Over the past year, TensorOps has observed a recurring scenario: organizations invest significant time—often 5-7 months—fine-tuning their AI models to achieve higher accuracy before launching them. Yet, when these applications finally go live, user adoption remains disappointingly low. When end-users are asked why they're not engaging with the new tools, the common refrain is, "It didn't provide a good enough answer."
In response, product managers often interpret this feedback as a mandate to further improve the model's accuracy, initiating yet another cycle of development. However, this approach overlooks a critical insight into user behavior and needs. When users encounter unsatisfactory results, they don't simply give up; they seek alternative solutions. Many admit, "I just copied my data into ChatGPT and adjusted the prompts until I got the answer I needed."
It's revealing that users turn to a general-purpose AI tool like ChatGPT—untuned for their specific use case—and yet achieve satisfactory results through interactive refinement.
This pattern highlights a fundamental truth: users value the ability to interact with AI systems, refining and guiding them to meet their specific needs. They prefer tools that allow them to participate actively in the problem-solving process rather than passively receiving outputs from a supposedly perfect model. Therefore, instead of solely focusing on enhancing model accuracy, companies should prioritize designing user experiences that empower users to fine-tune outputs themselves. By facilitating user interaction and enabling prompt engineering, applications can become more effective, user-friendly, and ultimately see higher adoption rates.
Let's explore four companies that have successfully embraced this human-centered UX approach in their generative AI applications—and one that has room for improvement.
1. ChatGPT: Empowering Users Through Interactive Dialogue
ChatGPT has become an indispensable tool for developers, writers, and professionals worldwide, even though it wasn't specifically designed for specialized tasks like debugging code or drafting legal documents. One key to its widespread adoption lies in its conversational interface (alongside really good AI engine, but this you can get from them as API, for the most part). Users can provide detailed context, ask follow-up questions, and iteratively refine their queries. This back-and-forth interaction allows users to guide the AI toward the solutions they need. Features like syntax highlighting and markdown support enhance readability and usability. By prioritizing collaboration over one-off accuracy, ChatGPT empowers users to achieve their goals more effectively.
2. Fathom: Customizable Summaries with User Input
Fathom is an exceptional note-taking tool that addresses the challenge of generating meaningful summaries from transcripts—a task where context and nuance are crucial. Recognizing that automated summaries can't capture every detail, Fathom offers standard conversation templates like sales calls with automatic BANT (Budget, Authority, Need, Timeline) detection. More importantly, it provides a textbox where users can input specific instructions or context for summarizing the content. Users can request shorter summaries, focus on action items, or highlight particular topics, tailoring the output to their unique needs. This interactive approach ensures that the summaries are not just accurate but also relevant and useful to the user.
3. Azure AI Studio: Personalization Through System Prompts
Azure AI Studio offers a chat experience similar to ChatGPT but takes customization a step further by allowing users to modify the system prompt—the underlying instructions that guide the AI's responses. Users can set the context for the entire conversation, define roles, or provide few-shot examples directly in the interface. This empowers users to tailor the AI's behavior to specific use cases without requiring deep technical expertise. By putting control in the hands of the users, Azure AI Studio lets users generate a high end customized chat experience powered by the same OpenAI models behind ChatGPT but with VERY little work to tune them.
4. Notion AI: Infinite Editing Possibilities Within the Platform
Notion AI understands that writing and text editing are highly personal processes. While it offers standard manipulations like formatting, bullet points, and shortening text, it also allows users to input free-text prompts as instructions for editing. This means users can stay within the platform and direct the AI to adjust the text in highly specific ways—whether that's changing the tone to be more formal, adding detailed explanations, or rephrasing sections for clarity. By facilitating direct interaction with the AI, Notion AI enhances productivity and by making sure that the user doesn't need to jump to any other tool for editing the text or the Gen AI responses.
The Missed Opportunity: LinkedIn's Limited Customization
In contrast, LinkedIn introduced a "Write with AI" feature designed to help users craft posts that align with the platform's professional style—adding hashtags and refining language to enhance visibility and engagement. While it does improve the text to be more "LinkedIn-esque," the feature falls short by not allowing users to provide specific instructions on how to adjust the output. Users can either accept the generated text, revert to their original, or provide generic positive or negative feedback. This lack of interactive refinement means that whenever the AI doesn't fully cater to individual user needs, LinkedIn sends them to look for another platform to finish off the work. End result is a less engaging experience and a missed opportunity to extend user involvement.
Conclusion: The Power of User-Centric Design in Generative AI
These examples highlight a lesson for companies developing generative AI applications: empowering users to interact with and refine AI outputs not only enhances user satisfaction but also drives adoption more effectively than a singular focus on accuracy. Users are not merely passive recipients; they are active participants who appreciate tools that accommodate their input and adapt to their needs.
As the AI landscape continues to evolve, organizations should reorient their strategies from solely perfecting models in isolation to creating interfaces that foster collaboration between users and AI. This means investing in user-centric design, intuitive interfaces, and features that allow for personalization and interactive refinement. After all, even the most advanced model is only as valuable as its ability to meet the diverse and dynamic needs of its users.
The message is clear: to truly succeed in the realm of generative AI, companies must embrace user-centric design principles that empower users to shape their own experiences. By doing so, they will not only enhance user satisfaction but also shorten time to market.
Comments