Advancements in AI technologies have reached a tipping point where what was once squarely in the realm of science fiction is seemingly within reach, overnight.
In the face of this massive sea change, as we scramble to adjust plans and shift resources, tech leaders are challenged to ensure their efforts drive real impact and not just flashy headlines.
1. Focus on your users
AI is all the rage right now, but unlike recent hype cycles, the natural language and reasoning capabilities of LLMs have the potential for broad impact across a wide range of products. But, adding AI to a well-functioning product just for the sake of being on-trend will only confuse users.
Start by asking yourself if AI addresses problems or gaps before incorporating it. To find the right opportunities, focus on your users. Think about ways AI could enhance their experience, while staying aligned with your product strategy. Look for interactions that would benefit from a natural language interface, or manual workflows that AI could streamline.
Airbnb is a good example. When searching for a place to stay, users filter parameters like price range or number of bedrooms. Simply replacing a straightforward checkbox with a user typing a question as an AI query wouldn’t enable anything new and would arguably create a more cumbersome and unreliable interface. But even with multiple filters, it’s not always easy to find what you’re looking for. Even with over 100 filter categories, people can’t always find exactly what they want. This is where AI’s ability to understand the nuance of natural language can be game changing. With an AI-powered search experience, there are no limits to the customization you can offer.
What to watch out for: While it’s relatively easy to create a compelling demo utilizing new AI, a usable product is more difficult to nail. Plan for an iterative process of getting feedback from users, and assume you have much to learn when it comes to delivering a valuable product.
2. Don’t fine-tune your own model
One of the exciting facets of this latest AI wave is its ability to be hyper-customized. What used to require human-level comprehension of nuance and intent, can now be digitized and made accessible at scale.
But don’t let your fascination with the technology get in the way of everything you know about practical product development. While fine-tuning a bespoke model can seem enticing, it is a form of premature optimization that locks you into a set of choices before you’ve found your product fit. Fine-tuning an AI model prematurely slows down your rate of iteration and increases your maintenance cost, ultimately slowing down your innovation velocity.
So how do you create a custom experience? It’s all about the prompt. The prompt is a great place to set the tone for the interaction: confidence, cultural or industry adjustments, your brand voice, and more. Make sure to pass through any proprietary information you’d like the model to leverage. Think about the context you’d need to provide to a new hire to accomplish this task and distill it into your prompt.
This approach gives you the flexibility to iterate and adapt, as both the underlying technology and your understanding of how to leverage it advances. Your level of sophistication in structuring your prompt will end up being a key differentiator for your product.
What to watch out for: AI models are like black boxes — you pass in a prompt, and get a response. Even minor tweaks can lead to massive shifts in quality! Establish a quality validation process from the start to assess improvements effectively and catch degradations swiftly.
3. Establish a foundation for rapid innovation
Keep up in today’s dizzying pace by equipping your team with the ability to rapidly iterate. But while you want to foster a spirit of innovation, letting everyone just do their own thing can backfire quickly.
Make sure you start with a strong foundation by building out an AI platform to provide a paved path for developers, facilitating both rapid iteration and consistency across the product. Consider standardizing on sanctioned vendors and models, a foundational prompt framework, a quality testing approach, and base patterns and capabilities to extract relevant data from common data sources to serve as context in the prompt.
What to watch out for: While there are many challenges an AI platform can simplify, don’t over-rotate on centralizing. Remember — it’s not about the technology, it’s about its infusion into the product. Teams that own a particular aspect of your product are best suited for identifying and iterating on appropriate use cases. Aim to empower everyone on your product development team to bring AI into their domain successfully.
Taking the time to think where AI can bring the most value to your users, pulling customization and differentiation into the crafting of your prompt, and building a paved path for AI innovation across your product will unlock your ability to deliver practical AI innovation now and into the future.