Contextual Inquiry or Walkthrough¶
Field-research method — instantiates Affordance Shaping
Learns what agents are really trying to do — and where they go wrong — by observing them in their own context of use rather than reasoning about them from a desk.
Contextual Inquiry or Walkthrough goes to where the action happens and watches it — interviewing and observing real agents in their own environment (contextual inquiry), or stepping through a task exactly as the agent would, decision by decision (a walkthrough). Its distinguishing commitment is situated understanding: it treats the agent's goals, skills, workarounds, and error paths as things you can only learn in context, not deduce from a spec. Where an audit reads the environment and a test scores an outcome, this mechanism builds the picture of the agent — what they are trying to accomplish, what they can actually do, and the exact point at which the intended path and their real path diverge.
Example¶
A startup ships a crop-scouting app that lets agronomists log pest sightings; adoption stalls. Instead of guessing, a researcher spends two days walking fields with the agronomists. In context, she sees what no spec captured: the agronomist's hands are dirty and full, the phone stays in a chest pocket, gloves defeat the touchscreen, and the real goal isn't "log a sighting" — it's "get back to the truck with the count already tallied." The intended path (stop, peel off gloves, tap through five screens) collides with every one of those constraints. She walks the task herself, gloves on, to feel where it breaks.
The deliverable is a model of the agent and the goal: "the user is a moving, glove-handed, time-pressed expert whose actual goal is a running tally, and the harmful path is that in-field logging is unusable, so sightings get batched into unreliable end-of-day memory." That reframing — voice input, a one-tap tally — is something the design team could never have reached from the office.
How it works¶
- Go to the context; don't summon it. Observe in the real setting, where the environmental constraints that shape behaviour are actually present, rather than in an interview room or from a requirements doc.
- Follow the agent's goal, not the designer's task. Reconstruct what the person is really trying to achieve — often different from the feature's stated purpose — and the point where harm or failure enters.
- Walk it yourself. Step through the task in the agent's role and conditions to feel the friction and the capability limits from the inside.
- Master–apprentice stance. Let the agent lead and teach; the researcher's job is to surface the tacit, not to confirm a hypothesis.
Tuning parameters¶
- Naturalistic vs. structured — pure observation (catches the real, unprompted behaviour) versus a guided walkthrough of set steps (faster, but sees only what you thought to script).
- Breadth of agents — how many and how varied the people observed; more captures the range of goals and skills, at roughly linear cost.
- Interruption level — how often the observer probes with "why did you do that?"; probing surfaces tacit reasoning but perturbs the behaviour being observed.
- Real vs. simulated context — the true field setting (highest fidelity) or a reconstructed one (cheaper and safer, but softer on the constraints that matter).
When it helps, and when it misleads¶
Its strength is that it is the only sibling that recovers the agent's actual goal and real error path — the two things every downstream shaping decision quietly assumes it already knows. It routinely overturns the designer's model of the user, which is exactly where nonuse and misuse are born.
Its central failure mode is that observation changes what is observed: people work more carefully, follow the rules more faithfully, and narrate more coherently when watched, so a study can capture the compliant version of the behaviour rather than the real one — the Hawthorne effect.[n1] Small, unrepresentative samples over-generalise; and staged as theatre, a "walkthrough" can be run to confirm the design the team already loves. The discipline is to observe enough varied agents, unobtrusively, and long enough for the watched-behaviour gloss to wear off — and to look hardest for the workaround that contradicts the plan.
How it implements the components¶
Contextual Inquiry or Walkthrough fills the agent-understanding side of the archetype — the components that model the person the environment must fit:
agent_capability_model— it builds a grounded, observed model of what the real agent can perceive, reach, and do, including tacit skill and situational limits.goal_and_harm_path_definition— it surfaces the agent's true goal and the specific path by which they reach a wrong or harmful outcome, defining the target the shaping must redirect.
It does not inventory the environment's affordances (action_possibility_inventory, environment_substrate_map) — that is the Affordance Audit — and it produces no controlled outcome measure; that belongs to Prototype A/B or Multivariate Test and Usability or Field Test. Its formal, decompositional cousin is Task and Capability Analysis, which derives the capability model from task structure rather than from observation.
Related¶
- Instantiates: Affordance Shaping — supplies the model of the agent and goal the shaping is fitted to.
- Sibling mechanisms: Affordance Audit · Prototype A/B or Multivariate Test · Desire Path Observation · Friction Adjustment · Physical or Digital Keying · Task and Capability Analysis · Usability or Field Test · Signifier Prototyping · Robot Action-Space Mapping · Safe Default or Preselected Path · Wayfinding Marker
Editorial Notes¶
Form Classification¶
Form family: Communication, Facilitation & Learning
Rationale: Learns what agents are really trying to do — and where they go wrong — by observing them in their own context of use rather than reasoning about them from a desk, making its operative form a designed message, facilitated interaction, ritual, or learning activity that changes shared understanding.
Independent corroboration: The frozen evidence defines Contextual Inquiry or Walkthrough as 'Learns what agents are really trying to do — and where they go wrong — by observing them in their own context of use rather than reasoning about them from a desk', so its operative form is Communication, Facilitation & Learning.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Human-Computer Interaction
Origin pattern: Single lineage
Present-day reach: Specialized
Rationale: User-centered design cohered situated inquiry and task walkthroughs to recover real goals, workarounds, skills, and error paths from use rather than specification.
Related originating lineages:
- Ethnography & Qualitative Methods — Field observation supplies the contextual-inquiry lineage for understanding work in its natural setting.
Review resolution: Both named variants belong to the HCI user-research lineage and share ethnographic observation; the method's origin is specialized despite later operational uses.
Attribution caveat: The combined label joins contextual inquiry and cognitive or task walkthrough traditions within user research.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
This mechanism is upstream of almost everything else in the set: it produces the goal a Prototype A/B or Multivariate Test later scores against and the error path a Friction Adjustment later redirects. Skip it and every downstream mechanism optimises a goal nobody verified.
[n1] The Hawthorne effect — the tendency of people to alter their behaviour when they know they are being observed. It is the central threat to validity in any in-context observation method, and the reason unobtrusive, prolonged observation is preferred over a single watched session. ↩