3  Empathize and Define

Field Research, Interview Evidence, Empathy Maps, Personas, Pain Points, POV Statements, HMW Prompts, and Success Metrics

design-methodology
thinking
empathize
define

3.1 Start With the Frustrated User

Picture an operator who knows a cold room is drifting out of range but has no trustworthy timeline, owner, or next action. Empathize-and-define work turns that messy moment into evidence: who is affected, what they are trying to decide, which pain points are real, and what success would look like before a device, dashboard, or alert workflow is designed.

3.2 Learning Objectives

By the end of this chapter, you will be able to:

  • Plan field research for an IoT product without biasing users toward a preselected device or feature.
  • Separate what users say, think, do, and feel so the team can distinguish evidence from inference.
  • Convert observations into pain points, personas, journey moments, and clear problem frames.
  • Write Point-of-View statements and How Might We prompts that keep the solution space open.
  • Define success metrics that combine user outcomes with IoT constraints such as reliability, privacy, installation, and maintenance.
In 60 Seconds

Empathize and Define turn a vague connected-product idea into a defensible problem frame. First, gather field evidence from users and affected stakeholders. Then synthesize the evidence into empathy maps, pain points, personas, and journey moments. Finally, write a Point-of-View statement, create How Might We prompts, and define success metrics before the team commits to sensors, connectivity, or custom hardware.

3.3 Empathy Changes Architecture

In IoT, empathize and define work should change what the team builds. Field research may reveal that the problem is not “add a sensor” but “make responsibility visible after hours,” “make setup recoverable without Wi-Fi,” or “avoid exposing bystanders in a shared space.” Those insights affect hardware placement, radio choice, local feedback, notification routing, and support operations.

The Define stage turns messy field notes into a problem frame that engineering can act on. A good Point-of-View statement names the user, context, need, and reason without prescribing BLE, LoRaWAN, Wi-Fi, MQTT, computer vision, a dashboard, or a mobile app before the team knows which path fits the situation.

For example, a facilities team that asks for a “smart freezer sensor” may really need a freezer-loss recovery workflow. Interviews may show that overnight staff cannot identify which freezer alarm is real, that the Wi-Fi access point drops when a metal door closes, and that the person who can move stock does not receive the alert. The problem frame then becomes an operational handoff problem with sensing, networking, escalation, and accountability parts. That frame may lead to a DS18B20 probe plus door switch, local buzzer mute, battery-backed gateway, MQTT QoS 1 publish, SMS escalation, and a support playbook instead of only a nicer dashboard.

The same discipline prevents over-engineering. If the evidence says the pain is two minutes of uncertain setup during installation, the first prototype may be a paper provisioning flow, a Figma status screen, or a bench ESP32 that shows pairing states clearly. If the pain is auditability after an incident, the prototype may be an event log and acknowledgement model before any enclosure design. Empathy work is therefore not a soft preface to real engineering; it is the mechanism that decides which engineering risk is worth spending time on next.

Field evidence loop moving from observe and interview to synthesize, define, and gate the next prototype.
Empathize and Define form an evidence loop: observe the real context, synthesize patterns, frame the problem, and gate the next prototype.
  • Person boundary: Include installer, maintainer, support agent, data subject, alert recipient, and buyer when those roles affect the product.
  • Place boundary: Capture power, mounting, enclosure, light, noise, weather, network coverage, safety, and shared-space constraints before choosing hardware.
  • Decision boundary: Frame the decision or action that must improve, then let sensing, connectivity, automation, and UI follow from that decision.

3.4 Research the Whole Service Path

A useful IoT research plan follows the event from origin to response. For a cold-room alert, that path may include probe placement, door state, local display, gateway connectivity, MQTT publish, cloud rule, APNs or FCM notification, shift handoff, acknowledgement, support ticket, and incident closeout. Interviewing only the person who wants the dashboard misses most of that path.

Use mixed methods: contextual inquiry for the physical task, interviews for intent and trust, support-ticket review for repeated failures, telemetry review for timing and state gaps, and a journey map for handoffs. Tools such as Miro, FigJam, Figma, Jira, Linear, Zendesk, Node-RED, Home Assistant, AWS IoT Core, Azure IoT Hub, or a simple MQTT broker can support the work, but the artifact is only useful if it preserves context and points to a next decision.

Plan the research sample around roles and failure modes, not around convenience. For a home-energy monitor, the sample may need a renter, homeowner, installer, call-centre agent, utility program manager, and someone who opts out of sharing fine-grained data. For an industrial vibration monitor, include maintenance technicians, line supervisors, safety staff, network administrators, and the person who signs the downtime report. Each interview should separate observed facts from interpretation: “the gateway LED was red for three hours” is evidence; “operators do not care about reliability” is usually a weak inference.

Strong field notes connect a quote or observation to a decision the team must make. A photo of the mounting location informs enclosure ingress rating, cable strain relief, antenna clearance, and tamper risk. A support-ticket cluster about failed pairing informs BLE provisioning recovery, QR-code fallback, captive-portal wording, and what diagnostics the agent needs in Zendesk. A telemetry gap around shift changes informs alert throttling, role routing, and acknowledgement ownership. The output should be a traceable board of evidence, assumptions, open questions, and prototype tests, not a decorative persona poster.

  1. Recruit by role and scenario. Include the person who installs, uses, maintains, receives alerts, appears in data, and owns operational risk.
  2. Observe failure and recovery. Ask how the team handles offline devices, stale readings, dead batteries, pairing failure, false alerts, and manual override.
  3. Map the information path. Trace what the device knows, what the app shows, who receives messages, and who can safely act.
  4. Define success before ideation. Name the user outcome, technical behavior, privacy constraint, and support requirement that a later prototype must test.

3.5 Findings to System Constraints

Research findings become useful when they map to system constraints. “Technicians check the local display before opening the app” can become a requirement for device-side status, readable labels, offline diagnostics, and local acknowledgement. “Residents worry about who sees leak data” can become role-based access, retention limits, and a clear sharing model.

Translate each insight into actors, states, events, and measurements. Actors may include device, gateway, user, installer, automation rule, support agent, and external service. States may include unpaired, provisioning, normal, pending, offline, stale, faulted, acknowledged, overridden, and recovered. Events need timestamps, source, confidence, freshness, and ownership so later prototypes can test the real workflow.

The output should be a small constraint set that links user need to technical behavior: BLE provisioning must recover after a failed QR scan; Matter commissioning must explain what failed; a device shadow or twin must show desired versus reported state; an alert rule must avoid paging the wrong role; and a support console must show firmware, battery, last heartbeat, and recent command outcomes.

Make the constraint set precise enough that a prototype can fail honestly. A vague finding such as “users need trust” is not testable. A stronger constraint says that after a sensor misses three reporting intervals, the interface must show stale status, last successful heartbeat, battery estimate, network path, and next recovery action within one screen. A vague finding such as “installation is confusing” becomes a commissioning state machine with entry criteria, timeout handling, retry copy, and logs for QR scan, BLE pairing, Wi-Fi credential exchange, cloud claim, and first telemetry publish.

This translation also exposes privacy and safety obligations early. If research shows that the device observes a shared corridor, the data model needs purpose limitation, field-level retention, access roles, and possibly edge redaction before cloud upload. If an alert can trigger a physical response, the system needs confidence thresholds, duplicate suppression, audit records, manual override, and clear ownership for who acts. The Define stage is complete only when the team can point from every major product claim to an observed need, a system behavior, a measurable outcome, and a prototype or field test that can prove or disprove the assumption.

  • Finding to state: Turn observed confusion into explicit product states, labels, and recovery paths.
  • Finding to metric: Measure task completion, correct alert classification, setup recovery, acknowledgement time, and privacy comprehension.
  • Finding to prototype: Choose paper flow, Figma screen, bench device, MQTT mock, or field pilot based on the current uncertainty.

3.6 Why Empathize Comes First

IoT products operate in messy physical environments. A device may be installed by one person, configured by another, monitored by a third, and affect people who never touch the app. Discovery therefore has to include the full service context, not only the buyer or the most enthusiastic user.

The main risk is not that the team cannot build the device. The main risk is building a device that fits the team’s imagined workflow instead of the real one.

3.7 Research Scope for IoT

Start by naming the people and contexts that may change the problem definition. A narrow participant list creates a narrow product.

Primary User

Person Doing the Task

The person whose workflow, effort, risk, or decision should improve because of the product.

Installer

Person Who Sets It Up

The person who mounts, pairs, powers, calibrates, labels, or explains the device.

Maintainer

Person Who Keeps It Working

The person who replaces batteries, responds to faults, updates firmware, or handles support calls.

Affected Stakeholder

Person Impacted by Data

The person who receives alerts, appears in data, is evaluated by metrics, or owns the risk.

Planning Red Flag

If the team has only interviewed the buyer, it has not finished empathy work. IoT products often fail at installation, support, maintenance, privacy, or alert response rather than at the feature demo.

3.8 Research Methods

Use more than one method when the stakes are high or the context is unfamiliar. Interviews reveal stories and motivations. Observation reveals behavior and workarounds. Support records reveal repeated failure patterns. Field trials reveal what changes under real constraints.

Method
Best For
Evidence Output
IoT Watchpoint
Contextual observation
Seeing how the task is actually performed in the real environment.
Workflow notes, photos or sketches where allowed, constraints, workarounds.
Notice power access, mounting surfaces, network coverage, safety, lighting, noise, and weather.
Open interview
Understanding goals, frustrations, language, trust, and decision criteria.
Quotes, stories, recurring needs, objections, and unresolved questions.
Ask about the last real incident, not whether the user wants a feature.
Journey mapping
Finding handoffs, waiting, repeated effort, alert fatigue, and support burden.
Timeline, touchpoints, pain points, actors, and decision moments.
Include setup, failure, maintenance, data review, and end-of-life moments.
Diary or field log
Capturing repeated events that do not happen during one visit.
Time-stamped events, triggers, context, user response, and comments.
Useful for intermittent connectivity, comfort, energy, alarms, and maintenance tasks.
Existing evidence review
Understanding known problems before collecting new data.
Support tickets, incident reports, maintenance logs, telemetry, compliance notes.
Existing data may show symptoms but not the user’s reasoning or hidden workarounds.

3.9 Interview Guide

Good interview questions ask for behavior, context, and recent examples. Poor questions ask users to design the product for you.

Use
Avoid
Reason
“Walk me through the last time this happened.”
“Would you use a smart device for this?”
Recent behavior is better evidence than speculative agreement.
“What did you do next, and why?”
“Do you want automatic alerts?”
Sequence and reasoning reveal decision points and constraints.
“Who else was involved?”
“Is the app easy enough?”
IoT workflows often include handoffs, permissions, and affected stakeholders.
“What makes this hard to trust?”
“Would privacy be a concern?”
Trust concerns become more concrete when tied to a real situation.
“Show me where this happens.”
“Where should we install the sensor?”
Observation exposes physical constraints before the team chooses hardware.
Consent and Data Care

Field research can collect sensitive information about homes, workplaces, habits, health, safety, security, and operations. Explain what will be recorded, get informed consent, minimize identifiable data, and avoid collecting details the team does not need.

3.10 Empathy Maps

An empathy map is a synthesis tool, not a decoration. It helps the team compare direct quotes, inferred thoughts, observed behaviors, and emotional signals. The most useful insights often appear when the quadrants disagree.

Empathy map template with Says, Thinks, Does, and Feels quadrants for organizing interview and observation findings.
The empathy map separates interview quotes, inferred concerns, observed actions, and emotional signals so the team can compare evidence types before defining the problem.
Quadrant
What Belongs There
Example Evidence
Design Use
Says
Direct quotes from interviews, support calls, or field sessions.
“I only check it when something feels wrong.”
Use the user’s language in problem framing and onboarding copy.
Thinks
Inferred beliefs, assumptions, worries, or mental models.
The user may believe the device is hard to reset after a fault.
Mark as inference until later research confirms it.
Does
Observable behavior, workaround, sequence, or repeated action.
Silences alerts, then checks the equipment manually.
Find workflow friction and places where automation could help or harm.
Feels
Emotional signals, stress, embarrassment, confidence, relief, or distrust.
Looks anxious when asked who receives alerts.
Design trust, control, and dignity into the product.

3.11 From Evidence to Pain Points

Pain points should connect evidence to a user outcome. Avoid writing them as missing features.

Weak Pain Point
Better Pain Point
Why It Is Better
“Needs a dashboard”
“The operator cannot tell whether an alert is urgent without walking to the equipment.”
Names the user, situation, and consequence without prescribing UI.
“Needs more sensors”
“The maintainer cannot locate which part of the machine changed state before the fault.”
Focuses on the decision gap that sensing might address.
“Needs cloud analytics”
“The supervisor needs a trend view because single incidents do not reveal repeated misuse.”
Explains why history and aggregation may matter.
“Needs a mobile app”
“The field technician needs confirmation at the device because network access is unreliable in the basement.”
May lead away from app-first design and toward local feedback.

Pain points usually fall into several categories:

  • Functional: the task is slow, unreliable, physically hard, or error-prone.
  • Emotional: the user feels anxiety, embarrassment, distrust, or loss of control.
  • Operational: the handoff, maintenance, support, or compliance process breaks down.
  • Financial: the current process wastes time, materials, energy, or service visits.
  • Social or privacy-related: the product affects relationships, surveillance concerns, dignity, or accountability.

3.12 Personas and Journey Moments

A persona is useful only when it comes from real evidence. It should not become a fictional biography. Keep it compact and operational.

Persona

Evidence-Based Segment

Name the role, goals, context, constraints, technical comfort, and decision criteria. Tie each claim to research evidence.

Journey Moment

Critical Situation

Show the moment where the user makes a decision, receives an alert, performs setup, or recovers from failure.

Constraint

Design Boundary

Record physical, social, technical, safety, privacy, or support constraints that shape the solution space.

Evidence Gap

Next Research Need

Name what the team still does not know before ideation, prototyping, or architecture commitment.

3.13 Define the Problem

The Define stage turns research into a problem frame that can guide ideation and prototype testing.

Point-of-View formula:

[Specific user] needs [verb-based need] because [research insight].

The need should be a verb phrase, not a product. The insight should come from evidence, not the team’s preference.

Problem Frame
Issue
Improved Version
“Users need a smart lock.”
Generic user, solution noun, no insight.
Short-term rental hosts need to grant time-bounded access without meeting guests in person because key handoff delays disrupt remote check-in.
“Factories need predictive maintenance.”
Broad domain phrase, no actor, no decision.
Line supervisors need to recognize repeated vibration changes before a shift ends because repairs are only practical during scheduled downtime.
“Residents need a water sensor.”
Prescribes device type before testing the workflow.
Apartment residents need early confidence that a leak is contained because they are accountable to neighbors and cannot always reach building staff quickly.

3.14 How Might We Prompts

How Might We prompts bridge Define and Ideate. They should be open enough to invite options and constrained enough to keep the team tied to evidence.

Start with the POV Use the specific user, need, and insight as the source.
Remove the solution Do not name the sensor, network, algorithm, or app unless it is already evidence-backed.
Keep the user outcome Preserve the decision, relief, safety, coordination, or maintenance outcome.
Create several prompts Write prompts for trust, setup, failure, privacy, maintenance, and data use.

Examples:

  • How might we help field technicians confirm a device is healthy when the network is unavailable?
  • How might we let caregivers know that a routine is on track without making the user feel watched?
  • How might we help operators distinguish urgent alerts from noisy background events?
  • How might we make setup recoverable for a non-expert installer in a physically constrained space?

3.15 Define Success Metrics

Success metrics should reflect the problem frame. In IoT, the metric set should combine user outcomes with device, data, and service constraints.

Metric Type
Question
Example Metric
Evidence Source
User outcome
Did the user’s task, confidence, safety, or decision improve?
Correct response to an alert during a field scenario.
Usability session, field trial, interview follow-up.
Operational outcome
Did the workflow reduce support burden, repeated effort, or delay?
Fewer manual checks after reliable status evidence is available.
Support logs, maintenance records, journey review.
Technical behavior
Does the device behave well under real constraints?
Alert delivery, offline handling, battery behavior, sensor accuracy.
Lab test, field telemetry, reliability test.
Trust and privacy
Do users understand, accept, and control the data flow?
Consent comprehension, role-based access, unwanted alert escalation.
Research session, privacy review, support feedback.

3.16 Shared Cold-Room Monitoring

Scenario: A university lab group wants a connected monitor for shared cold rooms. The original request is “add more sensors and a dashboard.” The team pauses to empathize and define before choosing hardware.

Research plan:

  • Observe lab staff during sample storage, retrieval, and end-of-day checks.
  • Interview researchers, lab managers, maintenance staff, and safety officers.
  • Review incident logs, access rules, support tickets, and current alarm response steps.
  • Map the journey from temperature anomaly to user action and incident closeout.

Research findings:

Evidence
Interpretation
Risk if Ignored
Design Implication
Staff silence alarms during loading.
The alarm may be correct but poorly timed.
Alert fatigue and missed true events.
Support temporary context-aware suppression with timestamped event history.
Researchers are unsure who owns after-hours response.
The technical alert is not enough; ownership is unclear.
Slow response even when data is available.
Include escalation roles and visible acknowledgement.
Maintenance staff distrust readings after door-open events.
Raw temperature alone lacks context.
Unnecessary service calls or ignored warnings.
Record door state, recovery trend, and confidence notes.
Safety officers need incident records.
The system must support accountability, not only alerts.
Manual reconstruction after incidents.
Keep event history and acknowledgement context by equipment area.

Empathy map synthesis:

  • Says: “The alarm goes off all the time during normal loading.”
  • Thinks: “I do not know whether this is my responsibility after hours.”
  • Does: Silences alarms, checks the display manually, sends informal messages.
  • Feels: Worried about losing samples but annoyed by repeated non-actionable alerts.

POV statement:

Lab staff sharing cold-room equipment need to distinguish normal loading events from risky temperature excursions because repeated ambiguous alarms reduce trust and delay the right response.

HMW prompts:

  • How might we help staff see whether an alarm is actionable before they ignore it?
  • How might we make after-hours responsibility visible without broadcasting sensitive research details?
  • How might we show recovery trends after a door-open event instead of only showing a threshold crossing?
  • How might we preserve incident evidence for safety review without overloading routine users?

Success metrics:

  • Staff can correctly classify a simulated event as normal loading, watch, or urgent response.
  • Alerts include owner, context, and acknowledgement state.
  • Field test records show fewer repeated manual checks after trusted status is available.
  • Safety review can reconstruct what happened without collecting unnecessary personal data.

3.17 Evidence Record

Use a short record after each research and synthesis cycle. Keep it brief enough that the team will maintain it.

Evidence record flow linking observation, role, context, risk, requirement, evidence, and next test trigger.
An evidence record keeps research findings traceable from observation and role context through risk, requirement, evidence, and the next test trigger.
Field
What to Record
Why It Matters
Research question
What uncertainty was the session meant to reduce?
Prevents vague research and makes findings traceable.
Participant context
Role, environment, workflow, constraints, and excluded perspectives.
Shows whether evidence covers the real service context.
Observed evidence
Quotes, behaviors, artifacts, handoffs, exceptions, and environmental facts.
Separates evidence from the team’s interpretation.
Insight
Pattern or mismatch that changes the team’s understanding.
Feeds the POV statement and HMW prompts.
Design consequence
What should change in the problem frame, prototype, metric, or risk register?
Turns research into planning evidence.

3.18 Knowledge Check

3.19 Code Challenge: Research Insight Record

3.20 Common Pitfalls

Questions such as “Would you use a smart sensor for this?” produce weak evidence because users often answer politely or speculate. Ask about recent behavior, current workarounds, and actual constraints instead.

The Thinks and Feels quadrants often contain inference. Mark uncertain interpretations and test them in later sessions.

“Users need an app” or “users need a sensor” skips Define and jumps to a solution. Rewrite the frame as a user need and research insight.

An IoT product may succeed in a demo and fail in the field if setup, calibration, power, cleaning, firmware updates, and support are not part of empathy work.

3.21 References

3.22 See Also

3.23 What’s Next

Continue to Ideate, Prototype, and Test after your team has a research-backed POV statement, a small set of How Might We prompts, and success metrics that connect user outcomes with IoT constraints.

Previous
Current
Next
Human-Centred Design for IoT
Empathize and Define
Ideate, Prototype, and Test

3.24 Summary

The empathize and define stages translate real user context into a focused IoT problem statement. This includes observing workflows, identifying pain points, defining stakeholders, and separating symptoms from root needs.

3.25 Key Takeaway

Do not start IoT design with a sensor list. Start with the user, environment, decision, and failure consequence, then define the problem the system must solve.