Myna AI
| 2025
Pivoting an AI product from novelty to trusted execution system that drove +65% task completion for restaurant owners in 12 weeks
Pivoting an AI product from novelty to trusted execution system that drove +65% task completion for restaurant owners in 12 weeks
Outcome
Outcome
Task Completion
Task Completion
+65%
+65%
Shifted from swipe evaluation to weekly task ownership. Task completion increased from <20% → +65%
Shifted from swipe evaluation to weekly task ownership. Task completion increased from <20% → +65%
User Confusion
User Confusion
-70%
-70%
Used progressive disclosure to reduce cognitive load. User confusion dropped 70%
Used progressive disclosure to reduce cognitive load. User confusion dropped 70%
AI Complaints
-50%
Replaced free-form prompts with tap-based flows. AI-related complaints dropped 50%
Replaced free-form prompts with tap-based flows. AI-related complaints dropped 50%
Early signals also showed improved repeat usage and clearer paths toward retention and monetization, as owners began completing work instead of merely reviewing suggestions.
Early signals also showed improved repeat usage and clearer paths toward retention and monetization, as owners began completing work instead of merely reviewing suggestions.
Context
Context
Myna is an AI marketing assistant for independent restaurant owners.
Myna is an AI marketing assistant for independent restaurant owners.
It helps owners respond to reviews, manage social media, and run campaigns without hiring agencies or learning complex software.
It helps owners respond to reviews, manage social media, and run campaigns without hiring agencies or learning complex software.
The job wasn’t “more AI.” It was getting owners to actually complete marketing work while running a restaurant.
The job wasn’t “more AI.” It was getting owners to actually complete marketing work while running a restaurant.
Primary user:
Primary user:
Independent restaurant owners managing daily operations while juggling marketing as a secondary responsibility.
Independent restaurant owners managing daily operations while juggling marketing as a secondary responsibility.
Constraints:
Constraints:
Used in short, interrupted sessions between service tasks
High cognitive load, low tolerance for ambiguity
Mobile-first usage
Used in short, interrupted sessions between service tasks
High cognitive load, low tolerance for ambiguity
Mobile-first usage
Used in short, interrupted sessions between service tasks
High cognitive load, low tolerance for ambiguity
Mobile-first usage
Not for:
Not for:
Agencies managing multiple brands
Enterprise restaurant chains
Agencies managing multiple brands
Enterprise restaurant chains
Agencies managing multiple brands
Enterprise restaurant chains
TL;DR
Myna launched as a swipe-based “approve/dismiss” AI assistant optimized for speed. Early testing showed that speed didn’t translate to trust or action.
Myna launched as a swipe-based “approve/dismiss” AI assistant optimized for speed. Early testing showed that speed didn’t translate to trust or action.
I led a product pivot from a novelty interaction model to a task-driven execution system with weekly cadence and constrained AI interactions. This shifted prioritization from the user to the system, allowing owners to act with confidence instead of constantly evaluating options.
I led a product pivot from a novelty interaction model to a task-driven execution system with weekly cadence and constrained AI interactions. This shifted prioritization from the user to the system, allowing owners to act with confidence instead of constantly evaluating options.
Role
Role
Product Designer
Product Designer
Team
Team
Lean founding team (CEO, CTO, AI Engineer, Designer)
Lean founding team (CEO, CTO, AI Engineer, Designer)
Domain
Domain
B2B SaaS · AI for Restaurant Tech
B2B SaaS · AI for Restaurant Tech
Platform
Platform
Agnostic (iOS and Android)
Agnostic (iOS and Android)
Timeline
Timeline
12 Weeks (from audit → pilot)
12 Weeks (from audit → pilot)
Company
Company
Early-stage / finding product-market fit
Early-stage / finding product-market fit
Problem
Problem
Owners didn’t trust the system enough to act, so the AI value never landed.
Owners didn’t trust the system enough to act, so the AI value never landed.
The swipe interaction looked fast in theory (“one gesture, zero friction”). In practice, it removed the context owners needed to feel confident. The experience felt like a stream of suggestions, not progress.
The swipe interaction looked fast in theory (“one gesture, zero friction”). In practice, it removed the context owners needed to feel confident. The experience felt like a stream of suggestions, not progress.
When asked to explain the experience, users said:
When asked to explain the experience, users said:
“What am I looking at?”
“What am I looking at?”
“It would make things faster.. but I wouldn’t pay for it.”
“It would make things faster.. but I wouldn’t pay for it.”
“I can just use ChatGPT and get all this done for free."
“I can just use ChatGPT and get all this done for free."
Baseline signals:
Baseline signals:
Returning users: <10%
Returning users: <10%
Task completion: <20%
Task completion: <20%
Willingness to pay: Low
Willingness to pay: Low
The issue wasn’t usability alone; it was clarity and trust.
The issue wasn’t usability alone; it was clarity and trust.
Problem statement
Restaurant owners didn’t trust AI-generated suggestions enough to complete real marketing work, preventing value, retention, and monetization.
What Went Wrong
What Went Wrong
I reviewed usability tests and interaction recordings. The same breakdowns showed up in every session.
I reviewed usability tests and interaction recordings. The same breakdowns showed up in every session.
UX Audit and user test issues mapped on swipe cards
UX Audit and user test issues mapped on swipe cards
Root Cause
Root Cause
1
The swipe concept was defined without any user validation.
The swipe concept was defined without any user validation.
2
Early decisions were assumption-driven (speed-to-market pressure + investor expectations)
Early decisions were assumption-driven (speed-to-market pressure + investor expectations)
As a result, the product optimized for interaction speed instead of confidence in decision-making.
As a result, the product optimized for interaction speed instead of confidence in decision-making.
Assumptions (before research)
Assumptions (before research)
1
Initial Assumptions
Initial Assumptions
Faster interaction (swipe-based approval) would increase task completion
Less friction would automatically translate to trust
Restaurant owners wanted more AI-generated options, not structured guidance
Faster interaction (swipe-based approval) would increase task completion
Less friction would automatically translate to trust
Restaurant owners wanted more AI-generated options, not structured guidance
Faster interaction (swipe-based approval) would increase task completion
Less friction would automatically translate to trust
Restaurant owners wanted more AI-generated options, not structured guidance
2
Why We Believed Them
Why We Believed Them
Investor narratives around “AI speed”
Founder's intuition based on consumer swipe patterns
Industry momentum around ChatGPT-style free-form input
Investor narratives around “AI speed”
Founder's intuition based on consumer swipe patterns
Industry momentum around ChatGPT-style free-form input
Investor narratives around “AI speed”
Founder's intuition based on consumer swipe patterns
Industry momentum around ChatGPT-style free-form input
3
What We Expected to Validate
What We Expected to Validate
That faster interaction → higher completion
That users would feel confident approving AI output without added context
That faster interaction → higher completion
That users would feel confident approving AI output without added context
That faster interaction → higher completion
That users would feel confident approving AI output without added context
Strategy
Reframe Myna from “idea generator” to “execution system.”
Reframe Myna from “idea generator” to “execution system.”
Restaurant owners were time-poor and decision-fatigued. They didn’t want more options. They wanted:
Restaurant owners were time-poor and decision-fatigued. They didn’t want more options. They wanted:
1
A small set of high-impact actions
A small set of high-impact actions
2
Clear priorities
Clear priorities
3
Scope they could finish
Scope they could finish
4
Guidance that reduced thinking
Guidance that reduced thinking
Research sprint. (2 weeks)
Research sprint. (2 weeks)
1
13 restaurant owner interviews
13 restaurant owner interviews
2
Competitive benchmarking and SWOT
Competitive benchmarking and SWOT
3
Affinity mapping and behavioral synthesis
Affinity mapping and behavioral synthesis
4
Quantitative & qualitative patterns from usability sessions
Quantitative & qualitative patterns from usability sessions
Sample size:
13
Competitor Landscape Review
Market Positioning & SWOT Findings
Insight Synthesis: Affinity Mapping
Research Repository: Interviews & Analysis
“If you can bring me catering orders, I’ll pay you tomorrow.”
“If the app could make everyday processes easy, I’d pay for it."
“I want something that’s actively chasing opportunities for me.”
Sample size:
13
Competitor Landscape Review
Market Positioning & SWOT Findings
Insight Synthesis: Affinity Mapping
Research Repository: Interviews & Analysis
“If you can bring me catering orders, I’ll pay you tomorrow.”
“If the app could make everyday processes easy, I’d pay for it."
“I want something that’s actively chasing opportunities for me.”
Sample size:
13
Competitor Landscape Review
Market Positioning & SWOT Findings
Insight Synthesis: Affinity Mapping
Research Repository: Interviews & Analysis
“If you can bring me catering orders, I’ll pay you tomorrow.”
“If the app could make everyday processes easy, I’d pay for it."
“I want something that’s actively chasing opportunities for me.”
Sample size:
13
Competitor Landscape Review
Market Positioning & SWOT Findings
Insight Synthesis: Affinity Mapping
Research Repository: Interviews & Analysis
“If you can bring me catering orders, I’ll pay you tomorrow.”
“If the app could make everyday processes easy, I’d pay for it."
“I want something that’s actively chasing opportunities for me.”
Key Insight
Key Insight
Rather than designing for an “average user,” synthesis revealed 4 distinct behavioral groups:
Rather than designing for an “average user,” synthesis revealed 4 distinct behavioral groups:
Overwhelmed Operator
Overwhelmed Operator
Daily operations · Low time · Low tolerance
Daily operations · Low time · Low tolerance
Skeptical Pragmatist
The Skeptical Pragmatist
Cautious · ROI-driven · Trust-sensitive
Cautious · ROI-driven · Trust-sensitive
Outcome-Driven Owner
Outcome-Driven Owner
Results first · Opportunist · Revenue focused
Results first · Opportunist · Revenue focused
Time-Poor Solo Manager
Time-Poor Solo Manager
Single decision-maker · High cognitive load
Single decision-maker · High cognitive load
Across all groups, owners consistently struggled with feeds, dashboards, and swipe mechanics. But they responded reliably to clear, outcome-driven tasks with defined scope and completion.
Across all groups, owners consistently struggled with feeds, dashboards, and swipe mechanics. But they responded reliably to clear, outcome-driven tasks with defined scope and completion.
What Research changed
What Research changed
Invalidated
Invalidated
Speed alone builds trust
Speed alone builds trust
Validated
The Skeptical Pragmatist
Clear scope + completion increases follow-through
Clear scope + completion increases follow-through
Discovered
Discovered
Owners preferred the system to prioritize work for them
Owners preferred the system to prioritize work for them
Direction Shift
Direction Shift
From reactive evaluation → system-driven execution
From reactive evaluation → system-driven execution
Design Principles
Design Principles
These principles guided every decision during the pivot
These principles guided every decision during the pivot
Clarity
Clarity
Make priorities obvious in seconds
Make priorities obvious in seconds
Execution
Execution
Make action feel lightweight and finishable
Make action feel lightweight and finishable
Trust
Trust
Make AI predictable before powerful
Make AI predictable before powerful
Evaluated Directions
Evaluated Directions
I explored three structural approaches before committing to a direction, evaluating each through the lens of small-screen behavior, interruption-heavy use.
I explored three structural approaches before committing to a direction, evaluating each through the lens of small-screen behavior, interruption-heavy use.
Refined swipe model
Refined swipe model
Fast, daily execution of the top-priority action
Fast, daily execution of the top-priority action
Felt reactive and transactional
Felt reactive and transactional
But framed work as isolated reactions
But framed work as isolated reactions
No clear signal of how actions compounded into progress
No clear signal of how actions compounded into progress
AI-ranked suggestions feed
AI-ranked suggestions feed
More context and options surfaced
More context and options surfaced
Shifted prioritization back to the user
Shifted prioritization back to the user
Required scanning and comparison
Required scanning and comparison
Increased hesitation and decision
fatigue
Increased hesitation and decision
fatigue
Task-based workflow
Task-based workflow
Explicit responsibilities and completion
Explicit responsibilities and completion
Clear priorities and scope
Clear priorities and scope
Stronger sense of control and progress
Stronger sense of control and progress
Reduced thinking, increased follow-through
Reduced thinking, increased follow-through
Decision
Decision
We committed to the task-based workflow.
We committed to the task-based workflow.
Swipe and feed models still required owners to manage work mentally. Tasks moved that responsibility into the system.
Swipe and feed models still required owners to manage work mentally. Tasks moved that responsibility into the system.
Designing the System
Designing the System
However, early versions exposed too much detail upfront, causing hesitation and drop-off.
However, early versions exposed too much detail upfront, causing hesitation and drop-off.
I simplified the entry point to show only what mattered now, revealing details after intent through progressive disclosure suited for small screens.
I simplified the entry point to show only what mattered now, revealing details after intent through progressive disclosure suited for small screens.
Before: Tasks visible upfront
Before: Tasks visible upfront
After: Shows overall progress to orient the user
After: Shows overall progress to orient the user
How this translated into the final product
Overview
Overview
What matters now
What matters now
Prioritization
Prioritization
Why it matters
Why it matters
Execution
Execution
How to complete it
How to complete it
This structure reduced initial scanning cost and helped users orient quickly during short, interrupted sessions.
This structure reduced initial scanning cost and helped users orient quickly during short, interrupted sessions.
Behavior shift we saw
Home oriented users
Home oriented users
Task list helped them prioritize
Task list helped them prioritize
Task Chat supported execution
Task Chat supported execution
Users moved from browsing to completing.
Users moved from browsing to completing.
Reframing Work from Daily Pressure to Weekly Progress
Reframing Work from Daily Pressure to Weekly Progress
Daily tasks felt like an obligation. Missing a day felt like failure, even when business impact was minimal.
Daily tasks felt like an obligation. Missing a day felt like failure, even when business impact was minimal.
I grouped actions into weekly task cycles, intentionally limiting how much work appeared at once.
I grouped actions into weekly task cycles, intentionally limiting how much work appeared at once.
Before: Daily pressure
Before: Daily pressure
After: Weekly task cycles
After: Weekly task cycles
Weekly framing reduced anxiety, gave owners flexibility to act when they had time, and created a clearer sense of completion.
Weekly framing reduced anxiety, gave owners flexibility to act when they had time, and created a clearer sense of completion.
A Small but Necessary Adjustment
A Small but Necessary Adjustment
In real usage, campaign work took longer than a single week and couldn’t be completed within the task cycle.
In real usage, campaign work took longer than a single week and couldn’t be completed within the task cycle.
To account for this, I introduced a tab structure that separated high-impact, quick tasks from longer-running campaigns.
To account for this, I introduced a tab structure that separated high-impact, quick tasks from longer-running campaigns.
Before: Tasks and campaign steps were combined in a single list.
Before: Tasks and campaign steps were combined in a single list.
After: Quick tasks separated from ongoing campaign work.
After: Quick tasks separated from ongoing campaign work.
Testing in the Real World
Testing in the Real World
We piloted the redesigned experience with 10 restaurants in 2-week cohorts.
We piloted the redesigned experience with 10 restaurants in 2-week cohorts.
What worked:
What worked:
Owners completed weekly tasks within 1-2 days.
Owners completed weekly tasks within 1-2 days.
Guided chat felt natural and supportive.
Guided chat felt natural and supportive.
The experience felt focused, not overwhelming.
The experience felt focused, not overwhelming.
What surfaced
What surfaced
Campaigns still felt heavy for some users.
Campaigns still felt heavy for some users.
Free-form prompt led to vague questions, generic responses, and occasional hallucinations.
Free-form prompt led to vague questions, generic responses, and occasional hallucinations.
AI Design Trade-off: Flexibility vs Trust
AI Design Trade-off: Flexibility vs Trust
Free-form chat sounds powerful. But in early adoption, inconsistent AI behavior and unclear outcomes broke trust quickly.
Free-form chat sounds powerful. But in early adoption, inconsistent AI behavior and unclear outcomes broke trust quickly.
I reduced flexibility to make outcomes predictable, Using Claude Code, I prototyped a shift from free-form input to constrained, tap-based flows that kept AI inside a known lane.
I reduced flexibility to make outcomes predictable, Using Claude Code, I prototyped a shift from free-form input to constrained, tap-based flows that kept AI inside a known lane.
Before: users type anything → AI response quality varies
After: users tap through a constrained flow → AI stays inside a known lane
This significantly reduced AI-related complaints during early adoption.
This significantly reduced AI-related complaints during early adoption.
How We Measured Trust and Execution
How We Measured Trust and Execution
Success was defined by completed work, not interaction speed.
Success was defined by completed work, not interaction speed.
We defined success around whether restaurant owners could confidently complete real marketing work, not how quickly they could interact with AI. Metrics were chosen as signals of trust, clarity, and execution, helping us validate design decisions and guide trade-offs.
We defined success around whether restaurant owners could confidently complete real marketing work, not how quickly they could interact with AI. Metrics were chosen as signals of trust, clarity, and execution, helping us validate design decisions and guide trade-offs.
+65%
+65%
Task Completion
Task Completion
Completion was the clearest indicator of trust and value. After shifting from swipe interactions to weekly task ownership, task completion rose from under 20% to +65%.
Completion was the clearest indicator of trust and value. After shifting from swipe interactions to weekly task ownership, task completion rose from under 20% to +65%.
-70%
-70%
User Confusion
User Confusion
Tracked through usability sessions and support feedback. Progressive disclosure and a simplified entry point reduced confusion by 70%.
Tracked through usability sessions and support feedback. Progressive disclosure and a simplified entry point reduced confusion by 70%.
-50%
-50%
AI-Related Complaints
AI-Related Complaints
Used as an early warning for AI reliability during onboarding. Constraining AI interactions reduced complaints by 50%.
Used as an early warning for AI reliability during onboarding. Constraining AI interactions reduced complaints by 50%.
See impact breakdown here
Early Retention & Monetization Signals
Early Retention & Monetization Signals
Repeat usage and willingness to engage with paid workflows were tracked as secondary signals. As owners started completing weekly tasks (instead of only reviewing suggestions), retention improved and paths toward monetization became clearer.
Repeat usage and willingness to engage with paid workflows were tracked as secondary signals. As owners started completing weekly tasks (instead of only reviewing suggestions), retention improved and paths toward monetization became clearer.
Constraints & Design Trade-offs
Constraints & Design Trade-offs
We were a lean team, so every trade-off prioritized speed, clarity, and core usability over visual delight.
We were a lean team, so every trade-off prioritized speed, clarity, and core usability over visual delight.
1
AI Breadth vs. Task Completion
AI Breadth vs. Task Completion
We focused on a small set of high-value workflows instead of a broad assistant. Follow-through improved because value became easier to see.
2
Flexibility vs. AI Reliability
Flexibility vs. AI Reliability
We constrained chat to stabilize trust early. Predictability beat power during adoption.
3
Novelty vs. Business Outcomes
Novelty vs. Business Outcomes
We replaced swipe novelty with outcome-focused tasks. Completion went up, confusion went down, and owners became more willing to act.
These patterns became reusable, enabling scale without reintroducing cognitive load.
These patterns became reusable, enabling scale without reintroducing cognitive load.
My Learnings
My Learnings
The work required navigating uncertainty, experimenting early, and treating failure as a signal to iterate toward better outcomes.
The work required navigating uncertainty, experimenting early, and treating failure as a signal to iterate toward better outcomes.
1
Trust must come before intelligence in AI products. People return for usefulness, not novelty.
Trust must come before intelligence in AI products. People return for usefulness, not novelty.
2
Validate early, observed behavior is more reliable than assumptions
Validate early, observed behavior is more reliable than assumptions
3
Constraint can be a feature when users are overloaded
Constraint can be a feature when users are overloaded
Next Steps
Next Steps
1
Strengthen trust and accuracy before expanding flexibility
Strengthen trust and accuracy before expanding flexibility
2
Identify the single workflow users would pay for
Identify the single workflow users would pay for
3
Build retention loops tied to real restaurant activity.
Build retention loops tied to real restaurant activity.
I'm glad you made it here;
I'm currently open for new and exciting opportunities. Let's connect and create something nice.