AI and Data-Informed Interior Design

Introduction

Interior design is increasingly shaped by evidence generated from buildings, users, environmental sensors, project records and digital design tools. This development does not replace professional judgement. Rather, it gives designers additional ways to understand how spaces are occupied, how environments perform and how users experience them. Data-informed design uses evidence to support decisions; artificial intelligence (AI) extends this process by identifying patterns, generating alternatives, interpreting images and automating routine work.

The central professional challenge is not simply learning to use AI software. It is deciding what evidence is trustworthy, what should remain under human control and how design values such as dignity, inclusion, cultural meaning and beauty can be protected. Interior designers must therefore combine spatial reasoning, environmental knowledge, research literacy, ethical judgement and technical competence.

Learning Objectives

By the end of this chapter, students should be able to:

1. Explain how occupancy, environmental and user data inform interior design decisions.
2. Select suitable data-collection methods and distinguish useful evidence from misleading data.
3. Describe applications of AI in research, ideation, visualisation and workflow automation.
4. Critically assess accuracy, bias, privacy, authorship, intellectual property and human oversight.
5. Develop a transparent, human-centred workflow for using data and AI in a design project.
6. Communicate evidence-based recommendations to clients and project teams.

1. From Intuition to Evidence-Informed Practice

Design intuition is valuable because it integrates experience, tacit knowledge and sensitivity to context. However, intuition can also reproduce assumptions. A designer may believe that a lounge is underused because it is unattractive when the real cause is glare, poor acoustics, inconvenient access or a mismatch between opening hours and user schedules.

Data-informed design introduces a disciplined cycle:

Question → collect evidence → interpret patterns → generate options → test → decide → evaluate.

Data is not the same as truth. A sensor records what it is configured to detect; a survey records what respondents choose to report; an AI model produces an output based on its training and instructions. Each source has scope, error and bias. Professional judgement is required to connect evidence to design action.

Three useful data categories

A strong evidence base normally combines quantitative and qualitative data. A count may show that a meeting room is rarely occupied; interviews may reveal that users avoid it because its acoustic privacy is inadequate. The count identifies a pattern, while the interview helps explain it.

2. Occupancy and Spatial Performance

Occupancy information concerns who uses a space, when, for how long and for what purpose. Sources include manual observation, room-booking systems, badge data, Wi-Fi or Bluetooth sensing, cameras, pressure mats and anonymous people-counting sensors. Every method involves trade-offs between accuracy, cost, intrusiveness and privacy.

Two basic indicators are useful. The occupancy rate compares observed occupants with a stated capacity:

The capacity must be defined consistently, and the observation period must be stated. A room with an average of 12 occupants and a design capacity of 20 has an average occupancy rate of 60%. This does not mean that the room is unsuitable: peak occupancy, duration and activity type may be more important than the average.

The utilisation rate measures use over time:

For example, if a studio is occupied for 30 of 50 available hours in a week, its utilisation rate is 60%. Occupancy rate concerns intensity at a particular time; utilisation concerns temporal use. They should not be treated as interchangeable.

A designer should examine distributions rather than only averages. A room might have a weekly utilisation of 50% while being completely full every morning and empty in the afternoon. The appropriate response could be flexible furniture, shared booking, improved scheduling or additional space—not necessarily a larger room.

Data quality questions

Before using occupancy data, ask:

• What exactly was measured: people, visits, bookings or duration?
• Were sensors calibrated and were there blind spots?
• Does the dataset include weekends, seasonal variation and unusual events?
• Are some users excluded because they do not use booking systems or digital devices?
• Could monitoring change user behaviour?
• Is the proposed design decision proportionate to the quality of evidence?

3. Environmental and User Data

Environmental data supports decisions about thermal comfort, lighting, air quality and acoustics. Common measurements include air temperature, relative humidity, illuminance, sound pressure level and carbon dioxide concentration. These measurements are indicators, not complete descriptions of experience. For example, carbon dioxide can indicate ventilation adequacy in an occupied space, but it is not a complete measure of indoor air quality.

A useful data record includes the measurement, unit, location, time, instrument and conditions. A temperature reading without time and location is difficult to interpret. Sensors should be calibrated or checked against a reliable reference, and anomalous readings should be investigated rather than silently deleted.

User research adds lived experience. Methods include semi-structured interviews, focus groups, diary studies, accessibility walk-throughs, post-occupancy questionnaires and behaviour mapping. Questions should avoid leading respondents toward a preferred design. Participation should be voluntary, and users should know how their responses will be used.

A mixed-method interpretation might identify the following:

• Sensors show excessive afternoon illuminance near a west-facing window.
• Observations show users moving away from the window after 15:00.
• Interviews report screen glare and visual fatigue.
• The design response is tested through shading, screen orientation and lighting controls.

This is stronger than concluding from a single measurement that the entire lighting system is defective.

4. A Framework for Data-Informed Decisions

A practical framework is the D-A-T-A cycle:

1. Define the decision and the people affected.
2. Acquire relevant, proportionate and lawful evidence.
3. Test and interpret data quality, uncertainty and competing explanations.
4. Act and assess through prototypes, implementation and post-occupancy review.

Figure: A cyclical framework for connecting evidence, interpretation and design action.

The cycle prevents a common error: collecting large quantities of data without a clear decision. It also makes evaluation part of design rather than an afterthought.

5. Artificial Intelligence in Interior Design

AI refers broadly to computational systems that perform tasks associated with human intelligence, such as classification, prediction, language processing, image analysis and generation. In interior design, AI applications can be grouped into four areas.

5.1 Research and analysis

AI can summarise large document collections, extract themes from interview transcripts, classify photographs, compare specifications and identify repeated issues in post-occupancy records. It can also assist with coding qualitative data, although the designer must check whether themes have been incorrectly merged or stripped of context.

5.2 Ideation and generative options

Text-to-image and image-to-image systems can generate mood studies, material combinations, spatial atmospheres and rapid alternatives. These images are useful for conversation and exploration, but they are not automatically buildable designs. They may contain impossible structures, inaccurate materials, unsafe circulation, inaccessible layouts or invented products.

Generative design tools can vary parameters such as area allocation, adjacency, daylight exposure or workstation density. The designer must define meaningful constraints and judge the results against human, cultural, technical and regulatory requirements.

5.3 Visualisation and communication

AI can assist with image enhancement, style variations, background removal, upscaling, virtual staging and rendering. It can help clients compare options, but visual realism may create false confidence. A persuasive image is not evidence that a scheme meets fire safety, accessibility, acoustic or construction requirements.

5.4 Workflow automation

AI can draft meeting summaries, populate schedules, create preliminary room data sheets, identify missing specification fields and support project correspondence. Automation is most appropriate for repetitive, reviewable tasks. It is risky when it silently changes contractual information, compliance statements or cost assumptions.

6. Designing Effective AI Prompts and Workflows

An effective prompt describes the task, context, constraints, audience, output format and evaluation criteria. For example, instead of asking an AI system to “design a sustainable office,” a designer might specify the building type, user group, area, accessibility objectives, reuse requirements, daylight conditions, budget range and requested output.

A responsible workflow is:

1. State the design question.
2. Provide only authorised and necessary information.
3. Identify constraints and non-negotiable standards.
4. Ask for alternatives and explicit assumptions.
5. Check outputs against drawings, specifications, standards and user evidence.
6. Record the tool, date, prompt purpose, source material and human revisions.
7. Obtain approval before issuing work externally.

AI should be treated as an assistant whose output requires verification, not as an expert whose authority is assumed. Retrieval systems that use a controlled project library can reduce unsupported answers, but they can still misinterpret documents or rely on outdated files.

7. Critical Issues: Bias, Accuracy, Privacy and Authorship

Bias and representation

AI systems learn patterns from data. If training images overrepresent particular cultures, body types, income groups, architectural styles or geographic regions, generated recommendations may reproduce those patterns. Bias can also enter through the designer’s selection of data. A workplace survey conducted only during daytime may exclude cleaners, security staff, shift workers and people with caring responsibilities.

Bias mitigation includes diverse participation, disaggregated analysis, accessibility review, comparison of alternative interpretations and consultation with affected communities. The goal is not to pretend that data is neutral, but to identify whose experience is visible and whose is missing.

Accuracy and uncertainty

Generative systems can produce fluent but false text, incorrect dimensions, fictional products and unsupported citations. Image systems may misrepresent colour, texture and scale. Designers must verify:

• dimensions and quantities;
• product identity, availability and performance claims;
• regulatory and accessibility implications;
• environmental declarations and sustainability claims;
• calculations, sources and dates;
• compatibility with the actual building and services.

Confidence should not be inferred from polished presentation. Where uncertainty matters, it should be stated explicitly.

Privacy and data protection

Occupancy and user data may become personal data when individuals can be identified directly or indirectly. Facial images, device identifiers, access records and detailed movement histories are particularly sensitive. Good practice includes data minimisation, purpose limitation, informed notice, access controls, retention limits, secure storage and deletion procedures. Anonymisation should be tested rather than assumed; combining several datasets can re-identify people.

Designers should not upload confidential client drawings, personal information or commercially sensitive data into an unapproved public AI service. Organisational policy and applicable data-protection law must guide every project.

Authorship and intellectual property

Authorship is not solved merely by naming a software tool. The designer remains responsible for professional decisions and issued work. Records should distinguish client-provided material, designer-created work, AI-assisted material and third-party content. Image-generation tools may have different licence conditions, training-data policies and restrictions on commercial use. A project team should confirm rights before using generated or transformed material publicly or commercially.

8. Human Oversight and Governance

Human oversight is strongest when responsibility is assigned rather than vaguely promised. A project can use an AI register containing the tool, purpose, input type, output, reviewer, verification method and approval status. High-impact decisions—such as accessibility, safety, surveillance, allocation of resources or user classification—require qualified human review and, where appropriate, stakeholder consultation.

A useful risk classification is:

Governance should address procurement, data ownership, security, auditability, accessibility, model updates, incident reporting and the right to reject an automated recommendation. A design manager should also consider whether automation changes professional roles, workload or accountability.

9. Worked Application: Flexible Learning Commons

The following is an illustrative case, not a report of a real project.

Background

A university intends to refurbish a 600 m² learning commons used by individual learners, project groups and staff. Complaints include lack of quiet study space, glare near the south façade and underused fixed computer desks.

Problem and evidence

For four representative weeks, anonymised counts and observations are collected. The space is available for 60 hours per week. Average weekday occupancy is 96 people, while the assessed operational capacity is 160 people. Peak observations reach 148 people. Booking data shows that group rooms are heavily booked between 10:00 and 14:00 but lightly used late in the day. A survey receives 210 responses; 68% report glare at some point, and interviews identify acoustic distraction as a major concern.

The average occupancy rate is:

This average conceals a peak rate of:

The data suggests that total capacity is not the only issue. Time-based peaks, zoning and environmental performance require attention.

AI-supported analysis

A controlled AI tool is used to cluster anonymised survey comments into preliminary themes. The designer checks a sample of original responses and finds that the tool has combined “privacy” and “quietness,” which users distinguish. The themes are therefore revised manually. An image-generation tool produces three atmosphere studies for a proposed gradient from collaborative to quiet space. These are used only for discussion; dimensions and specifications are developed separately in CAD and schedules.

Design response

The team proposes movable acoustic screens, a quiet zone away from the entrance, adjustable solar shading, fewer fixed computers and more power-accessible shared tables. A booking review encourages late-day use of group rooms. A small pilot zone is tested for two weeks before full implementation.

Outcome and evaluation

After implementation, post-occupancy evaluation compares peak occupancy, glare reports, acoustic satisfaction and room bookings with the baseline. The results are interpreted cautiously because the evaluation period has different examination schedules. The team documents the evidence, limitations and decisions rather than claiming that AI caused the improvement.

Learning points

1. Average occupancy alone would have hidden peak pressure.
2. User comments were necessary to interpret numerical patterns.
3. AI accelerated theme sorting and visual exploration but did not determine the design.
4. A pilot reduced risk before committing to the full refurbishment.
5. Evaluation must account for changing conditions and alternative explanations.

10. Contemporary Professional Practice and Best Practices

Contemporary practice is moving toward digital twins, sensor-linked building management systems, generative space planning, computer vision, natural-language interfaces and automated specification checking. These developments may improve adaptability and operational efficiency, but they also increase dependence on data infrastructures and vendors. Designers should ask who owns operational data, who can access it, how long it is retained and what happens when a platform changes or fails.

Best practice can be summarised as follows:

• Begin with a human and spatial question, not a fascination with technology.
• Use the minimum data necessary for the stated purpose.
• Combine measurements with observation and lived experience.
• Record assumptions, uncertainty, sources and version dates.
• Test sensors and validate AI outputs against primary evidence.
• Protect confidential and personal information.
• Include affected users, especially underrepresented groups.
• Keep a human decision-maker accountable for professional outputs.
• Use prototypes and post-occupancy evaluation.
• Disclose material AI assistance where authorship or trust could be affected.
• Treat accessibility, safety, dignity and cultural meaning as constraints—not optional improvements.

Chapter Summary

Data-informed interior design uses occupancy, environmental and user evidence to support decisions about planning, comfort, inclusion and performance. Occupancy rate and utilisation rate answer different questions and must be interpreted with time, capacity and activity context. Environmental sensors provide useful indicators, while interviews, observations and accessibility research explain how spaces are experienced.

AI can assist with research, qualitative coding, ideation, visualisation, generative options and workflow automation. Its outputs are probabilistic and may be biased, inaccurate, non-buildable or legally problematic. Responsible practice requires defined questions, proportionate data collection, verification, privacy protection, authorship records and human accountability. The most reliable workflow is cyclical: define, acquire, test and interpret, then act and assess. AI is therefore best understood as a powerful design assistant within a human-centred professional system—not as a substitute for judgement.

Key Terms

Artificial intelligence: Computational systems that perform tasks associated with human intelligence.

Bias: A systematic tendency that produces unfair, distorted or unrepresentative results.

Data-informed design: Design practice that uses collected evidence to support professional decisions.

Data minimisation: Collecting and retaining only the data necessary for a stated purpose.

Generative AI: AI that produces new text, images, layouts or other content from learned patterns.

Human oversight: Review and control by responsible people over automated systems and outputs.

Illuminance: The amount of light falling on a surface, normally measured in lux.

Occupancy rate: Observed occupants expressed as a proportion of stated capacity.

Post-occupancy evaluation: Systematic assessment of a completed space in use.

Privacy: Protection of people’s personal information, identity and reasonable expectations of confidentiality.

Prompt: An instruction or input supplied to an AI system.

Qualitative data: Non-numerical evidence such as comments, observations and interview transcripts.

Quantitative data: Numerical evidence that can be counted or measured.

Retrieval system: A system that supplies an AI model with information from a selected document collection.

Sensor calibration: Checking or adjusting an instrument against a recognised reference.

Utilisation rate: The proportion of available time during which a space is occupied.

Validation: Checking whether a method or output is fit for its intended purpose.

Verification: Checking whether an output is correct against requirements or evidence.

Visualisation: A graphic representation used to communicate a design proposal.

Review Questions

Short Answer Questions

1. Distinguish occupancy rate from utilisation rate.
2. Name three categories of data used in interior design and give one example of each.
3. Why should environmental sensor data be recorded with time and location?
4. Identify two risks associated with generative AI images.
5. What is meant by data minimisation?

Long Answer Questions

1. Explain how quantitative and qualitative evidence can be combined in a post-occupancy evaluation.
2. Discuss the benefits and limitations of AI for interior design ideation and visualisation.
3. Explain how bias can enter an occupancy or user-research project and propose mitigation measures.
4. Develop a responsible workflow for using AI to summarise confidential project documents.
5. Discuss why human oversight remains necessary when AI is used for generative space planning.

Analytical / Application Questions

1. A study room has a capacity of 24 and an average observed occupancy of 18. During peak periods it reaches 23 users. Calculate the average and peak occupancy rates, then explain why the peak figure may be more useful for planning.
2. A sensor reports low use of a collaboration area, while interviews report that users avoid it because of noise. Analyse possible explanations and propose a mixed-method investigation.
3. You are asked to upload identifiable staff movement data and confidential drawings to a public AI platform. Prepare a professional response identifying the risks, information required before proceeding and a safer alternative workflow.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes occupancy rate from utilisation rate?

A. Occupancy rate measures time used, whereas utilisation rate measures lighting quality.

B. Occupancy rate compares occupants with capacity, whereas utilisation rate compares occupied time with available time.

C. Occupancy rate measures user satisfaction, whereas utilisation rate measures acoustic performance.

D. They are identical terms used in different software systems.

Correct Answer: B. Occupancy rate compares occupants with capacity, whereas utilisation rate compares occupied time with available time.

Question 2

A room has a design capacity of 40 people and an observed occupancy of 30 people. What is its occupancy rate?

A. 40%

B. 60%

C. 75%

D. 133%

Correct Answer: C. 75%

Question 3

Why should occupancy data be combined with qualitative research?

A. Qualitative research removes the need for measurement.

B. Interviews can help explain why a numerical pattern occurs.

C. Qualitative research guarantees that data is unbiased.

D. It allows sensors to operate without calibration.

Correct Answer: B. Interviews can help explain why a numerical pattern occurs.

Question 4

Which is the most appropriate use of a generative image tool?

A. Issuing the image as proof of code compliance

B. Replacing measured drawings with an attractive visual

C. Exploring atmosphere and alternatives for discussion

D. Confirming that a product is commercially available

Correct Answer: C. Exploring atmosphere and alternatives for discussion

Question 5

What is data minimisation?

A. Deleting all project data immediately

B. Collecting only data necessary for a stated purpose

C. Compressing files to reduce storage space

D. Using the smallest possible sensor

Correct Answer: B. Collecting only data necessary for a stated purpose

Question 6

Which action best addresses possible AI bias in user research?

A. Accepting the first generated theme without review

B. Using only data from the most frequent users

C. Including diverse participants and checking outputs against original evidence

D. Removing all qualitative information

Correct Answer: C. Including diverse participants and checking outputs against original evidence

Question 7

A space is available for 50 hours and occupied for 35 hours. What is its utilisation rate?

A. 35%

B. 50%

C. 70%

D. 85%

Correct Answer: C. 70%

Question 8

Which task normally requires the strongest human oversight?

A. Drafting a non-confidential agenda

B. Generating informal mood-board keywords

C. Issuing accessibility or safety advice

D. Sorting generic reference images

Correct Answer: C. Issuing accessibility or safety advice

Question 9

Why can a polished AI-generated image be misleading?

A. AI images can never contain colour

B. Visual realism does not prove that dimensions, safety or buildability are correct

C. AI tools only generate black-and-white outputs

D. A polished image always indicates a final design

Correct Answer: B. Visual realism does not prove that dimensions, safety or buildability are correct

Question 10

What is the first stage of the D-A-T-A cycle described in the chapter?

A. Act and assess

B. Acquire

C. Define

D. Test and interpret

Correct Answer: C. Define