should be asked for a prioritisation of their own: where along the process is the most significant potential for AI to assist in never-before-seen ways? These lenses provide a heat map for prioritisation, considering impact and value, feasibility, and desirability.
Rather than solving these issues with a feature or new product, product teams should prioritise functionality based on what would connect best into a smooth E2E offering, effectively defining a modernised and improved incarnation for each step in the original journey map. As a16z wrote more recently:
Context switching is deadly. With AI, users should never have to break the flow of their work. Information, ideas, or examples (should) magically appear where and when needed. This might manifest in tools that own an entire workflow, like translating research notes to a final blog post and graphics. Or, it may be an assistant that "lives" wherever a user does work and can interject appropriate context.
This also means that while the solution may use AI heavily, in many cases it won’t all be AI. What’s more important is that it seamlessly integrates into users’ existing workflows and tech stack while augmenting and enhancing all of it.
When it comes to development, one new aspect is the management of unknown quantities alongside traditional product delivery. Let an R&D workstream run a couple of weeks ahead of product design, which is, in turn, ahead of engineering. Instead of delivering functionality, this effort solely focuses on answering feasibility questions and defining the technical approaches to solving a problem. If you find what you thought feasible isn’t doable or practical with available technologies, you’ll know so at the start of the design phase, focusing the solution space on what can be done instead.
Shifting focus towards connected user experiences supported by a deeply integrated understanding of technological advancements may just bring about the pivotal moment in your own product’s development.
Takeaways
To summarise:
AI solutions that address individual tasks without integrating with a broader workflow are cumbersome to use in practice, fail to achieve the desired uptake, and are easily replaced by competitors.
In a crowded market where few players monopolise foundational technology, owning a specific problem is the next best path to a moat.
Map out opportunities informed by deep domain expertise, insights on user pain points and workflows, and a keen understanding of the technology and its multitude of uses.
Construct an end-to-end value proposition that takes customers from zero to achieved outcomes. Don’t just focus on the AI piece; knit experiences together using the lowest-tech solutions possible. Augment existing workflows and routines and refrain from inventing new ones where possible.
Plan for R&D activities and manage them as part of the product roadmap.
I hope you’ve found this inaugural issue of Product Innovation insightful. Please share your thoughts in the comments.
In the following parts of this series, we’ll look at a case study, human-AI interaction patterns for the next generation of AI applications, and how to shore development up against risks.
See you soon!
Tom