Bridging the Gap: Translating User Needs to Model Capabilties

In a previous piece, we introduced the idea of the Blank Box Barrier – the anxiety of the empty text box that keeps many users from becoming repeat users of Large Language Models (LLMs). We discussed how Use Case Taxonomies can help AI companies identify high-value applications for their technology. But uncovering user needs is just the first step. The next hurdle lies in translating those insights into specific features that drive product adoption.

This article dives deeper into this critical challenge, exploring strategies that bridge the gap between user-led design and cutting-edge AI research. We'll examine how AI companies can leverage their Use Case Taxonomies to navigate the tension between prioritizing solutions for real user problems and pursuing technical advancements to maintain a competitive edge.

Technology-First or Customer-First?

Foundational LLMs thrive on the brilliance of their researchers – typically PhD-level scientists focused on pushing the boundaries of artificial intelligence and machine learning. While most tech companies increasingly embrace customer-centricity, the focus within AI research often leans towards achieving breakthroughs in a model's capabilities. This can lead to a gap between new technical advancements and product experiences that actually help people achieve specific and useful outcomes.

To bridge this gap, AI companies need to establish a strong feedback loop between cutting-edge research and user-centered design. This requires balancing two approaches: technology-first and customer-first innovation. 

• Technology-First (Inside-Out): This approach starts internally, focusing on a model's capabilities and then seeking existing problems it can address. This inside-out approach is driven by unique technical capabilities that can be broadly beneficial across many use cases, and often requires many years of research to achieve. Many genAI companies are technology-led based on their underlying model and the interests of their founders and research teams.
• Customer-First (Outside-In): Conversely, this approach starts externally, examining customer segments and use cases to understand their needs and challenges. It leverages frameworks like Use Case Taxonomies to categorize users by requirements, workflows, and pain points.  Most traditional B2C companies follow this outside-in strategy, focusing on a customer segment and a specific problem, then developing technology to address it. While effective, it can miss out on groundbreaking applications enabled by cutting-edge technology.

The goal is to find the intersection between the two: what can the technology do, and what does the customer genuinely need? This allows us to create products that are both technically advanced and highly relevant to users.

Case Study: Multi-Modal Video Creation

Imagine we're part of an AI company specializing in creating multi-modal models for content creation. Multi-modal models can process and integrate multiple formats of media like images, audio, and video together. 

Based on enthusiasm from early adopters and market research, the company has identified two potential use cases to focus on: creating short videos for social media and producing full scenes for feature-length films. A simplified version of their Use Case Taxonomy might look like the following:

1. Short videos for social media
   • Profile/Demographic: Amateur content creators seeking self-expression and personal gratification.
   • Customer Goal: Grow an audience by creating viral, entertaining content.
   • Workflow: Capture footage on the go using their iPhone, edit it in mobile apps like CapCut utilizing simple templates and tools. Post to social media platforms at least 3-5 times a week.
   • Challenges: Limited editing expertise, mobile editing constraints, translating creative ideas into finished products.
   • Outcomes: Creation of short, captivating videos under one minute.
2. Feature-length film editing
   • Profile/Demographic: Professional movie editors working on high-budget films.
   • Goal: Produce award-winning, 120+ minute films
   • Workflow: Extensive multi-month processes involving storyboards, hundreds of hours of footage, and collaboration with specialized teams, and editing using advanced technology like Final Cut Pro.
   • Challenges: Manage highly collaborative workflows and integrate diverse content elements into a cohesive narrative.
   • Outcomes: Deliver long-form content characterized by high precision and stylistic cohesiveness.

After constructing their Use Case Taxonomy, the company realizes that creating short social media clips and producing scenes for feature-length films are quite different use cases. Is it possible to build a single model and product that meets the needs of both?

Mapping technical capabilities to user needs

In order to make this decision, the company wants to understand the intersection between technical capabilities and user needs. They start by creating a detailed workflow for both amateur content creators and professional editors, describing every step they take in the creation process and the tools and outcomes they want to achieve along the way.

Next, they review each step of the workflow and brainstorm what type of technical capabilities could be useful to each customer; given the outcome the customer is trying to achieve, how could the model help them reach their goal?

As they build this map of technical capabilities to customer needs, they discover that both segments want to combine small pieces of “reference” media with original video clips to create brand new content. For the amateur content creator, it might be a clip they discovered on TikTok  they want to use as stylistic inspiration for their own video. They may want to instruct the model to edit their video by copying the style of the reference clip.

The professional editor wants to combine a script with raw video footage to create “conceptual edits” of different scenes. They want the model to use the script as a guide to compose a first-pass cut of a scene across all of the different takes filmed by the director, which can serve as a starting point before they delve into their full editing process.

The professional film editor has a much more precise and sophisticated workflow than the amateur content creator but by digging deeper into the needs of both, the AI company was able to find a very common need across both that their model can address. 

Mapping technical capabilities to customer needs enabled this company to thread the needle between these two segments. Now, they can confidently create features that are both meaningful to customers as well as impressive technological advancements.

Combining Approaches for Holistic Innovation

As you combine the technology-centric approach with the customer-focused approach, you’ll likely run into the following categories of opportunities:

1. Existing Home-Runs: Places where your model’s technical capabilities are already closely aligned to user needs. These are the use cases that your model excels at and your customers already know and use you for. 
2. Low Hanging Fruit: Places where incremental improvements to your model can allow it to perform better across different use cases. Customers might already be using your product for these use cases but with mediocre or inconsistent results. Smaller technical investments can boost model performance and lead to higher adoption.
3. High-risk, High Reward: Places where technical innovation is needed to capture entirely new use cases, each of which represent significant market opportunities.

Teams will need to effectively balance across these categories to create a diverse portfolio of “bets” on how to improve their models and products.

Conclusion

Conquering the Blank Box Barrier requires a dual approach that combines technological innovation with a deep understanding of user needs. By developing a Use Case Taxonomy grounded in UX Research, companies can better segment their users and tailor their products to meet specific needs. Balancing a technology-first approach with a customer-first mindset allows for the creation of AI tools that are both cutting-edge and highly relevant to users. This holistic strategy ensures that AI products not only stand out in the market but also provide consistent, repeatable value, driving adoption and long-term success.

Misha Cornes is a User Research executive with experience building and scaling research teams at companies like McKinsey and Lyft. Kene Anoliefo is the co-founder of HEARD.

‍