Why my initial design failed and what it taught me about designing a product restaurant owners actually pay for
Myna is an AI agent designed for small to medium-sized businesses, enabling quick marketing actions and attracting more customers without expensive agencies or complicated software.
PRODUCT EVOLUTION
From this
A swipe-based app
AI-powered for quick marketing, but lacked clear guidance.
To this
A task-focused experience
Simplified flows so users could act, stay motivated, and track progress easily.
And finally this
A multimodal chat experience
That brought AI insights and conversation together.
Each iteration was guided by user feedback, balancing speed-to-market with clarity of value.
At this time, Myna is seeking $1.5M in investment and working to find product-market fit.
MY ROLE & TEAM
My Role
Founding Designer
Team
CEO, CTO, AI Engineer, Founding Designer, 2 Developers and 2 Junior Designers
Timeline
Early-stage startup
(8 months)
Domain
B2B, SaaS,
AI Restaurant tech
My initial responsibility was crafting the Myna bird persona, logo, and brand, an effort that later played a key role in doubling investor demo conversions.
Simultaneously,
I Led and mentored a team of junior designers, aligning efforts across product, marketing, and social media to deliver a cohesive, consistent, high-impact brand presence.
Partnered with the CTO and AI Engineer to shape interactions and lead product design - balancing usability, brand, and tech feasibility.
MY POV (THE CHALLENGE)
I joined the founding team at a stage where the product concept existed but no UX research had been done.
No discovery phase
Decisions relied on assumptions and past research.
High speed-to-market pressure
Investors expected fast results.
Abstract concept
“customer sentiment” meant different things to different stakeholders.
THE FOUNDER'S VISION
A Tinder-like swipe interface letting owners approve marketing actions in seconds.
The swipe card experience was meant to be the magic- One gesture, one action, zero friction.
THE PROBLEM
On paper it seemed simple. In reality, restaurant owners were confused:
01
One restaurant owner swiped through a few cards, then asked:
“What am I looking at?”
They were unclear about what the swipe gesture would do, where the content came from, and why they should trust it.
02
Another told me:
“It would make things faster… but I wouldn’t pay for it.”
03
Some were even more blunt:
“I can just open ChatGPT and get all this done for free. Why would I pay for another app?”
The Root cause:
Swipe interaction was unclear: Users didn’t know they could swipe or what the cards meant.
Scroll was unattempted: Users didn't know there was more info available as they scroll further.
Some users misundertook the swipe lable as an option to scroll further.
No clear focus or hierarchy: Users are unsure of what to do first.
Confusing task flow: steps and progression feel disconnected.
Information isn’t transparent: It doesn’t explain why or how it helps, making decisions harder.
Figure shows: UX Audit and user test issues mapped on swipe cards
IMPACT OF THE PROBLEM
More than 60% users felt the swipe was a gimmick for something that required context and trust.
The product executed tasks they could already do 20% faster. But, not not valuable enough to pay for
Half of them couldn’t find value, having to take extra steps instead of getting results immediately.
And the other half couldn't even get through the entire card to execute the task. It was a disaster.
RESEARCH
So I set out to uncover real-world use cases that users would pay for.
Interview purpose
During the conversations
01
One owner told me:
“If the app could make my everyday processes easy, I’d pay for it. I want something that’s actively chasing opportunities for me.”
02
Another was Blunt
“If you can bring me catering orders, I’ll pay you tomorrow.”
They described their ideal AI assistant as one that could tell them:
What’s happening nearby today?
What are my competitors up to?
How will today’s weather affect my sales?
Quick check-ins on my sales/labor metrics.
Industry trends that matter right now.
Ingredient pricing changes.
Key operational reminders.
Motivational boosts for my team.
Opportunities to book catering orders.
Desk research and interviews revealed the major challenges restaurant owners were facing with their visibility.
Figure shows: Affinity mapping on Figjam.
01
Reputation Management
Owners were spending 6–10 hours/week responding to reviews, up to 20% of an owner’s weekly administrative time, directly hurting their ability to run the business.
02
Social Media & Content
Over 70% struggled to keep up with trends and content creation due to limited time and resources, leading to lower engagement.
03
Customer Acquisition Pressure
Customer acquisition costs were going up 15–25% every year, making it harder and more expensive for owners to bring in new customers.
Competitor benchmarking and SWOT Analysis revealed exsisting solutions and best practices, weakneses and opportunities for Myna.
Figure shows: Competitor Analysis on Figjam.
Figure shows: SWOT Analysis on Figjam.
Research revealed something critical -
Restaurant owner's didn't want to pull information from an app - They wanted an app that pushed opportunities to them. Making it clear we needed deeper, restaurant-specific intelligence they couldn’t find elsewhere.
The swipe-card concept wasn’t the right experience to deliver it.
IDEATION
Before jumping into solutions, I aligned the team on three things:
Clarity of user value that we identified from the research
Feasibility with our engineering bandwidth - (Partnering with the CTO)
A workflow flexible enough to evolve as the company matures
After uncovering all of this and a lot of back-and-forth with the founding team, It led me ask: How can I make Myna less of a tool… and more of a partner?
THOUGHT PROCESS
Instead of leaving owners to explore or guess, I structured actions as clear, guided tasks that showed what to do, why it mattered, and how it affected the business. To make this happen I defined a hypothesis:
“If we reduce cognitive load and reframe marketing as a sequence of meaningful, bite-sized tasks every week, owners will feel more in control and take consistent action.”
Once engineering could support AI chat, I worked with the founding team to map all core flows into a lightweight systems blueprint showing how AI, tasks, analytics, and user actions connected. This became a shared reference that aligned product and engineering, and helped spot issues early.
Figure shows: Task focused process flow on figjam
SOLUTION: PIVOT
From Swipe Cards to a focused task based experience: Hidden Actions to Transparent, Guided Tasks.
Recap:
Owners didn’t want to swipe through content to find value.
So I designed the task model to surface value before the user even interacted.
Figure shows: Task-focused App (Deployed in september 2025)
01
I introduced the Myna Score on the top. It became the primary orienting signal - a single metric representing marketing health and revenue potential.
What exactly is the Myna Score?
The AI engineer devised a simple metric that blends 10 key metrics into one simple, credit-score-like number that shows a business’s revenue potential, making it easy to measure and track financial health at a glance.
Product-wise, this let us center the experience around a single, easy-to-understand KPI, giving owners a clear reference point while allowing the system to personalize tasks automatically.
02
A weekly progress card to show improvements and to nudge them to stay engaged and demonstrate tangible impact.
This went beyond UX to behavioral design. By turning hidden marketing tasks into visible progress, we aimed to keep owners engaged and help them see Myna as making real progress, not just work.
Recap:
Owners hesitated with the swipe UI because they couldn’t see the context behind each action. To address this:
01
I replaced swipe cards with explicit, structured tasks that clearly showed what was being done, why it mattered, and how it impacted their business.
Figure shows: Low-fi wireframe
Figure shows: Task-focused App (Deployed in september 2025)
02
Each task included supporting context - the data source, expected outcome, or urgency - making decisions feel informed rather than arbitrary.
Figure shows: Detailed campaign view screen (Deployed in september 2025)
Figure shows: Review response/task execution screen (Deployed in september 2025)
03
I worked with the tech team to introduce mechanisms that flipped the interaction model:
Push notifications for urgent opportunities (e.g., reviews, catering leads, weather-based promos). Updated with ver.1.4.5
Smart reminders based on business patterns, not calendar time. (Work in progress)
A future pathway for daily streaks, reinforcing habit without pressure. (Work in progress)
Together, these features created a more anticipatory system. One that acts on behalf of the owner, not just responds to taps and swipes. It shifted the product posture from 'reactive' to 'proactive'.
Since we were on a tight timeline, as the founding product designer, my plan was to ship fast: design key screens, hand off to the dev team to vibe code using Claude, test with users, gather quick feedback, and iterate to improve the overall UI and experience.
Figure shows: Dev hand-off screen on Figma.
To maintain quality at that pace, I delegated developed screens to junior designers for structured QA guiding them on interaction fidelity, spacing, hierarchy, and edge cases. Their reviews helped ensure about 85% of screens met the same design standards, created a parallel QA stream, and helped us catch issues early without slowing momentum.
All identified issues were then documented directly in Figma and then pushed to GitHub as design fixes, creating clear, trackable backlogs for the dev team to work through in upcoming sprints.
Figure shows: Dev hand-off screen on Figma.
TESTING
We shipped the feature to all cohort 1 restaurants and gathered daily and weekly feedback. I designed the flow to feel rewarding and easy to follow, but some users felt overwhelmed by what felt like an endless stream of tasks.
After the initial tests, it became clear something still wasn’t right. Because the agent pulled data from multiple knowledge bases, it started hallucinating and offering generic insights. The dev team needed more time to resolve the underlying issues.
With only a few days left before launch, I had to move fast. I created a plan that would buy the team time while allowing me to continue testing with owners, gathering feedback, and refining the experience.
See user comments here
0E0E10
QUICK FIX
I removed the chat input field, and temporarily made the chat fully tap-based. This gave the dev teams time to work on the AI chat while keeping the product usable.
Before (With Chat input feild)
After (Removing the chat input field and switching to tap-based interactions)
Screen grab of the current experience
BUSINESS IMPACT (Weeks 1-4 post launch)
Cut user confusion
The strategic redesign that cut user confusion by 70% and increased weekly task completion during testing.
Shift to Guided Interactions
Moving from open multi-modal chat to tap-based guided interactions cut AI-related complaints by 50% and helped engineers stabilize the agent.
Low Willingness to Pay (WTP)
Despite positive reactions to the UI, 0% of interviewed owners expressed willingness to pay for the swipe-based or task-based versions.
Weak Retention After Week 1
Even after introducing tasks and guidance, over 60% of owners stopped returning after completing the weekly tasks, usually by Day 2.
Low Daily Active Usage (DAU)
Most owners spent under 3 mins per session, and only 1–2 sessions per week.
Low Trust in AI Output
Because early models hallucinated or surfaced generic insights, over 70% of owners said they wouldn’t act on the AI’s suggestions.
REDESIGNING THE EXPERIENCE AGAIN
While engineering worked on the fixes, I turned my attention to improving the overall experience. I went back to how restaurant owners naturally communicate quick, casual, back-and-forth messages.
During testing, I noticed something important:
01
The weekly tasks were too easy to finish. Most owners completed everything within the first two days. After that, they had nothing meaningful to come back for, and our usage data reflected it. Once all tasks were done, their engagement dropped and daily return visits fell off.
02
At the same time, whenever the chat input was available, owners instinctively tried to ask the agent questions. They wanted a conversation, not a checklist. This behavior told me something valuable: owners expected the AI to guide them, surface insights, and proactively tell them what mattered rather than making them search for it.
That shift in expectation was my signal to rethink the experience beyond just the task list. It pushed me to reimagine the product as something that supports owners throughout the week, not just something they open to “get things done.”
PIVOT
This redesign shifts Myna from a linear, task-heavy workflow to a multimodal chat experience that matches how restaurant owners actually work.
The new home screen helps owners jump back into context instantly. High-value actions sit at the top, while chat, tasks, and insights flow together naturally so owners never wonder “Where do I start?”
Figure shows: Home screen of the redesigned multimodal experience
I didn’t want users to feel they had to complete everything at once.
Instead, I wanted them to enjoy small wins by completing tasks that contributed towards long-term business growth without overwhelming the owner.
Myna also sends timely, personalized notifications throughout the day. These alerts are driven by the restaurant’s actual activity patterns, surfacing moments that matter most: reviews that need attention, social media opportunities, and real-time insights that can improve operations.
Figure shows: Notification screen of the redesigned multimodal experience
For example, tapping a 2-star review notification drops the owner directly into the exact chat thread so they can respond in under a minute.
And if they skip a notification, nothing is lost. Every pending task is neatly stored in the command center under “Tasks to complete.”
Figure shows: Command center of the redesigned multimodal experience
I didn’t want them to feel pressured to respond to every single review immediately.
Instead, I wanted them to feel in control being able to manage reviews, social posts, and insights at a pace that matches their day.
After a task is finished, it moves into the Library keeping the main view focused on what’s next while giving owners a simple way to track everything they’ve already completed.
Figure shows: Library section of the redesigned multimodal experience
Multi-modal chat based experience explorations
WRAPPING IT UP
Joined the founding team with a core product concept but no UX research, and faced high pressure to ship fast.
Built the brand, persona, and visual language, helping double investor demo conversions.
Tested the initial swipe-based design, only to discover restaurant owners didn’t understand it, couldn’t see value, and wouldn’t pay for it.
Conducted interviews, research, and competitive analysis to uncover real restaurant owner pain points and the need for proactive AI assistance.
Pivoted the product from swipe cards to a task-focused experience, introducing the Myna Score, weekly progress tracking, and transparent contextual tasks.
Iterated further to a multimodal chat experience, combining AI insights, and notifications to match how owners actually work and support long-term engagement.
Currently, the app is being shipped out to restaurants in cohorts to gather feedback, and the team is actively working to develop a fully functional multimodal experience.
MY LEARNINGS
This project humbled me in the best way possible.
Good design can’t save a weak value proposition.
I learned that no amount of polish, cleverness, or visual craft can make people care about something that doesn’t solve a real problem.
Speed without direction leads to beautiful mistakes.
We moved fast and shipped often, but without validating the core value, our momentum only got us to the wrong place faster.
Engagement isn’t the same as value, and delight can’t come before trust.
We celebrated interaction spikes, but none of it mattered without recurring, daily usefulness. Restaurant owners return for value, not novelty.
Observe actual user behavior, not assumptions.
Interviews and real-world testing revealed owners wanted proactive guidance, not a reactive checklist. Watching how people naturally work led to the shift from task lists to a multimodal chat experience.
AI is only as valuable as the human reality it fits into.
Intelligence means nothing if it doesn’t respect people’s time, mental load, and goals. Real impact happens when technology understands context.
Failure is a mirror, not a verdict.
Every design that fell flat showed me exactly what I wasn’t seeing yet and that reflection shaped me far more than anything that went right. It helped me improve the product, my thinking, and my approach to user-centered design.