Observation

Purpose

Observation elicits information by watching an activity being performed in its real context, with the tools and resources it uses. What it targets is the work as actually done, as opposed to the work as reported, which an interviewee describes the way they believe it runs or the way it ought to run. The gap between the two is not the interviewee's carelessness, it is structural: part of the work is tacit, those who hold it perform it without being able to put it into words, and another part consists of workarounds that nobody treats as steps and that no one therefore thinks to mention.

BABOK assigns the technique five uses: identifying needs and improvement opportunities, understanding a business process, setting performance criteria, evaluating the performance of a solution in use and supporting training. Two of these pull the technique in opposite directions and fix the choice of stance. Understanding a process and spotting improvement opportunities call for reaching the reasoning, the why behind the action. Setting performance criteria and evaluating a solution call instead for measurement, time, volumes and quality, without disturbing what is being measured. The first objective wants a stance that asks questions, the second a stance that stays silent.

The strength of the technique lies in the nature of its evidence. It shows productivity firsthand and compares it against a standard, it surfaces the informal tasks and the workarounds, and it rests its recommendations on observed facts, often quantified, rather than on a second-hand report. Its deliverable is an observation log, the structured record of what was seen, which captures for each step the time, the tool, the quality, the anomalies and the departures from the procedure. BABOK also states its limits: it can be intrusive, it demands significant time to prepare and to run, and it does not suit knowledge-based work, the kind that happens in the head and cannot be watched.

Usage

When to use it

• The declared process departs from the real one: what is said in an interview does not match what is done, and the gap is the very object of the study.
• Establishing a quantified baseline: measure the times, the volumes and the rework rate of an activity before a change, to assess its effect afterwards.
• Reaching the reasoning behind an action: understand why an exception is handled a certain way, by asking the operator to explain while they do it.
• Evaluating a solution in real use: watch the new tool used in the field, where the gap between the demo and the practice shows.
• Manual or physical activity: a counter, a warehouse, a reception desk, a care practice, where the work is visible and can be followed step by step.
• Capturing tacit knowledge: the move repeated without thinking, that its author cannot describe and that no manual contains.

When not to use it

• The work is cognitive: you cannot watch someone decide or design, pair observation with the interview to reach the invisible reasoning.
• You need coverage and numbers across a population: observation is costly and reaches only the few people followed, prefer a survey or questionnaire when breadth matters more than depth.
• Trust is missing or the exercise cannot be separated from staff appraisal: the observer effect dominates the signal and the work slides toward prohibited surveillance, fall back on document analysis and application logs.

Description

The decision that governs everything: reasoning or measurement

BABOK sets out two observation stances, and the choice between them is made before the session, from the objective. The two stances collect different things, and one destroys what the other preserves. Choosing the active stance when the objective was measurement yields times distorted by interruptions; choosing the passive stance when the objective was reasoning leaves the analyst with a mute chronology half of which they do not understand. The rule therefore comes down to one question asked before anything else: are we trying to understand why the work is done this way or to measure how much it costs and how well it succeeds.

Active observation and its facilitated variant

In active observation, also called noticeable, the observer asks questions as they arise, even at the cost of interrupting the activity. It reaches the reasoning quickly, because it asks the person to explain what they are doing while they do it, at the moment the action is still in front of both of them. This is the stance for an objective of understanding: reconstructing the process as it is really carried out, with its exceptions and the decisions that govern them. Its most directed variant is facilitated observation, where the observer prompts the participant to perform specific tasks, to shorten a session that would otherwise drag or to target a particular piece of information that the spontaneous activity would not surface.

Passive observation and measurement

In passive observation, also called unnoticeable, the observer does not interfere, holds their questions until the end and lets the activity run its course. This stance preserves the natural flow of the work, and it alone allows the measurement of time and quality, which a question would distort. This is the stance for an objective of performance: timing the steps, counting the reworks, recording the anomalies, producing the baseline against which the solution will later be judged. Its variant records the activity on video, then reviews it with the participant to resolve ambiguities, which defers the questioning to a moment where it no longer disturbs the measurement. Many real observations chain the two over time, passive first to see and measure without deforming, active afterwards to ask the questions the silent phase raised.

Contextual inquiry

BABOK names a variation it does not develop, contextual inquiry: the observer examines the person's work environment, their tools and their information resources, to understand the activity from the inside. The model is that of master and apprentice, described by Beyer and Holtzblatt: the person works and comments, the analyst observes and questions in the same movement, at the place of the activity. This variant adds attention to the artefacts that the work leaves around it, the annotated sheet stuck to the screen, the reminder note, the parallel binder, each of them the trace of a need the official system does not cover.

Running an observation

Four elements structure the conduct. The objectives come first and determine the stance, the people to follow and what will be recorded; a vague objective produces a session where you watch everything and retain nothing. Preparation turns the objective into a plan, who is observed, on which activity, at what moment, taking account of the person's level of experience, the frequency of the activity and the existing documentation, and it makes sure the stakeholders agree on the purpose. The conduct opens with an explanation that carries half the success: the observer says why they are there, assures that individual performance is not being judged and that results are aggregated, reminds the person that they can stop at any moment and invites them to think aloud; during the activity, the observer notes the ordinary and the rare steps, the use of tools, the time, the quality, the anomalies and their own questions. The confirmation finally: the notes are reread and completed with the person, and this sharing eases the concerns the session raised, then the observations are cross-checked against one another, aggregated and analysed against the objectives before the needs are communicated. An unconfirmed observation note remains the observer's interpretation, which often starts from an action whose reason they did not grasp.

The bias that threatens measurement: the observer effect

The central fact of the technique is that people change how they work when they know they are watched. This phenomenon is called the observer effect, or Hawthorne effect, after the studies run at Western Electric's Hawthorne plant between 1924 and 1932. Their quantitative interpretation has since been contested, so it is better to treat the effect as a well-established risk to guard against. The risk hits the measuring stance hardest: a person who knows they are being timed tidies up their moves, follows the official procedure they usually bypass and produces a time that is not their ordinary time. Four practices hold it down. Reassure, by stating that the person's performance is not the object and that results are aggregated. Observe long enough for the performance to slip, which a twenty-minute session never allows. Measure in the passive stance, since any question reminds the person that they are being followed. Triangulate, finally, by cross-checking the observation against interviews, documents and system logs before treating a finding as a fact.

Staying on the right side of surveillance

Watching employees work touches, in Switzerland, a line the law draws firmly. Art. 328b of the Code of Obligations holds that an employer may process data concerning an employee only insofar as it bears on their suitability for the job or is necessary to perform the employment contract: notes taken to specify a process are therefore not available to appraise the person who performs it. Art. 26 of Ordinance 3 to the Employment Act prohibits surveillance systems intended to monitor workers' behaviour at their station; where a system is needed for other reasons, it must impair neither health nor freedom of movement. An elicitation observation is consented, limited to its purpose and time-boxed, and this boundary must stay sharp: an observation never converts into performance monitoring. The Federal Act on Data Protection, in force in its revised version since 1 September 2023, adds proportionality, purpose limitation, transparency and the duty to inform the person of the collection, its aim and the recipients of the notes. The Federal Data Protection and Information Commissioner reminds employers of these requirements and advises consulting staff representation where it exists. In practice: agreement from the person and their manager, purpose stated in writing, retention of only the notes useful to the studied activity, the exercise kept clear of any HR process.

AI considerations

Observation resists automation more than the other elicitation techniques, because its core is a human judgement made in the field, in real time. AI helps downstream. Transcribing and annotating a recording reduces the sifting: locating the passages where a step appears, time-stamping the transitions, cutting the activity into tasks a first time. Assisted measurement extends the passive stance: extracting from a consented video the step durations and rework rates the observer would record by hand, which sharpens the quantified baseline without adding an interruption. Synthesis across several observations is the use where the machine outperforms the human: proposing groupings of actions, flagging the points where two operators proceed differently, signalling the recurring departures from the procedure. This list of departures is the raw material of the analysis, and a human produces it badly across ten logs.

The limits are hard. A model that segments a video does not see the workaround: it recognises the on-screen entry and misses the parallel binder set beside the keyboard, because it classifies what it has learned to recognise and the workaround is by nature what no one documented. And that workaround is very often the finding that justified the observation in the first place. A duration extracted automatically also remains a measure of a behaviour perhaps altered by the observer effect, and the tool does not correct this bias, it quantifies it precisely. Field judgement, deciding that a trivial move deserves a question, sensing that an exception has just gone by, remains the technique itself. The Swiss constraint finally weighs on the toolchain before quality: a video of the workstation is a processing of sensitive personal data, its transmission to a cloud transcription service is a disclosure to a processor, possibly abroad, and the tool is chosen on where the video is processed and stored before it is chosen on its accuracy. Filming without consent or keeping the raw video beyond the confirmation of the notes turns a legitimate elicitation into surveillance.

Examples

A group of three physiotherapy practices based in Yverdon-les-Bains (postcode 1400) is replacing its scheduling and billing software and wants to quantify the expected gain. The analyst runs a passive observation of invoice preparation, without intervening, to measure the time per case and the frequency of rework before the change. The stated objective is to establish a baseline, not to judge the person, and results are aggregated. The measurement log covers six consecutive cases.

What the passive stance delivers here is an undistorted time: no question asked during the measurement interrupted the person, and the durations recorded are those of their ordinary work. The log then brings out the cost driver: both reworks share the same cause, an incomplete prescription, and the two cases concerned are the longest in the run, 6:20 and 7:50 against 2:40 to 5:10 for those that go through first time. The median of 4 minutes 57 becomes the baseline against which the new software will be judged, and the cost of rework on an incomplete prescription becomes the priority requirement, that of a "prescription pending" status which avoids restarting a case from scratch.

Visualizations

The technique produces two objects of different kinds. The choice of stance is a decision read in space: an objective of reasoning leads to active or facilitated observation, an objective of measurement leads to passive observation, and contextual inquiry overlays both by adding the examination of the environment. A branching figure renders that decision at a glance, where a paragraph spells it out. The observation log is the opposite, rows and columns, with a duration column that totals and a rework column that reads down the diagonal. It is a table, selectable and searchable, and rendering it as an image would lose that for nothing.

Cost

Tooling

The minimum tooling for observation is a notebook and an observation grid: a sheet with columns prepared in advance, step, duration, tool, departure, question, which turns note-taking into data entry and avoids forgetting a dimension while the activity rolls by. For the passive stance, a stopwatch or the time-stamping of a note-taking app is the measurement instrument, and a durations-and-rework recording sheet prepared in advance beats an improvised timing.

The video or screen recording lifts the constraint of seeing everything live and lets you review a fast move in slow motion or measure durations after the fact. It is decided, though, on consent, on where the images are processed and stored and on the deletion deadline, before any other consideration. Screen capture suits software tasks, video suits physical activities, and both require the prior agreement of the person filmed. For contextual inquiry, a camera documents the work environment, with the same consent.

For the analysis of a series of observations, a spreadsheet with one row per finding and a labels column is enough up to a dozen sessions. Beyond that, and especially when several analysts sift the same material, qualitative analysis tools (Dovetail, MAXQDA, NVivo, Atlas.ti) hold the link between the excerpt, its label and the session of origin. Downstream, every identified need keeps a link back to the observation that produced it, in the requirements management tool.

Sources

• IIBA, A Guide to the Business Analysis Body of Knowledge (BABOK Guide) v3, §10.31 Observation.
• Nielsen Norman Group, Contextual Inquiry: Inspire Design by Observing and Interviewing Users in Their Context: the observe-in-context method, from the master-apprentice model to the examination of tools and information resources.
• Nielsen Norman Group, The Hawthorne Effect (Observer Bias) in User Research: the risk that the observed person alters their work and the measures that reduce it.
• Hugh Beyer and Karen Holtzblatt, Contextual Design: Defining Customer-Centered Systems, Morgan Kaufmann, 1998: the originating source of contextual inquiry, presented as one author team's method.
• FDPIC, Data processing by the employer: the Swiss data-protection frame applied to observing employees.
• FDPIC, Monitoring systems in the workplace: the surveillance boundary the technique must stay clear of.
• Code of Obligations (CO), SR 220, art. 328b: the limitation of processing an employee's data to their suitability for the job or the performance of the contract.
• Ordinance 3 to the Employment Act (OLT 3), SR 822.113, art. 26: the prohibition of surveillance systems intended to monitor workers' behaviour at the workplace.
• Federal Act on Data Protection (FADP), SR 235.1: proportionality, purpose limitation, transparency and the duty to inform at collection.