back

Designing an AI-powered POS System for Smarter Store Management

Designed an intelligent POS dashboard that unified orders, analytics, and AI insights helping store owners turn data into effortless business clarity.

Problem

Existing POS system used disconnected tools for orders, inventory, and reporting, creating inconsistent workflows and scattered insights. Admins lacked real-time data visibility, slowing decisions and increasing manual work.

Outcome

Redesigned the POS experience with a unified, insight-driven dashboard featuring real-time analytics, clearer data flows between admin and manager roles, and AI-assisted summaries that transformed raw numbers into actionable guidance, making daily operations faster and easier to navigate.

My role

Product Designer

Timeline

4 Sprints | Product Cycle

Team

Sole designer, 1 Product Lead, 1 Full stack develpoer

TL;DR

Challenge

Store owners and admins struggled to manage fragmented tools for reporting, analytics, and inventory visibility.

Approach

Conducted AI analysis, competitor analysis, current store manager and stakeholder interviews, and iterative dashboard prototyping with AI inclusion for insights and reports.

Soultuion

Designed a unified POS dashboard that simplified data visualization, introduced assistive AI for reporting, and improved information flow between admin and store manager sides.

Result

Reduced reporting time by 40%, improved data accuracy by 25%, and established a scalable admin-to-store data system that encouraged confident, real-time decisions.

Let's preview the final solution before going through the steps

Admin dashboard

Reports analytics

Adding new inventory

WHAT WAS BROKEN: OLD POS SYSTEM

PROJECT CONTEXT

Objective

An AI-powered POS assistant designed to help restaurant staff manage orders, inventory, and customer service seamlessly acting as a thought partner rather than just another tool.

Why This Matters

The store operates on razor-thin margins, with staff turnover rates exceeding 70% annually. Every minute spent fumbling with technology is a minute not spent creating memorable guest experiences. But here's what most people miss: the problem isn't that staff need faster buttons they need better thinking support during high-pressure moments.

The Discovery Process

I positioned myself as a user inside the system, asking:

“Where does the data come from? Who needs it? And why does it feel so hard to use?”

What I noticed:

Staff frequently interrupted each other with basic questions, managers spent significant time on repetitive inquiries and checking inventory across multiple systems, and new hires required 3-4 weeks of heavy mentorship to become confident.

Key Insight

Admin, Staff and managers weren't struggling with the POS system itself they were struggling with accessing the knowledge they needed to make good decisions quickly.

MAJORLY I FOCUSED ON

Immersing in the Admin Workflow

Mapping tasks, reports, filtering, legacy screens, and inconsistent data flows with no clear starting point.

Studying the Store Manager Side

How data moves from store to admin database, delays, missing fields, accuracy issues

Reviewing Existing Dashboards & Competitor Tools

What tools offer features but lack clarity, show numbers without meaning

-how I began

EXPLORING THE SOLUTION SPACE FOR AI

I explored Three options before landing on AI:

Option 1: Better information architecture

Reorganize menus, add search functionality, create shortcuts

Why it wasn't enough: Still requires staff to know what to look for and where to find it

Option 2: Comprehensive training program

Detailed onboarding, reference materials, cheat sheets

Why it wasn't enough: Doesn't help in real-time moments of need; information becomes outdated

Option 3: AI-powered conversational assistant

Natural language queries, contextual suggestions, proactive support

Why this made sense: Meets staff where they are, adapts to their language, provides answers in context

Why I Chose Option 3 (The Research Behind It)

Cognitive Load Theory

Based on: John Sweller's Cognitive Load Theory (1988)

Staff can only retain 5-7 pieces of information at once. Conversational AI reduces cognitive load by providing answers on demand.

Just-In-Time Learning

Inspired by: Xerox's Just-In-Time training approach

People learn best when they need the information RIGHT NOW, not 3 weeks ago in training. AI delivers knowledge exactly when decisions need to be made.

Contextual Intelligence

Research: Contextual design by Karen Holtzblatt

AI understands context (what the staff member is doing, what customer is asking) and surfaces relevant information without making them dig for it.

Aspect

Real-time Support

Reduces Interruptions

Contextual (understands situation)

Adapts to Staff Language

Works for High-Pressure Moments

Scales with Team Growth

Better UI

Manual search needed

Still interrupt supervisors

Shows all options equally

Need to integrate

Too slow under stress

Same training needed

Training

Requires recall

Still forget things

Dense + overwhelming

Need to learn

Forget under pressure

3+ weeks per person

AI Assistant

Instant answers

Self-sufficient

22% easier to interpret

Just ask naturally

Easy guide

Infinite scale

STORYBOARD: A DAY IN THE STORE FOR STAFF/MANAGER

Persona 1: Retail Cashier

Current Problems

Confusing UI - Complex menu system, hard to find products

Long Learning Curve - New staff takes 2+ weeks to feel confident

Slow Transactions - Takes 3-4 minutes per customer

High Error Rates - Wrong prices, missed items, payment failures

Customer Frustration - Impatient customers, complaints, poor reviews

Persona 2: Store Manager

Current Problems

No Real-Time Data - Dashboard outdated, reports come 24 hours late

Manual Inventory Checks - Walking around store checking shelves physically

Guessing on Orders - No clear insights on what to reorder, excess stock or stockouts

No Analytics - Can't see which products sell best, staff performance unclear

Time Wasted - Spends 3+ hours daily on manual tasks instead of strategy

Defining AI Design Principles

Before designing any interface, I established principles based on what makes AI feel helpful rather than intrusive:

1. Feel Like a Partner, Not a Black Box

Users should understand why AI suggests what it suggests
Transparency in decision-making builds trust
Application: Show confidence levels, explain reasoning, allow users to see the "why"

2. Humanize the Experience

Conversational, not robotic
Acknowledges uncertainty rather than faking confidence
Application: Natural language processing, empathetic responses, admits when it doesn't know

3. Focus on Tasks That Unlock New Outcomes

AI shouldn't just speed up existing processes—it should enable entirely new capabilities
Application: Proactive insights (inventory running low), predictive suggestions (popular pairings), learning from patterns

4. Communicate Predictable Behavior

Users should know what AI can and can't do
Consistent interaction patterns build confidence
Application: Clear capabilities, consistent response formats, explicit limitations

5. Build User Trust and Control

Users always have final say
Easy to override or ignore AI suggestions
Application: Suggestions, not commands; one-tap to dismiss; manual controls always available

-the good mess

THE BRAINSTORMING PROCESS: FINDING THE EDGE CASES

How I Approached Ideation:

I didn't start with screens, I started with scenarios. Using the principle that AI should be a thought partner, I mapped out every moment a staff member might need support.

Scenarios explored:

Server taking an order with specific preferences
Customer asked for wine pairing recommendation
Staff noticing ingredient running low
New employee's first day on the floor
Manager reviewing end-of-day analytics
System encountering an item not in the database

Testing the Limits: Edge Cases That Shaped the Design

From Concepts to Prototypes

Early Prototyping Philosophy:

Following the principle of "prototyping as a tool for adding depth," I created multiple fidelity levels to test different aspects:

-the solution

THE SYSTEM BEYOND THE UI

DEFINING WHAT I BUILT

Key areas which I an showing the study on

Reports

Dynamic and AI-summarized, replacing manual PDF exports.

Analytics Dashboard

Visualized sales, restocks, and top sellers across stores.

Data Flow System

Created transparent visibility from store transactions to the admin database, ensuring real-time sync.

//USER PERSONA FOR EACH TYPE OF SYSTEM USER

//USER JOURNEY MAPPING FOR EACH TYPE OF SYSTEM USER

-wire framing and finalizing design decisions

BUILDING THE LAYOUTS

Left Navigation for Predictability

Predictive Trend Blocks

Modular Card Layout

Central Analytics Hub

Parameter Controls Placed Above Insights

Detailed SKU Table for Granular Review

Smart Filter Bar With Dynamic Tags

Manage Tabs for Customization

Centralized Pricing Controls

Dedicated Price Reduction Scheduler

Algorithm Options for Pricing

Integration of AI

//WIREFRAMES

SYSTEM THINKING

//POS MIND MAP

//SITEMAP

NOTES THAT HAD GREATER IMPACT TOWARDS THE FINAL OUTCOME

identifying the problems

Complicated dashboards, slowed daily task

No real-time insights

Difficult to forecast demand

Manual restocking and price adjustments

No standardized multi-store reporting,

noting out solutions

simplified dashboard that surfaces the important data

Real-Time Sales & Inventory Sync

AI- driven insights for restocking and reporting

AI-powered recommendation pricing strategies

Multi-store consolidated reporting

TRANSLATING INSIGHTS INTO RESULTS

Smart Restock Predictions

AI tracked item velocity and forecasted when to reorder.

Important Narratives

Each report ended with an AI summary like “Chips sales up 12% due to weekend promotions.”

Admin Alerts

Automatic notifications about anomalies, like unexpected stock dips or licensing deadlines.

Final design outcome

Delivered a unified payment hub that removed repetitive navigation, reduced friction, and made every transaction faster, clearer, and more intuitive.

MEASURING THE IMPROVEMENT

I asked participants (admin + store manager personas) to complete the same 5 tasks

The same tets were alloted on work on the current POS systems and the new designed POS system

Generate today’s sales report with the report section.

Check weekly performance and revenue

Add one item and update it in the inventory

Identify a low-stock items with stores with AI

Update the promotional pricing for a product

BEFORE AFTER COMPARISON TABLE

//POS DASHBOARD FOR EVERYDAY ANALYTICS

Design tokens

//THE MOST IMPORTANT THAT DRIVES THE BUSINESS- REPORTS

//CUSTOMIZABLE SETTINGS FOR INVENTORY PRICING

- the numbers that defined success

FINAL RESULT

32% Faster Reporting

Reporting time was reduced allowing teams to access insights instantly instead of navigating multiple disconnected tools.

15% Faster Task Completion

Clearer navigation, simplified dashboards, and visual summaries reduced the steps needed to complete daily tasks — making workflows smoother and more intuitive.

22% Reduction in Cognitive Load

The redesigned dashboard minimized clutter and turned dense information into digestible insights, helping admin users process data without overwhelm.

WHAT WOULD I DO DIFFERENTLY WITH EXTRA TIME

Deeper staff, manager and admin collaboration

What I'd do: Embed in restaurants for full shifts, not just observations
Why: Would catch micro-moments of friction I missed in shorter sessions
Impact: More nuanced understanding of stress triggers and flow states

Long-Term Learning Study

What I'd do: Deploy for 3-6 months and track how AI accuracy improves
Why: Real learning patterns only emerge over time with real data
Impact: Proof of value; insights for training improvements

Accessibility Deep Dive

What I'd do: Test with staff who have visual, hearing, or motor impairments
Why: Restaurant industry is diverse; technology should be inclusive
Impact: More universally usable design

Integration Stress Testing

What I'd do: Prototype the full technical architecture with developers
Why: AI response speed under load is critical; beautiful UI means nothing if it's slow
Impact: Realistic expectations and technical constraints surfaced early

- always a learner

KEY TAKEAWAY AND LEARNINGS

AI Should Assist, Not Replace

Transparency mattered more than complexity. Showing why an AI insight appeared built trust and usability.

Design for Flow, Not Features

Instead of adding more tools, I focused on connecting the existing ones and working on the functionality first rather than the design creating a seamless loop between admin data and the store manager’s cache.

Data Should Speak

I learned that analytics design is storytelling. Meaningful visuals transformed raw data into confident actions.

Browse more work

Trust & Security

Designed Security Flow: Smart ID at Shoptaki, boosting usability by 15% and creating engaging interfaces, creating seamless flow

Finance Multipayment hub

Revamped SmartID’s Finance Flow to streamline multi-payment actions, reducing steps by 40% and improving user control across transactions.

Live Audio Chat & Real-Time Engagement

Researched and redesigned the organizer experience for an event services and ticketing app, improving user flows, incerasing user engagement by 22%, and enhancing usability testing.

back

Designing an AI-powered POS System for Smarter Store Management

Designed an intelligent POS dashboard that unified orders, analytics, and AI insights helping store owners turn data into effortless business clarity.

Problem

Outcome

My role

Product Designer

Timeline

4 Sprints | Product Cycle

Team

Sole designer, 1 Product Lead, 1 Full stack develpoer

TL;DR

Challenge

Store owners and admins struggled to manage fragmented tools for reporting, analytics, and inventory visibility.

Approach

Conducted AI analysis, competitor analysis, current store manager and stakeholder interviews, and iterative dashboard prototyping with AI inclusion for insights and reports.

Soultuion

Designed a unified POS dashboard that simplified data visualization, introduced assistive AI for reporting, and improved information flow between admin and store manager sides.

Result

Reduced reporting time by 40%, improved data accuracy by 25%, and established a scalable admin-to-store data system that encouraged confident, real-time decisions.

Let's preview the final solution before going through the steps

Admin dashboard

Reports analytics

Adding new inventory

WHAT WAS BROKEN: OLD POS SYSTEM

PROJECT CONTEXT

Objective

An AI-powered POS assistant designed to help restaurant staff manage orders, inventory, and customer service seamlessly acting as a thought partner rather than just another tool.

Why This Matters

The Discovery Process

I positioned myself as a user inside the system, asking:

“Where does the data come from? Who needs it? And why does it feel so hard to use?”

What I noticed:

Key Insight

Admin, Staff and managers weren't struggling with the POS system itself they were struggling with accessing the knowledge they needed to make good decisions quickly.

MAJORLY I FOCUSED ON

Immersing in the Admin Workflow

Mapping tasks, reports, filtering, legacy screens, and inconsistent data flows with no clear starting point.

Studying the Store Manager Side

How data moves from store to admin database, delays, missing fields, accuracy issues

Reviewing Existing Dashboards & Competitor Tools

What tools offer features but lack clarity, show numbers without meaning

-how I began

EXPLORING THE SOLUTION SPACE FOR AI

I explored Three options before landing on AI:

Option 1: Better information architecture

Reorganize menus, add search functionality, create shortcuts

Why it wasn't enough: Still requires staff to know what to look for and where to find it

Option 2: Comprehensive training program

Detailed onboarding, reference materials, cheat sheets

Why it wasn't enough: Doesn't help in real-time moments of need; information becomes outdated

Option 3: AI-powered conversational assistant

Natural language queries, contextual suggestions, proactive support

Why this made sense: Meets staff where they are, adapts to their language, provides answers in context

Why I Chose Option 3 (The Research Behind It)

Cognitive Load Theory

Based on: John Sweller's Cognitive Load Theory (1988)

Staff can only retain 5-7 pieces of information at once. Conversational AI reduces cognitive load by providing answers on demand.

Just-In-Time Learning

Inspired by: Xerox's Just-In-Time training approach

People learn best when they need the information RIGHT NOW, not 3 weeks ago in training. AI delivers knowledge exactly when decisions need to be made.

Contextual Intelligence

Research: Contextual design by Karen Holtzblatt

AI understands context (what the staff member is doing, what customer is asking) and surfaces relevant information without making them dig for it.

STORYBOARD: A DAY IN THE STORE FOR STAFF/MANAGER

Persona 1: Retail Cashier

Current Problems

Confusing UI - Complex menu system, hard to find products

Long Learning Curve - New staff takes 2+ weeks to feel confident

Slow Transactions - Takes 3-4 minutes per customer

High Error Rates - Wrong prices, missed items, payment failures

Customer Frustration - Impatient customers, complaints, poor reviews

Persona 2: Store Manager

Current Problems

No Real-Time Data - Dashboard outdated, reports come 24 hours late

Manual Inventory Checks - Walking around store checking shelves physically

Guessing on Orders - No clear insights on what to reorder, excess stock or stockouts

No Analytics - Can't see which products sell best, staff performance unclear

Time Wasted - Spends 3+ hours daily on manual tasks instead of strategy

Defining AI Design Principles

Before designing any interface, I established principles based on what makes AI feel helpful rather than intrusive:

1. Feel Like a Partner, Not a Black Box

Users should understand why AI suggests what it suggests
Transparency in decision-making builds trust
Application: Show confidence levels, explain reasoning, allow users to see the "why"

2. Humanize the Experience

Conversational, not robotic
Acknowledges uncertainty rather than faking confidence
Application: Natural language processing, empathetic responses, admits when it doesn't know

3. Focus on Tasks That Unlock New Outcomes

AI shouldn't just speed up existing processes—it should enable entirely new capabilities
Application: Proactive insights (inventory running low), predictive suggestions (popular pairings), learning from patterns

4. Communicate Predictable Behavior

Users should know what AI can and can't do
Consistent interaction patterns build confidence
Application: Clear capabilities, consistent response formats, explicit limitations

5. Build User Trust and Control

Users always have final say
Easy to override or ignore AI suggestions
Application: Suggestions, not commands; one-tap to dismiss; manual controls always available

-the good mess

THE BRAINSTORMING PROCESS: FINDING THE EDGE CASES

How I Approached Ideation:

I didn't start with screens, I started with scenarios. Using the principle that AI should be a thought partner, I mapped out every moment a staff member might need support.

Scenarios explored:

Server taking an order with specific preferences
Customer asked for wine pairing recommendation
Staff noticing ingredient running low
New employee's first day on the floor
Manager reviewing end-of-day analytics
System encountering an item not in the database

Testing the Limits: Edge Cases That Shaped the Design

From Concepts to Prototypes

Early Prototyping Philosophy:

Following the principle of "prototyping as a tool for adding depth," I created multiple fidelity levels to test different aspects:

-the solution

THE SYSTEM BEYOND THE UI

Learning and Improvement Systems

How the AI Gets Smarter

Staff feedback loop

Thumbs up/down on suggestions

Correction tracking

When staff override, system learns from the pattern

Usage analytics

Which questions are most common? Where does AI struggle?

Manager inputs

Seasonal menu changes, new policies, special events

DEFINING WHAT I BUILT

Key areas which I an showing the study on

Reports

Dynamic and AI-summarized, replacing manual PDF exports.

Analytics Dashboard

Visualized sales, restocks, and top sellers across stores.

Data Flow System

Created transparent visibility from store transactions to the admin database, ensuring real-time sync.

//USER PERSONA FOR EACH TYPE OF SYSTEM USER

//USER JOURNEY MAPPING FOR EACH TYPE OF SYSTEM USER

-wire framing and finalizing design decisions

BUILDING THE LAYOUTS

Left Navigation for Predictability

Predictive Trend Blocks

Modular Card Layout

Central Analytics Hub

Parameter Controls Placed Above Insights

Detailed SKU Table for Granular Review

Smart Filter Bar With Dynamic Tags

Manage Tabs for Customization

Centralized Pricing Controls

Dedicated Price Reduction Scheduler

Algorithm Options for Pricing

Integration of AI

//WIREFRAMES

SYSTEM THINKING

//POS MIND MAP

//SITEMAP

NOTES THAT HAD GREATER IMPACT TOWARDS THE FINAL OUTCOME

identifying the problems

Complicated dashboards, slowed daily task

No real-time insights

Difficult to forecast demand

Manual restocking and price adjustments

No standardized multi-store reporting,

noting out solutions

simplified dashboard that surfaces the important data

Real-Time Sales & Inventory Sync

AI- driven insights for restocking and reporting

AI-powered recommendation pricing strategies

Multi-store consolidated reporting

TRANSLATING INSIGHTS INTO RESULTS

Smart Restock Predictions

AI tracked item velocity and forecasted when to reorder.

Important Narratives

Each report ended with an AI summary like “Chips sales up 12% due to weekend promotions.”

Admin Alerts

Automatic notifications about anomalies, like unexpected stock dips or licensing deadlines.

Final design outcome

Delivered a unified payment hub that removed repetitive navigation, reduced friction, and made every transaction faster, clearer, and more intuitive.

MEASURING THE IMPROVEMENT

I asked participants (admin + store manager personas) to complete the same 5 tasks

The same tets were alloted on work on the current POS systems and the new designed POS system

Generate today’s sales report with the report section.

Check weekly performance and revenue

Add one item and update it in the inventory

Identify a low-stock items with stores with AI

Update the promotional pricing for a product

BEFORE AFTER COMPARISON TABLE

WHAT WOULD I DO DIFFERENTLY WITH EXTRA TIME

Deeper staff, manager and admin collaboration

What I'd do: Embed in restaurants for full shifts, not just observations
Why: Would catch micro-moments of friction I missed in shorter sessions
Impact: More nuanced understanding of stress triggers and flow states

Long-Term Learning Study

What I'd do: Deploy for 3-6 months and track how AI accuracy improves
Why: Real learning patterns only emerge over time with real data
Impact: Proof of value; insights for training improvements

Accessibility Deep Dive

What I'd do: Test with staff who have visual, hearing, or motor impairments
Why: Restaurant industry is diverse; technology should be inclusive
Impact: More universally usable design

Integration Stress Testing

What I'd do: Prototype the full technical architecture with developers
Why: AI response speed under load is critical; beautiful UI means nothing if it's slow
Impact: Realistic expectations and technical constraints surfaced early

- always a learner

KEY TAKEAWAY AND LEARNINGS

AI Should Assist, Not Replace

Transparency mattered more than complexity. Showing why an AI insight appeared built trust and usability.

Design for Flow, Not Features

Data Should Speak

I learned that analytics design is storytelling. Meaningful visuals transformed raw data into confident actions.

Browse more work

Built Trust

Product Designer

User Research

B2B

SaaS

Most impactful

Trust & Security

Designed Security Flow: Smart ID at Shoptaki, boosting usability by 15% and creating engaging interfaces, creating seamless flow

Finance Multipayment hub

Revamped SmartID’s Finance Flow to streamline multi-payment actions, reducing steps by 40% and improving user control across transactions.

Built Emotion

UX/UI Design

User Research

B2C

SaaS

Entertainment service

Most impactful

Live Audio Chat & Real-Time Engagement

Researched and redesigned the organizer experience for an event services and ticketing app, improving user flows, incerasing user engagement by 22%, and enhancing usability testing.

back

OVERVIEW

PROJECT CONTEXT

Designing an AI-powered POS System for Smarter Store Management

Designed an intelligent POS dashboard that unified orders, analytics, and AI insights helping store owners turn data into effortless business clarity.

Problem

Outcome

My role

Product Designer

Timeline

4 Sprints | Product Cycle

Team

Sole designer, 1 Product Lead, 1 Full stack developer

TL;DR

Challenge

Store owners and admins struggled to manage fragmented tools for reporting, analytics, and inventory visibility.

Approach

Conducted AI analysis, competitor analysis, current store manager and stakeholder interviews, and iterative dashboard prototyping with AI inclusion for insights and reports.

Solution

Designed a unified POS dashboard that simplified data visualization, introduced assistive AI for reporting, and improved information flow between admin and store manager sides.

Result

Reduced reporting time by 40%, improved data accuracy by 25%, and established a scalable admin-to-store data system that encouraged confident, real-time decisions.

Let's preview the final solution before going through the steps

Admin dashboard

Reports analytics

Adding new inventory

WHAT WAS BROKEN: OLD POS SYSTEM

PROJECT CONTEXT

Objective

An AI-powered POS assistant designed to help restaurant staff manage orders, inventory, and customer service seamlessly acting as a thought partner rather than just another tool.

Why This Matters

The Discovery Process

I positioned myself as a user inside the system, asking:

“Where does the data come from? Who needs it? And why does it feel so hard to use?”

What I noticed:

Key Insight

Admin, Staff and managers weren't struggling with the POS system itself they were struggling with accessing the knowledge they needed to make good decisions quickly.

MAJORLY I FOCUSED ON

Immersing in the Admin Workflow

Mapping tasks, reports, filtering, legacy screens, and inconsistent data flows with no clear starting point.

Studying the Store Manager Side

How data moves from store to admin database, delays, missing fields, accuracy issues

Reviewing Existing Dashboards & Competitor Tools

What tools offer features but lack clarity, show numbers without meaning

-how I began

EXPLORING THE SOLUTION SPACE FOR AI

I explored Three options before landing on AI:

Option 1: Better information architecture

Reorganize menus, add search functionality, create shortcuts

Why it wasn't enough: Still requires staff to know what to look for and where to find it

Option 2: Comprehensive training program

Detailed onboarding, reference materials, cheat sheets

Why it wasn't enough: Doesn't help in real-time moments of need; information becomes outdated

Option 3: AI-powered conversational assistant

Natural language queries, contextual suggestions, proactive support

Why this made sense: Meets staff where they are, adapts to their language, provides answers in context

Why I Chose Option 3 (The Research Behind It)

Cognitive Load Theory

Based on: John Sweller's Cognitive Load Theory (1988)

Staff can only retain 5-7 pieces of information at once. Conversational AI reduces cognitive load by providing answers on demand.

Just-In-Time Learning

Inspired by: Xerox's Just-In-Time training approach

People learn best when they need the information RIGHT NOW, not 3 weeks ago in training. AI delivers knowledge exactly when decisions need to be made.

Contextual Intelligence

Research: Contextual design by Karen Holtzblatt

AI understands context (what the staff member is doing, what customer is asking) and surfaces relevant information without making them dig for it.

Aspect

Real-time Support

Reduces Interruptions

Contextual (understands situation)

Adapts to Staff Language

Works for High-Pressure Moments

Scales with Team Growth

Better UI

Manual search needed

Still interrupt supervisors

Shows all options equally

Need to integrate

Too slow under stress

Same training needed

Training

Requires recall

Still forget things

Dense + overwhelming

Need to learn

Forget under pressure

3+ weeks per person

AI Assistant

Instant answers

Self-sufficient

22% easier to interpret

Just ask naturally

Easy guide

Infinite scale

STORYBOARD: A DAY IN THE STORE FOR STAFF/MANAGER

Persona 1: Retail Cashier

Current Problems

Confusing UI - Complex menu system, hard to find products

Long Learning Curve - New staff takes 2+ weeks to feel confident

Slow Transactions - Takes 3-4 minutes per customer

High Error Rates - Wrong prices, missed items, payment failures

Customer Frustration - Impatient customers, complaints, poor reviews

Persona 2: Store Manager

Current Problems

No Real-Time Data - Dashboard outdated, reports come 24 hours late

Manual Inventory Checks - Walking around store checking shelves physically

Guessing on Orders - No clear insights on what to reorder, excess stock or stockouts

No Analytics - Can't see which products sell best, staff performance unclear

Time Wasted - Spends 3+ hours daily on manual tasks instead of strategy

Defining AI Design Principles

Before designing any interface, I established principles based on what makes AI feel helpful rather than intrusive:

1. Feel Like a Partner, Not a Black Box

Users should understand why AI suggests what it suggests
Transparency in decision-making builds trust
Application: Show confidence levels, explain reasoning, allow users to see the "why"

2. Humanize the Experience

Conversational, not robotic
Acknowledges uncertainty rather than faking confidence
Application: Natural language processing, empathetic responses, admits when it doesn't know

3. Focus on Tasks That Unlock New Outcomes

AI shouldn't just speed up existing processes—it should enable entirely new capabilities
Application: Proactive insights (inventory running low), predictive suggestions (popular pairings), learning from patterns

4. Communicate Predictable Behavior

Users should know what AI can and can't do
Consistent interaction patterns build confidence
Application: Clear capabilities, consistent response formats, explicit limitations

5. Build User Trust and Control

Users always have final say
Easy to override or ignore AI suggestions
Application: Suggestions, not commands; one-tap to dismiss; manual controls always available

THE BRAINSTORMING PROCESS: FINDING THE EDGE CASES

How I Approached Ideation:

I didn't start with screens, I started with scenarios. Using the principle that AI should be a thought partner, I mapped out every moment a staff member might need support.

Scenarios explored:

Server taking an order with specific preferences
Customer asked for wine pairing recommendation
Staff noticing ingredient running low
New employee's first day on the floor
Manager reviewing end-of-day analytics
System encountering an item not in the database

Testing the Limits: Edge Cases That Shaped the Design

From Concepts to Prototypes

Early Prototyping Philosophy:

Following the principle of "prototyping as a tool for adding depth," I created multiple fidelity levels to test different aspects:

Learning and Improvement Systems

How the AI Gets Smarter

Staff feedback loop

Thumbs up/down on suggestions

Correction tracking

When staff override, system learns from the pattern

Usage analytics

Which questions are most common? Where does AI struggle?

Manager inputs

Seasonal menu changes, new policies, special events

DEFINING WHAT I BUILT

Key areas which I an showing the study on

Reports

Dynamic and AI-summarized, replacing manual PDF exports.

Analytics Dashboard

Visualized sales, restocks, and top sellers across stores.

Data Flow System

Created transparent visibility from store transactions to the admin database, ensuring real-time sync.

//USER PERSONA FOR EACH TYPE OF SYSTEM USER

//USER JOURNEY MAPPING FOR EACH TYPE OF SYSTEM USER

SYSTEM THINKING

//POS MIND MAP

//SITEMAP

NOTES THAT HAD GREATER IMPACT TOWARDS THE FINAL OUTCOME

identifying the problems

Complicated dashboards, slowed daily task

No real-time insights

Difficult to forecast demand

Manual restocking and price adjustments

No standardized multi-store reporting,

noting out solutions

simplified dashboard that surfaces the important data

Real-Time Sales & Inventory Sync

AI- driven insights for restocking and reporting

AI-powered recommendation pricing strategies

Multi-store consolidated reporting

-wire framing and finalizing design decisions

BUILDING THE LAYOUTS

Left Navigation for Predictability

Predictive Trend Blocks

Modular Card Layout

Central Analytics Hub

Parameter Controls Placed Above Insights

Detailed SKU Table for Granular Review

Smart Filter Bar With Dynamic Tags

Manage Tabs for Customization

Centralized Pricing Controls

Dedicated Price Reduction Scheduler

Algorithm Options for Pricing

Integration of AI

//WIREFRAMES

TRANSLATING INSIGHTS INTO RESULTS

Smart Restock Predictions

AI tracked item velocity and forecasted when to reorder.

Important Narratives

Each report ended with an AI summary like “Chips sales up 12% due to weekend promotions.”

Admin Alerts

Automatic notifications about anomalies, like unexpected stock dips or licensing deadlines.

Final design outcome

Delivered a unified payment hub that removed repetitive navigation, reduced friction, and made every transaction faster, clearer, and more intuitive.

MEASURING THE IMPROVEMENT

I asked participants (admin + store manager personas) to complete the same 5 tasks

The same tets were alloted on work on the current POS systems and the new designed POS system

Generate today’s sales report with the report section.

Check weekly performance and revenue

Add one item and update it in the inventory

Identify a low-stock items with stores with AI

Update the promotional pricing for a product

BEFORE AFTER COMPARISON TABLE

WHAT WOULD I DO DIFFERENTLY WITH EXTRA TIME

Deeper staff, manager and admin collaboration

What I'd do: Embed in restaurants for full shifts, not just observations
Why: Would catch micro-moments of friction I missed in shorter sessions
Impact: More nuanced understanding of stress triggers and flow states

Long-Term Learning Study

What I'd do: Deploy for 3-6 months and track how AI accuracy improves
Why: Real learning patterns only emerge over time with real data
Impact: Proof of value; insights for training improvements

Accessibility Deep Dive

What I'd do: Test with staff who have visual, hearing, or motor impairments
Why: Restaurant industry is diverse; technology should be inclusive
Impact: More universally usable design

Integration Stress Testing

What I'd do: Prototype the full technical architecture with developers
Why: AI response speed under load is critical; beautiful UI means nothing if it's slow
Impact: Realistic expectations and technical constraints surfaced early

- always a learner

KEY TAKEAWAY AND LEARNINGS

AI Should Assist, Not Replace

Transparency mattered more than complexity. Showing why an AI insight appeared built trust and usability.

Design for Flow, Not Features

Data Should Speak

I learned that analytics design is storytelling. Meaningful visuals transformed raw data into confident actions.

Browse more work

Built Trust

Product Design

Finance

Security

Mobile

User Research

B2B

SaaS

Most impactful

Trust & Security

Designed Security Flow: Smart ID at Shoptaki, boosting usability by 15% and creating engaging interfaces, creating seamless flow

Finance Multipayment hub

Revamped SmartID’s Finance Flow to streamline multi-payment actions, reducing steps by 40% and improving user control across transactions.

Built Emotion

UX/UI Design

User Research

B2C

SaaS

Entertainment service

Most impactful

Live Audio Chat & Real-Time Engagement

Researched and redesigned the organizer experience for an event services and ticketing app, improving user flows, increasing user engagement by 22%, and enhancing usability testing.