Designing AI products to succeed
If you’re a current or past parent of young children, you’ll understand the concept of “January 3rd Gifts”. On Christmas morning, the kids rip open packages of toys with glee (this happens in millions of households regardless of religious affiliation or geographic location, sometimes with a change in holiday & date). By the time the kids return to school, a subset of these toys get placed in a closet during the grand cleanup. Many toys spend most of their time in that closet until a spring cleaning or household move sees them transitioned to a box and eventually Goodwill.
During the AI Goldrush of 2024, many companies will create “January 3rd Products”. They will be exciting new AI-powered experiences that look wonderful with their bright, shiny bows. Usage will drop off, expensive servers re-tasked for other purposes and eventually, these products will get uninstalled. There’s no Goodwill for unloved software. The companies who developed these products will quietly write off the investment, and possibly reorg to shift the troublesome leaders out of harm’s way.
If you’re tasked with developing one of these AI-powered products, how can you reduce your risk of creating a “January 3rd Product”? What do you need to get to the daily active user metric that brings massive revenue gains? The AI equivalent of Legos (vs. Pet Rocks, Tamagotchi, or Bratz)? The answer comes down to product development fundamentals.
Legos endure. Tamagotchi - not so much.
AI-powered products serve a user (or fail to serve them). Eventually, AI systems might evaluate and purchase AI capabilities to empower themselves, leaving humans to do - something. But for the foreseeable future, there’s a person with a need who will decide what tools and capabilities they choose to use (and pay for). How do we effectively align product with user?
Step one is Design Strategy: how does this project align with your priorities, capabilities, brand, team, and market trends? (see ‘How do we approach Design Strategy?’ for more details)
Step two is to learn in detail who these users are, and what they want. User Research is still the best way we have to discover:
Who are our target users?
How do we describe variations in these users (archetypes)?
What are they asking for?
What do they really want? (it’s typically not what they’re asking for)
What are their capabilities, i.e. what technologies are they able to use?
What are their goals, and what do they need to satisfy them? (i.e. the “Job to be Done” framework)
Example of a functional Persona, confidential details blurred
User Researchers and Designers take this information and create Personas, which are incredibly helpful in ensuring the outcome of product development delivers value for real people. The goal of this effort is to ensure the ‘V’ in MVP stands for ‘Viable’, rather than ‘Vaguely [relevant to users, hopefully]’. Too often MVP products are defined by what can be built quickly and cheaply, these research-driven personas help keep teams focused on solving real user needs.
Step three is to define the system design fundamentals properly. These traditional issues in system design—from requirements gathering to interfaces, integrations, database management, testing, and deployment will determine whether the core concept functions correctly, and can be scaled to demand. (see Eliot’s LLM Whitepaper for an example of building a system defined by fundamentals aligned with the core need).
Step four is product design: designing an interface that is approachable, intuitive, and compelling to your target users. The greatest tool in the world won’t get traction if it scares off potential users, or if they get frustrated during their initial use. This is especially true of AI-powered products, as the user typically creates the prompts that drive outcomes. How does the product guide the user to create prompts that will deliver the desired results? How does it guide them to revise their prompts if they get unhelpful results? Can the tool correctly interpret and refine the user’s prompts to improve performance? Can it help them learn what works, and why? Can it become indispensable to these users?
An effective AI product designed with the above approach will provide the user such value that, should it be taken away someday, they will spend every day afterward wishing they had it back. In other words, an effective AI product design will change the baseline expectations so completely that returning to the old way of doing things will feel tedious and burdensome. That’s the baseline expectation your team should hold for their project.
The “sticky” AI-powered products (ones that see active usage and avoid the ‘January 3rd Closet’) will be the products that deliver immediate value to users, value that grows over time. Effective User-Centered Design will be core to those products that break through and succeed. It’s tempting to shortcut and under-invest in the Design Strategy, User Research, System Design & Product Design aspects of product development in the race to market an MVP during this era of explosion in AI opportunity, but a bit of time and investment in these areas can make or break your entire program.
