A year of lessons: a product designer’s journey in designing GenAI-powered B2B solutions at a startup
I remember how, just over a year ago, I looked at the idea of designing GenAI (LLM) features/app with almost childlike naivety. I thought it would be a simple adaptation of the consumer-facing app experiences I’d seen — the chatbot panel, the ✨ generate button, just wrapped in a more “corporate” UI.
But as a Product Designer diving headlong into complex enterprise workflows at Trustana, I quickly discovered that my assumptions were way off. In the B2C world, creating a “useful” feature might involve some smart floating chatbot recommendations that entertain users or enjoying a quirky AI-generated image. However, in the B2B realm, the “usefulness of AI” is all about saving time, reducing steps, and easing the day-to-day workload.
Early Struggles and a Paradigm Shift
At first, I struggled to design and propose GenAI-powered solutions that genuinely helped our end-users. Even though I grasped the business objectives, I constantly questioned how Large Language Models (LLMs) could truly enhance their daily workflows.
Over time, I realized that feeling out of my depth wasn’t a sign of failure — it was the catalyst for a paradigm shift. I began shedding old assumptions and rethinking what “good design” truly means in these complex enterprise settings.
Below are the key lessons I learned, each explored in detail…
Less hype, more reliability
Early on, I overestimated the power of “cool” features. Maybe a consumer-grade AI can afford to be entertaining, but in B2B, flashy doesn’t cut it. Teams don’t have patience for guesswork; they need to trust that the AI’s outputs can speed up their work, not slow it down. I realized that reliability is the gold standard here. Instead of wowing users with tricks, I had to focus on showing them that this AI wasn’t just a novelty item — it was a dependable coworker.
This meant being transparent about the AI’s reasoning, offering confirmation steps before changes took effect, and leaning on familiar UI patterns that didn’t force users to relearn anything. By doing so, I wasn’t trying to impress them with technology — I was trying to earn their trust. When I got it right, users responded with relief: “This makes my job easier.” That was the kind of delight that mattered.
Ensuring AI adds value to the workflows
One of the hardest lessons I learned was that a “neat” AI feature isn’t always a “valuable” one. In B2B contexts, real work is messy. It’s full of nested tasks, dependencies, and sometimes illogical sequences that aren’t going away anytime soon. GenAI ideas that looked great on paper often fell flat when inserted into these tangled workflows because they didn’t actually address a genuine problem.
I remember concepts that seemed brilliant in isolation — like automatically generating summaries of certain data sets — but ended up adding confusion or extra steps. After observing how real users interacted with these features, I learned to filter out what didn’t help and refine what did. It was humbling and reinforced the importance of staying closely connected to our customers and users. This experience clarified my priorities: strip away what’s not useful, invest in what is, and always keep the user’s everyday struggles at the forefront of my design decisions.
Integrating GenAI into my design workflow
(aka ‘think with AI’)
Ironically, the very technology I was wrestling with became a catalyst for my own growth. Leveraging GenAI as a design assistant — quickly generating mockup variations or brainstorming UI patterns — allowed me to iterate more rapidly. It enabled me to step back from pixel-perfect adjustments and focus on understanding how each element fit into the user’s complex environment.
While I still relied on my human judgment and domain expertise to shape the final product, this AI “assistant” opened up fresh perspectives. It was like having an endless brainstorming partner, always ready with innovative ideas to help me think beyond my usual design patterns.
Designers should build and experiment, not just use!
One of my biggest realizations was that designers need to understand at least some basics of how these LLMs systems work behind the scenes.
I’m not saying we should all start coding full-stack solutions, but a little technical curiosity can transform your design instincts. After struggling through certain feature projects, I decided to get my hands dirty. I played with no-code tools (Bubble.io), set up and built simple AI-powered apps on my own.
It wasn’t always smooth, but I learned about latency issues, quirky function-calling scenarios, and why some outputs took longer than others. I even explored different LLM platforms (OpenAI, Replicate, ComfyUI, Anthropic, Elevenlabs) to understand how different environments and capabilities could change the game.
Doing this helped me become a better collaborator with developers and prepared me to spot weird edge cases early. Ultimately, it grounded my UX and UI design solutions in reality, ensuring that what I proposed could actually work under the hood, not just visually appealing mockups.
TL;DR:
Essential UX & Design Lessons for Crafting LLM-Based B2B Solutions
1. Evaluate Workflow Impact:
- Understand the Workflows:
Dive deep into your users’ entire workflows — the exact steps they take to complete their tasks. Don’t just delegate steps to AI without fully grasping the process. Revisit those Service Blueprint exercises and identify where the real pain points lie. This ensures your AI solutions are addressing actual needs, not just adding shiny features. - Avoid Adding Complexity (Be intentional):
Sometimes, adding AI can inadvertently complicate things (sometimes, not everytime). For example, a complex AI-powered bulk data enrichment steps might sound fantastic on paper, but having to manually QC or review thousands of AI-generated outputs can turn a time-saver into a time-waster. Always assess whether the AI feature truly reduces workload or just shifts the effort elsewhere. Aim for simplicity and genuine efficiency gains, not more hoops for your B2B users to jump through.
2. Clarity & Transparency
- Show the Source:
Don’t make users play detective with the AI’s answers. Whether the output is based on solid data or just the AI’s best guess, let them know right upfront. In B2B, time is money, and nobody wants to waste it double-checking questionable info. By showing where the suggestions came from, you help users quickly decide if they need to adjust or can trust the result and move on. In short: no guesswork, no second-guessing, just transparency. - Maintain Context:
Keep track of what’s been said and highlight what data the model used, so users don’t feel lost in multi-step workflows.
3. Latency & Reliability Management:
- Clear Wait States:
Show loading indicators or progress messages if responses take time, so users know the system is working and aren’t left wondering if something went wrong. - ‘Plan B’ Options:
Provide backup methods like manual templates or saved drafts if the AI fails or lags, ensuring that users can keep moving forward without getting stuck.
4. Control & Confirmations:
- Empower User Choices:
Allow users to confirm or reject AI outputs before applying them. This prevents costly mistakes and ensures that users remain in control. - Fine-Tune Options:
Offer ways to adjust suggestions to fit the user’s specific context, such as modifying the tone or level of detail. - Human-in-the-Loop:
Ensure that users can guide, correct, or override the AI’s suggestions at key decision points, keeping the final call in their hands.
5. Discoverable & Contextual Affordances
Right Place, Right Time:
Think twice before slapping AI onto every corner of the product.
If you find yourself going, “Hey, we can use AI can do this part!” — take a breath. Instead of rushing in, actually talk to your users (or your Customer Success team) and see what they’re doing day-to-day. (Going back to Point 1 above)
Don’t scatter AI tools everywhere just because you can or because your competitors are doing it. Forcing users to sift through hundreds of low-quality AI-generated outputs is a fast track to frustration, not productivity. Aim for solutions that seamlessly fit into their workflow, making things smoother, not adding more complexity.
Final Note
After a year of diving into GenAI at a B2B startup, I’ve learned that success hinges on trust, context, and a willingness to understand the technology beneath the surface.
Don’t just paint a veneer of AI “coolness” over a product — focus on what genuinely helps the people using it.Embrace the uncomfortable learning curve, get closer to the tech, and keep the user’s messy reality front and center.
In the end, “B2B,” “GenAI,” and “Workflow Design” aren’t buzzwords; they’re signals of where the true value lies. By embracing the complexity and asking hard questions, we can build products that don’t just function, but truly help, inspire confidence, and ultimately make a tough job a little easier.
Highlights of My Mini LLM App Projects in 2024
- Portfolio Page generator: https://dribbble.com/shots/23381669-Ohhh-cool-Web-profile-builder-v1-0
- AI Headshot Generator: https://tourwithpic.bubbleapps.io/version-test/ai-headshot
- Blog Generator: https://tourwithpic.bubbleapps.io/version-test/blog-writer
- Daily Expense Tracker: https://tap2track.bubbleapps.io/version-test/
- Mini Pocket Tour Guide (Audio): https://tourwithpic.bubbleapps.io/version-test/pocket-tour-guide?debug_mode=true
So, these are my personal reflections from the past year. The pace of LLM development is incredibly fast, so some of these points might feel outdated by the time you read this. I encourage all UX and UI designers out there to keep learning and exploring the capabilities of LLMs as we move into 2025 and beyond! Your feedback and comments are always welcome — I’d love to hear your thoughts. Thanks again for taking the time to read about my journey. 😀
