Heaji Wi

AI 기반 냉난방 의사결정 지원 시스템 AI-powered HVAC decision support system

Operator

복잡한 냉난방 판단을 간단한 행동으로 Turning complex HVAC judgment into simple actions

매장 운영 중 반복되는 냉난방 판단 부담을 줄이기 위해,
비용과 쾌적도를 함께 고려하는 AI 기반 의사결정 지원 서비스를 설계했습니다.
To ease the HVAC judgment calls that keep coming back during store operations, I designed an AI-based decision-support service that weighs cost and comfort together.

  • AI Recommendations
  • Decision Support
  • Product Design

팀 Team

2 Product Designer

프로젝트 타입 · 기간 Project type · Duration

Concept project · 7주 Concept project · 7 weeks

역할 Role

  • Context Research — Co-led
  • System Design — Co-led
  • User Research — Co-led
  • Usability Testing — Led
Operator dashboard on a tablet

습도상승으로 냉방효율저하 감지

제습우선모드 + 목표온도 25도 조정

오퍼레이터가 추천솔루션을 적용 완료했습니다!

제습우선모드 + 목표온도 25도 조정

적용 완료

1-1

Start point 시작점

Running a store, I kept facing the same cooling-cost problem. 매장을 운영하며 반복되는 냉방비 문제를 경험했습니다.

When the store is busy, there's no time to check and fine-tune the HVAC.
You run the cooling hard so customers stay comfortable, and only feel the cost a month later when the bill arrives.

Knowing how to cut costs and actually doing it in the middle of operations turned out to be two different things.
매장이 바쁠 때는 냉난방 상태를 세밀하게 확인하고 조절할 여유가 없습니다.
일단 손님들이 쾌적하게 머물 수 있도록 냉방을 강하게 가동하고, 비용은 한 달 뒤 고지서를 받고 나서야 실감하게 됩니다.

비용을 줄이는 방법을 아는 것과 실제 운영 중에 이를 실행하는 것은 다른 문제였습니다.

So is this problem unique to one store, or does it repeat across the many spaces that manage heating and cooling? 그렇다면 이 문제는 특정 매장만의 문제일까요, 아니면 냉난방을 관리하는 다양한 공간에서 반복되는 문제일까요?

1-2

Target 타깃

I judged that the cost burden and the way heating and cooling are managed could differ by environment.
So I compared households, large buildings and small business stores by cost impact and whether a management system was already in place.
문제가 발생하는 환경마다 비용 부담과 관리 방식이 다를 수 있다고 판단했습니다.
이에 비용 영향과 기존 관리 체계의 유무를 기준으로 가정, 대형 건물, 자영업 매장을 비교했습니다.

  • Households 가정

    • Low consumption 낮은 사용량
    • Low cost impact 비용 영향 낮음
    • No dedicated management system 별도 관리 시스템 없음
  • Large buildings 대형 건물

    • High consumption 높은 사용량
    • High cost impact 비용 영향 큼
    • EMS in place EMS 관리시스템 존재
    • Already managed and optimized 이미 관리 및 최적화가 이뤄지는 영역
  • Selected 선택

    Small business stores 자영업 매장

    • High consumption 높은 사용량
    • High cost impact 비용 영향 큼
    • No dedicated energy
      management system
      별도 에너지관리 시스템 없음
    • Reliant on individual judgment 개별 판단에 의존

I set small business stores as the target. 자영업 매장을 타깃으로 설정했습니다.

For households, the business impact of cutting costs is relatively limited, and large buildings often already have an energy management system.
Small business stores, by contrast, feel HVAC costs heavily yet lack a way to manage them around day-to-day operations.
가정은 비용 절감의 사업적 영향이 상대적으로 제한적이고, 대형 건물은 기존 에너지 관리 시스템을 갖춘 경우가 많았습니다.
반면 자영업 매장은 냉난방 비용의 영향이 크면서도, 운영 상황에 맞춰 관리할 수 있는 체계가 부족하다고 판단했습니다.

So I set small business stores as the main target, and confirmed the real problem through community research and interviews. 이에 자영업 매장을 주요 타깃으로 설정하고, 커뮤니티 리서치와 인터뷰를 통해 실제 문제를 확인했습니다.

2-1

Secondary research 2차 리서치

Without clear criteria, HVAC adjustments
became a matter of trial and error.
명확한 기준이 없어,
냉난방 조절은 시행착오의 반복이 되었습니다.

I collected and categorized data from small business owner communities to explore the problems that keep recurring in HVAC management.

The analysis showed a pattern: owners weighed cost against comfort, yet changed settings without clear criteria,
and each responded in their own way depending on the store's situation.

This suggested that the difficulty of managing HVAC may not be a simple control problem, but one tied to the absence of decision criteria.
자영업자 커뮤니티 데이터를 수집하고 분류해, 냉난방 관리 과정에서 반복적으로 나타나는 문제를 탐색했습니다.

분석 결과, 비용과 쾌적함 사이에서 고민하면서도 명확한 기준 없이 설정을 바꾸고,
매장 상황에 따라 각자 다른 방식으로 대응하는 패턴이 나타났습니다.

이를 통해 냉난방 관리의 어려움이 단순한 조작 문제가 아니라, 판단 기준의 부재와 관련 있을 수 있다고 보았습니다.

01 Data collection 데이터 수집

Collected community posts
Collected community posts 커뮤니티 데이터 게시글 모음
Categorized community data analysis
Categorized community data analysis 커뮤니티 데이터 카테고리화 분석

02 Patterns 패턴

Analyzing community data revealed recurring patterns. 자영업자 커뮤니티 데이터를 분석한 결과 다음과 같은 패턴이 반복적으로 나타났습니다.

  • 1. Repeated adjustments from missing criteria 1. 기준 부재로 인한 반복 조정

    Without criteria, anxiety drives constant HVAC tweaking 기준이 없어 불안할 때마다 냉난방을 계속 바꾸게 된다

  • 2. Caught between comfort and cost 2. 쾌적함 vs 비용 사이의 고민

    Wanting comfort but worried about cost, the deliberation never stops 쾌적하게 유지하고 싶지만, 비용이 걱정돼 계속 고민하게 된다

  • 3. Hard to adjust across store situations 3. 매장별 상황에 따른 조절의 어려움

    Every store's situation differs, so it's hard to apply a consistent adjustment criterion 매장마다 상황이 달라 일관된 조절 기준을 적용하기 어렵다

  • 4. Acting first, judging never 4. 판단보다 조작에 의존하는 행동

    Rather than weighing the decision, they jump straight to manipulating the controls 어떻게 판단할지보다, 일단 조작으로 해결하려고 한다

03 Observation 관찰

Small business owners repeatedly adjusted HVAC settings as store conditions changed, but lacked clear criteria for judging which settings were right. 자영업자들은 매장 상황에 따라 냉난방 설정을 반복적으로 조정했지만, 어떤 설정이 적절한지 판단할 명확한 기준은 부족했습니다.

2-2

Primary research 1차 리서치

HVAC settings couldn't be decided on cost alone. 냉난방 설정은 비용만으로 결정할 수 없었습니다.

01 Interview 인터뷰

Interview protocol
Interview protocol 인터뷰 프로토콜

To see in what context the lack of decision criteria found in the community analysis shows up in real store operations, we interviewed 5 small business owners.

We explored when they adjust settings, what influences their decisions,
and how they assess the outcome.
커뮤니티 분석에서 발견한 판단 기준의 부재가 실제 매장 운영에서 어떤 맥락으로 나타나는지 확인하기 위해 자영업자 5명을 인터뷰했습니다.

언제 설정을 바꾸는지, 어떤 요인이 결정에 영향을 주는지,
조절 결과를 어떻게 판단하는지 살펴봤습니다.

02 Data Analysis 데이터 분석

Affinity mapping
Affinity mapping 어피니티 맵핑
Thematic coding
Thematic coding Thematic coding

Organizing the interviews through affinity mapping and thematic analysis showed that owners struggled to tell how much a settings change affected cost,
and that while they responded differently to each store situation, they lacked clear criteria for deciding.

In short, users were adjusting their HVAC, but had little basis for judging whether those choices worked.
인터뷰 내용을 어피니티 맵핑과 주제별 분석으로 정리한 결과, 설정을 바꿔도 비용이 얼마나 달라지는지 알기 어렵고,
매장 상황에 따라 다르게 대응하면서도 명확한 판단 기준이 부족하다는 점을 확인했습니다.

즉, 사용자는 냉난방을 조절하고 있었지만 그 선택이 효과적이었는지 판단할 근거는 부족했습니다.

03 Findings 발견

  • 1. Hard to predict the outcome 1. 결과를 예측하기 어렵다

    It's hard to know how much the electricity bill changes after a settings change, so they can't be sure of the result. 설정을 바꿔도 전기요금이 얼마나 달라지는지 알기 어려워, 조절 결과를 확신하기 어렵다.

  • 2. Torn between cost and comfort 2. 비용과 쾌적도 사이의 고민

    They want to cut HVAC costs, but turning settings down can make customers uncomfortable, so adjusting isn't easy. 냉난방 비용을 줄이고 싶지만, 설정을 낮추면 고객이 불편해질 수 있어 쉽게 조절하기 어렵다.

  • 3. They adjust by intuition 3. 직관에 의존해 조절한다

    Without clear criteria, they adjust settings based on how the store feels or on past experience. 명확한 판단 기준이 없어, 매장에서 느껴지는 체감이나 과거 경험을
    바탕으로 설정을 조정한다.

04 Insight! 인사이트!

The owner's decision dilemma 운영자의 의사결정 딜레마

Owners wanted to cut costs, but couldn't sacrifice customer comfort. 비용을 줄이고 싶어도, 고객의 쾌적함을 희생할 수는 없었습니다.

The interviews confirmed that owners' HVAC decisions hinge not only on cost but also on customer comfort. 인터뷰를 통해 운영자의 냉난방 결정에는 비용뿐 아니라 고객의 쾌적함도 중요하게 작용한다는 점을 확인했습니다.

2-3

Context research 맥락 리서치

To weigh cost and comfort together,
we first had to understand how HVAC costs are determined.
비용과 쾌적도를 함께 고려하려면,
냉난방 비용이 결정되는 구조부터 이해해야 했습니다.

01 Energy Pricing Structure 에너지 요금 구조

Pricing structure analysis
Pricing structure analysis 요금구조 분석 이미지

Energy costs are determined by multiple variables, including peak, mid-peak, and off-peak time slots,
as well as progressive billing based on usage.
에너지 비용은 피크·중간·경부하 시간대와
사용량에 따른 누진 과금 등 여러 변수로 결정됩니다.

For busy small business owners, understanding how these variables affect the final bill
is difficult during daily operations.
운영에 바쁜 자영업자가 이 변수들이 최종 요금에
어떤 영향을 주는지 매일 따져보기는 어렵습니다.

Building on the cost-management difficulties found in the interviews, we examined whether the pricing structure could serve as a basis for cost-saving decisions in a real store. 인터뷰에서 확인한 비용 관리의 어려움을 바탕으로, 요금 구조가 실제 매장에서 비용 절감의 판단 기준이 될 수 있는지 검토했습니다.

02 Scenario Test 시나리오 테스트

Scenario diagram
Scenario diagram 시나리오 이미지

We built store-operation scenarios that account for time-of-use rates and usage, and compared the results of adjusting settings to minimize cost. 시간대별 요금과 사용량을 고려한 매장 운영 시나리오를 구성하고,
비용을 최소화하는 방식으로 설정을 조정했을 때의 결과를 비교했습니다.

Cost savings came with trade-offs 비용 절감에는 대가가 따랐습니다

Some scenarios showed potential savings of around 5–10% a month, but these assumed a specific pricing structure and usage conditions. In a real store, applicability could vary with operating hours, weather and usage patterns. Cutting costs could also conflict with keeping the space comfortable.

So rather than aiming HVAC management at cost savings alone, we broadened the goal to managing cost efficiently while accounting for comfort.
일부 시나리오에서는 월 5~10% 수준의 절감 가능성을 확인했지만, 이는 특정 요금 구조와 사용 조건을 가정한 결과였습니다. 실제 매장에서는 운영시간과 날씨, 사용 패턴 등에 따라 적용 가능성이 달라질 수 있었습니다. 또한 비용을 줄이는 방식이 쾌적한 환경 유지와 충돌할 가능성도 있었습니다.

따라서 냉난방 관리의 목표를 비용 절감에만 두기보다,
쾌적도를 고려하면서 비용을 효율적으로 관리하는 것으로 확장했습니다.

03 Insight! 인사이트!

HVAC management needed decision criteria
that weigh cost and comfort together.
냉난방 관리에는 비용과 쾌적도를
함께 고려하는 판단 기준이 필요했습니다.

Minimizing cost alone could not fully reflect how a real store operates. 비용만을 최소화하는 방식으로는 실제 매장의 운영 맥락을 충분히 반영하기 어려웠습니다.

2-4

Competitor research 경쟁사 리서치

Existing HVAC systems supported checking status and making adjustments, but didn't help with decisions that weigh cost and comfort. 기존 냉난방 시스템은 상태 확인과 조작은 지원했지만, 비용과 쾌적도를 고려한 의사결정은 돕지 못했습니다.

We looked at how existing systems show the current state, such as temperature and humidity, and how they help users change settings. 기존 시스템이 온도와 습도 등 현재 상태를 어떻게 보여주고, 사용자가 설정을 어떻게 변경하도록 돕는지 살펴봤습니다.

Competitor analysis spreadsheet
Competitor analysis spreadsheet 경쟁사 분석 엑셀

This led us to note that checking status and making adjustments alone make it hard for users to judge which setting to choose.

Operator aimed to go beyond showing status, setting itself apart by offering adjustment options and the expected effect of each.
이를 통해 상태를 확인하고 조작하는 기능만으로는 사용자가 어떤 설정을 선택해야 할지 판단하기 어렵다는 점에 주목했습니다.

Operator는 상태를 보여주는 데서 나아가, 선택 가능한 조정안과 그에 따른 예상 효과를 제공하는 방향으로 차별화하고자 했습니다.

3

Problem statement 문제 정의

The problem with HVAC management wasn't simple control —
it was how hard it is to keep making decisions that fit operations.
냉난방 관리의 문제는 단순한 조작이 아니라,
운영 상황에 맞는 판단을 지속하기 어렵다는 것이었습니다.

Store owners want to manage HVAC costs efficiently, but during busy operations it's hard to keep checking the store environment and adjusting settings.
Yet putting cost savings first can lose the comfort customers need.
매장 운영자는 냉난방 비용을 효율적으로 관리하고 싶지만, 바쁜 운영 중에는 매장 환경을 지속적으로 확인하고 설정을 조정하기 어렵습니다.
그렇다고 비용 절감만을 우선하면 고객에게 필요한 쾌적도를 놓칠 수 있습니다.

4-1

Design direction 디자인 방향

We shifted from operating HVAC directly
to supporting better decisions.
냉난방을 직접 조작하는 방식에서,
더 나은 판단을 지원하는 방식으로 전환했습니다.

Competitor analysis showed that existing systems focused on checking HVAC status and controlling devices. What they lacked, relatively, was helping users see how an adjustment affects cost and the store environment, and decide their next action on that basis. 경쟁 서비스 분석 결과, 기존 시스템은 냉난방 상태를 확인하고 기기를 제어하는 기능에 집중하고 있었습니다.
반면, 조정이 비용과 매장 환경에 어떤 영향을 미치는지 파악하고, 이를 바탕으로 다음 행동을 결정하도록 돕는 기능은 상대적으로 부족했습니다.

Going beyond status checks and control, Operator was designed so that AI presents adjustments and their expected effects to fit the store's situation, and users review them and decide. This lets owners weigh cost and comfort together while easing the burden of checking information and making a judgment every time. Operator는 단순한 상태 확인과 제어를 넘어, AI가 매장 상황에 맞는 조정안과 예상 효과를 제시하고 사용자가 이를 검토해 직접 결정하는 방식을 설계했습니다.
이를 통해 운영자가 매번 정보를 확인하고 판단해야 하는 부담을 줄이면서, 비용과 쾌적도를 함께 고려할 수 있도록 했습니다.

Before - Existing systems Before - 기존 시스템

After - Our approach After - 우리의 접근

4-2

Solution direction 솔루션 방향

AI recommends HVAC adjustments,
and users decide whether to approve them.
AI가 냉난방 조정안을 추천하고,
사용자가 승인 여부를 결정합니다.

We envisioned an experience that proposes adjustments based on the current store state and the user's cost and comfort priorities, and shows the expected change alongside. 현재 매장 상태와 비용·쾌적도에 대한 우선순위를 바탕으로 조정안을 제안하고, 예상되는 변화를 함께 보여주는 경험을 구상했습니다.

  1. 01 Input 입력

    • Store temperature & humidity 매장의 온도·습도
    • Outdoor temperature 외부 기온
    • Time of day 시간대
    • User adjustment history 사용자 조작 이력
    • Time-of-use electricity pricing 시간대별 전기요금 구조
  2. 02 Criteria 판단 기준

    • Priority between cost and comfort 비용과 쾌적도 간 우선순위
    • Expected cost 예상 비용
    • Comfort change 쾌적도 변화
    • Using records from similar past situations 과거 유사 상황의 기록 활용
  3. 03 Output 출력

    • Candidate adjustments and expected results 후보 조정안과 예상 결과
    • Recommendation and its rationale 추천안과 추천 근거

*Rather than implementing a real AI model's prediction performance or HVAC device control, this project focused on designing and exploring a product experience that reduces owners' judgment and operating burden through AI's adjustment suggestions and expected effects. *본 프로젝트는 실제 AI 모델의 예측 성능이나 냉난방 기기 제어를 구현하는 대신, AI의 조정 제안과 예상 효과를 통해 운영자의 판단과 조작 부담을 줄이는 제품 경험을 설계하고 탐색하는 데 집중했습니다.

4-3

Ideal process 아이디에이션 과정

We turned users' key decision points into features and set MVP priorities. 사용자의 주요 판단 지점을 기능으로 구체화하고,
MVP의 우선순위를 정했습니다.

Building on the HVAC management difficulties found in research, we laid out what information users need to check and what decisions they need to make.

We turned these into features, separating what AI would suggest from what users would choose themselves.
We then selected the MVP features by importance and implementation priority.
리서치에서 발견한 냉난방 관리의 어려움을 바탕으로, 사용자가 어떤 정보를 확인하고 어떤 결정을 내려야 하는지 정리했습니다.

이를 기능으로 구체화하고, AI가 제안할 부분과 사용자가 직접 선택할 부분을 구분했습니다.
이후 중요도와 구현 우선순위를 기준으로 MVP에 포함할 기능을 선정했습니다.

01 Feature Definition 기능 정의

User stories
User stories 유저 스토리
Feature categorization
Feature categorization 기능 카테고리화

We identified the key decision points users face during daily operations and translated them into features,
separating what AI handles automatically from what remains under user control.
사용자가 매일 운영 중에 마주하는 주요 판단 지점을 식별해 기능으로 옮기고,
AI가 자동으로 처리할 부분과 사용자가 통제할 부분을 나눴습니다.

02 Feature Prioritization 기능 우선순위

MoSCoW prioritization
MoSCoW MoSCoW
2×2 prioritization matrix
2×2 Matrix 2×2 매트릭스

We evaluated the feature set by importance and implementation priority,
narrowing it down to the features essential for the MVP.
기능을 중요도와 구현 우선순위로 평가해, MVP에 꼭 필요한 기능으로 좁혔습니다.

03 MVP feature set MVP 기능

The prioritization process narrowed the product down to a focused MVP. 우선순위 정리를 거쳐 제품을 핵심에 집중한 MVP로 좁혔습니다.

  • Real-time comfort status 실시간 쾌적도 상태
  • AI recommendation AI 추천
  • Cost & comfort comparison 비용 · 쾌적도 비교
  • Predicted outcome 예상 결과
  • Open / Peak / Close guidance 오픈 / 피크 / 마감 가이드
  • Daily summary 일일 요약
  • Mode Setting 모드 설정

4-4

Structure design 구조 설계

Users decide quickly in the pop-up, and check the reasoning and expected effects on the main screen. 팝업에서는 빠르게 판단하고, 메인 화면에서는
추천의 근거와 기대 효과를 확인하도록 설계했습니다.

01 User flow 유저 플로우

Pop-up first, details when needed 팝업 먼저, 필요할 때 자세히

In busy moments, users can check the current state and the recommended action at a glance in the pop-up and approve it. When they need more detail, the main screen lets them compare the expected cost, comfort change and recommended solution together. 매장이 바쁜 순간에는 팝업에서 현재 상태와 추천 행동을 간결하게 확인하고 승인할 수 있도록 했습니다. 더 자세한 정보가 필요한 경우에는 메인 화면에서 예상 비용, 쾌적도 변화, 추천 솔루션을 함께 비교하도록 구성했습니다.

02 Information Architecture 정보 구조

Information architecture
Information Architecture 정보 구조

Four key data points, one screen 핵심 정보 네 가지, 한 화면에

We gathered the key information on one screen, so users can review a recommendation without moving between screens. 핵심 정보를 한 화면에 모아 여러 화면을 오가지 않고도
추천을 검토할 수 있도록 했습니다.

  • Predicted cost 예상 비용
  • Cumulative cost 누적 비용
  • Comfort change 쾌적도 변화
  • Recommended solution 추천 솔루션

03 Crazy 8 · Wireframes · Design System Crazy 8 · 와이어프레임 · 디자인 시스템

We explored layout ideas, refined the screen structure, and established visual consistency. 레이아웃 아이디어를 탐색하고, 화면 구조를 다듬고, 시각적 일관성을 확립했습니다.

Crazy 8 Crazy 8
Wireframes
Wireframe 와이어프레임
Design system
Design System 디자인 시스템

5-1

Key Features 핵심 기능

01 INSTANT DECISIONS VIA POP-UPS 팝업으로 즉시 결정

Even during busy operations, users can check a recommendation in the pop-up and approve it quickly. 바쁜 운영 중에도 팝업에서 추천을 확인하고
빠르게 승인할 수 있습니다.

The pop-up briefly shows the current state, the reason for the recommendation and the expected change after applying it, with visual emphasis focused on the key action: approval. This aimed to spare users from stopping store work to navigate between screens. 팝업에는 현재 상태와 추천 이유, 적용 후 예상 변화를 간결하게 보여주고, 핵심 행동인 승인에 시각적 강조를 집중했습니다.
이를 통해 사용자가 매장 업무를 중단하고 여러 화면을 탐색해야 하는 부담을 줄이고자 했습니다.

Design goal 설계 목표

Support quick decisions on the spot 현장에서 빠른 의사결정을 지원

  • Open / Peak / Close guidance ✅ 오픈 / 피크 / 마감 가이드 ✅
  • Real-time comfort status ✅ 실시간 쾌적도 상태 ✅
  • AI recommendation ✅ AI 추천 ✅

5-2

Key Features 핵심 기능

02 DATA LOG & LEARNING 데이터 로그 & 학습

We envisioned a data log so users can look back on how each decision affected cost and comfort. 각 결정이 비용과 쾌적도에 미친 영향을 돌아볼 수 있도록 데이터 기록 기능을 구상했습니다.

We envisioned a data log so users can check the results of their HVAC adjustments and help improve future recommendations.

A record of adjustments and the resulting cost and comfort changes lets users look back on past choices. Longer term, we envisioned using the accumulated records to understand each store's patterns and preferences.

AI learning from real accumulated data, and the resulting gain in recommendation accuracy, were outside the scope of implementation and validation.
사용자가 냉난방 조절의 결과를 확인하고,
향후 추천을 개선할 수 있도록 데이터 기록 기능을 구상했습니다.

냉난방 조절과 비용·쾌적도 변화의 기록을 통해 사용자가
과거의 선택을 돌아볼 수 있도록 설계했습니다.
장기적으로는 누적된 기록을 매장별 패턴과 선호를 이해하는 데 활용하는 방향을 구상했습니다.

실제 데이터 축적을 통한 AI 학습이나 추천 정확도 향상은
구현 및 검증 범위에 포함되지 않았습니다.

Design goal 설계 목표

Record and visualize cost and comfort changes 비용·쾌적도 변화의 기록 및 시각화

  • Daily summary ✅ 일일 요약 ✅
  • Cost & comfort comparison ✅ 비용 · 쾌적도 비교 ✅
  • Predicted outcome ✅ 예상 결과 ✅

5-3

Key Features 핵심 기능

03 MODE SETTING 모드 설정

Users choose whether cost or comfort takes priority. 비용과 쾌적도 중 어떤 기준을 우선할지
사용자가 선택하도록 했습니다.

We offered a cost-management mode and a comfort-first mode so users can choose the criterion they want to fit the store's situation. AI recommendations were envisioned to reflect the priority the user chooses. 비용 관리 모드와 쾌적도 우선 모드를 제공해, 매장 운영 상황에 따라 사용자가 원하는 기준을 선택할 수 있도록 설계했습니다.
AI 추천은 사용자가 선택한 우선순위를 반영하는 방향으로 구상했습니다.

Design goal 설계 목표

Recommendations that reflect the user's cost and comfort priorities 사용자의 비용·쾌적도 우선순위를 반영한 추천

  • Mode Setting ✅ 모드 설정 ✅

6-1

Usability test 사용성 테스트

We tested the decision-making flow
with 5 participants across 4 tasks.
참가자 5명, 4개 과제로
판단 흐름을 테스트했습니다.

01 UT Process UT 과정

A participant running the test at a store counter
UT Photo UT 사진

We simulated real store operations to evaluate whether users could: 실제 매장 운영 상황을 가정해, 사용자가 다음을 할 수 있는지 평가했습니다.

  1. Understand the current store state 현재 매장 상태 이해하기
  2. Understand how the reason for a recommendation relates to the store's situation 추천 이유와 매장 상황의 연관성 이해하기
  3. Decide whether to approve the recommendation 추천을 승인할지 결정하기
  4. Interpret the expected cost and comfort changes 예상 비용·쾌적도 변화 해석하기
UT protocol spreadsheets
UT Protocol UT 프로토콜

02 SUS Score SUS 점수

76

The average SUS score was 76 (n = 5; individual scores: 72.5–80), above the commonly cited benchmark of 68. SUS 평균 점수는 76점(n = 5, 개별 72.5–80)으로, 흔히 기준으로 삼는 68점을 웃돌았습니다.

Given the small sample, we treated this as directional usability evidence rather than a definitive measure of product performance. 표본이 작아, 제품 성과를 확정하는 지표가 아니라 사용성의 방향을 확인하는 참고값으로 해석했습니다.

03 Problems 문제

01 — What Went Wrong 01 — 문제 발생

Users struggled to understand how AI recommendations applied to their own store.

We ran a usability test with 5 participants. The core problem it revealed wasn't only that they didn't know how to operate it. They found it hard to fully understand how AI produces its recommendations, and how those recommendations connect to their own store's situation.
사용자는 AI 추천이 자신의 매장에 어떻게 적용되는지
이해하기 어려워했습니다.

5명을 대상으로 사용성 테스트를 진행했습니다. 테스트에서 확인한 핵심 문제는 단순히 조작 방법을 몰랐다는 것만이 아니었습니다.
AI가 추천을 만드는 방식과 그 추천이 자신의 매장 상황에 어떻게 연결되는지 충분히 이해하기 어려워했습니다.

02 — Why It Happened 02 — 원인

How recommendations were personalized wasn't conveyed well enough, and the sample data and expected effects didn't resonate with users.

In the prototype at the time, the way store-specific settings personalize recommendations didn't come across clearly. As a result, users couldn't be sure an AI recommendation fit their own store.

And because the sample data and expected effects on screen didn't connect to their actual store, users found it hard to feel what they meant for them.
추천의 개인화 과정이 충분히 전달되지 않았고, 가상의 데이터와 예상 효과도 사용자에게 와닿지 않았습니다.

당시 프로토타입에서는 매장별 설정을 통해 추천이 개인화되는 과정이 충분히 전달되지 않았습니다. 그 결과 사용자는 AI 추천이 자신의 매장 상황에 맞는 것인지 확신하기 어려웠습니다.

또한 화면에 제시된 가상의 데이터와 예상 효과가 실제 매장 상황과 연결되지 않아, 본인에게 어떤 의미가 있는지 체감하기 어려워했습니다.

6-2

Iteration 개선

We improved onboarding and how information is presented,
so users can understand what store-specific recommendations mean.
사용자가 매장별 추천의 의미를 이해할 수 있도록
온보딩과 정보 제공 방식을 개선했습니다.

01 The Problem 문제

Users found it hard to understand how AI recommendations reflected their own store's situation, and struggled to pick out what was meaningful to them in the sample data and expected effects. 사용자는 AI 추천이 자신의 매장 상황을 어떻게 반영하는지 이해하기 어려웠고,
가상의 데이터와 예상 효과에서도 본인에게 의미 있는 정보를 파악하기 어려워했습니다.

02 The Solution 해결

GUIDED ONBOARDING
Making an unfamiliar model easier to understand
가이드 온보딩
낯선 방식을 더 쉽게 이해하도록

We strengthened onboarding so users can understand how the product works and the role of store-specific settings. It also explains the difference between the cost-management and comfort modes, and shows expected savings over a month and a year, so users can grasp what a recommendation means more concretely.

After a pop-up approval, we added the next step and concrete action guidance so users can follow the flow that comes after.
제품의 작동 방식과 매장별 설정의 역할을 이해할 수 있도록 온보딩을 보완했습니다. 또한 비용 관리와 쾌적도 모드의 차이를 설명하고, 한 달과 일 년 기준의 예상 절감 효과를 함께 제시해 추천의 의미를 더 구체적으로 파악할 수 있도록 했습니다.

팝업 승인 이후에는 다음 단계와 구체적인 행동 안내를 추가해 사용자가 이어지는 흐름을 이해할 수 있도록 개선했습니다.

What changed 개선 내용

  • Guided users through the AI-based HVAC operating model and store-specific setup AI 기반 냉난방 운영 방식과 매장별 설정 과정을 안내했습니다
  • Explained the difference between the cost and comfort modes, and their expected effects, in concrete terms 비용과 쾌적도 모드의 차이 및 예상 효과를 구체적으로 설명했습니다

After the improvements, feedback from colleagues showed they could grasp the product's core concepts more easily. However, since the improved design wasn't retested with the same users, it's hard to say the usability gains have been proven. 개선 후 동료들의 피드백에서는 제품의 핵심 개념을 더 쉽게 이해할 수 있었다는 반응을 확인했습니다. 다만 동일한 사용자에게 개선안을 다시 테스트하지는 못했기 때문에,
실제 사용성 개선 효과가 입증됐다고 보기는 어렵습니다.

7

Reflection 회고

01 Experience Is a Starting Point, Not Evidence 경험은 근거가 아니라 출발점입니다

Running a store helped me spot the problem quickly, but I didn't assume from my own experience alone that other stores had the same problem. Community research, interviews and pricing-structure analysis confirmed the problem's scope and validity. 매장 운영 경험을 통해 문제를 빠르게 발견했지만, 개인 경험만으로 다른 매장에도 같은 문제가 있다고 단정하지 않았습니다.
커뮤니티 리서치와 인터뷰, 요금 구조 분석을 통해 문제의 범위와 타당성을 확인했습니다.

I learned that personal experience is a starting point for finding problems, but what can be generalized has to be judged through research. 개인 경험은 문제를 발견하는 출발점이지만, 무엇을 일반화할 수 있는지는 리서치로 판단해야 한다는 점을 배웠습니다.

02 Complex Information Must Become an Understandable Decision Structure 복잡한 정보는 이해할 수 있는 판단 구조로 바뀌어야 합니다

Usability alone is not enough. Users need to understand the reasoning behind the AI's judgment before they can trust its recommendations and use them in their decisions. I learned that this means designing not only the structure of information, but an experience that makes the AI's decision process understandable. 사용성만으로는 충분하지 않습니다. 사용자가 AI의 판단 근거를 이해할 수 있어야 추천을 신뢰하고 결정에 활용할 수 있습니다. 이를 위해 정보의 구조뿐 아니라 AI의 판단 과정을 이해할 수 있는 경험을 설계해야 한다는 점을 배웠습니다.

Good UX doesn't stop at making information easy to use; it helps users understand and trust the system's judgment. 좋은 UX는 정보를 쉽게 사용하는 데서 그치지 않고, 사용자가 시스템의 판단을 이해하고 신뢰할 수 있도록 돕는 것입니다.

03 What Remains to Be Validated 아직 검증이 남은 것

This was an exploratory project based on 5 interviews and a usability test with 5 participants. I couldn't re-validate the revised onboarding with the same participants, and real HVAC control and cost savings were not confirmed either.

Next, I want to test the improved experience again and validate how well recommendations are understood, and how effective they are, in real store conditions.
이 프로젝트는 인터뷰 5명과 사용성 테스트 5명을 바탕으로 한 탐색적 프로젝트였습니다.
개선한 온보딩을 동일한 참가자에게 재검증하지 못했고, 실제 냉난방 제어와 비용 절감 효과도 확인하지 못했습니다.

다음 단계에서는 개선된 경험을 다시 테스트하고, 실제 매장 상황에서 추천의 이해도와 효과를 검증하고자 합니다.

A clear design direction is only the beginning; its impact still needs to be validated. 명확한 디자인 방향은 시작일 뿐, 그 효과는 여전히 검증이 필요합니다.