Interviews

Goal

The interview elicits business analysis information from an individual by conducting a conversation with them that is directed by a goal and equipped with a prepared set of questions. Part of what an organisation knows is written down nowhere: why the control rule was introduced in 2019, what the field "status 4" actually designates, how the case that arrives after the deadline is handled, what happened last December when volume tripled. That knowledge lives in the heads of a few people, and the only way to get it is to ask them.

What the technique brings, and what separates it from every other form of elicitation, comes down to three properties. The probe first: the interviewer hears the answer and reacts within the second, pushes the general formula down to the concrete case and confronts two incompatible statements while the person who made them is still sitting across the table. No survey does that. Candour next: BABOK makes it an explicit strength of the technique, a person expresses in private the opinions they will not voice in public, especially where confidentiality has been promised. The reservation about the sponsor's plan, the workaround everybody uses, the disagreement with the head of department come out here or nowhere. The non-verbal signal last: the hesitation before "yes, that is the process" is information, and it calls for one more question.

BABOK gives the technique a second purpose and places it at the same rank as the first: the interview establishes the relationship between the analyst and the stakeholder and thereby increases that stakeholder's involvement in and support for the change. The information is the deliverable, the relationship is the asset. It makes the second request easier than the first, it decides the quality of the answers at the next workshop, and it is built in the hour spent one to one.

The technique produces two artifacts. Before, the interview guide: the goal, the interviewee and their role, the mode and the duration, the set of questions in their order, the planned probes, the timing per block and the questions that will be dropped if time runs short. After, a set of findings confirmed by the interviewee: the notes written up, sent back, corrected, marked as confirmed. BABOK states its own limitation with useful frankness, the documentation produced from an interview remains subject to the interpretation of whoever wrote it, and confirmation is the only control that closes that gap. Six factors decide success, and they are the technique's real cost drivers: the interviewer's knowledge of the domain, their experience of interviewing, their skill at documenting, the readiness of the interviewee to provide the information and of the interviewer to conduct the session, the clarity of the goal in the interviewee's mind and the rapport established between the two.

Usage

When to use it

• The information sits in one head: an expert holds the rule, the exception or the history that nothing documents.
• Depth over coverage: probe, pursue and confront two contradictory answers within the hour itself.
• Sensitive or political subject: reservations, workarounds and disagreement with the sponsor are voiced in private.
• Senior stakeholder: the director grants 45 minutes one to one and will attend no workshop.
• Early framing: nobody yet knows what to ask, and the first interviews dictate the questions later put to everyone.
• A relationship to build: a stakeholder the analysis will depend on for months, whose trust is won one to one.
• Validating a deliverable: put a model, a rule set or a matrix in front of the person able to demolish it.

When not to use it

• Numbers and coverage are needed: reach beats depth, run a survey by questionnaire (BABOK 10.45).
• A decision or a work product is needed: no interview makes a group converge, run a workshop (BABOK 10.50).
• The question is about work as actually performed: the declared process diverges from the real one, observe (BABOK 10.31).

Description

The two forms of the interview

BABOK sets out two interview types. The structured interview rests on a predefined set of questions: the same questions, in the same order, to every interviewee. Its answers are comparable across interviews, they can be summarised and counted, and it is the only form that lets anyone say "seven of the nine case managers describe the same exception". The unstructured interview gives up the predetermined format and order: the questions arise from the answers and from the interaction. It goes further into the unexpected and produces nothing comparable.

BABOK itself adds that practice combines the two, the analyst adding, dropping and reordering questions as the interview runs. The working form is therefore a prepared set of questions held loosely, and the real decision is taken question by question: a question is customised when the point of the exercise is this person's own perspective and standardised when the answers will be summarised and analysed across the whole set of interviews. A series of interviews from which a count is to be drawn is run with standardised questions, and that decision is taken before the first interview.

The group interview

BABOK allows more than one interviewee in the same session and describes the move that defines the technique: the interviewer is careful to elicit a response from each participant. The boundary with the focus group lies in the direction of the questions. They stay addressed to individuals, one after another, and the interviewer conducts a dialogue with each person; in a focus group the discussion runs between the participants. The failure mode is mechanical and it is common: one voice, often the most senior, fills the hour; the silent participants are recorded as agreeing with what it said; and the organisation has paid for four people's time to hear one person's. The remedy is to name people. The question is put to the group, then put again to each person by name, and a silence is recorded as a silence, never as assent.

Prepare

The goal

It sits at two levels. The first is the level of the series: which business need the whole set of interviews answers. The second is the level of this interview: what this person, and only this person, can give. BABOK is explicit on a point practice neglects, the goals are clearly stated and communicated to each interviewee. An interview with no goal is a conversation. A goal the interviewee has never heard remains the analyst's goal alone, and the interviewee spends the hour guessing what is wanted from them. BABOK distinguishes four families of goals, which produce different question sets: collecting data, understanding the stakeholder's perspective on the change, developing a proposed solution or establishing rapport and support.

The interviewees

They are identified from the goals, together with the project manager, the sponsor and other stakeholders. Three populations stand apart and are recruited differently: those who decide (senior enough to frame the stakes and to name the others), those who do the work (users and business experts, with whom the detail and the exception live) and those whose absence from the list would be a political incident. That selection is the study's sample, and it is through it that bias enters, silently, before the first question.

On a process subject, the ABPMP's BPM CBOK widens that population. It has the process owner interviewed, together with internal and external stakeholders (suppliers, customers, partners), those who work the process and those who pass inputs to it or receive its outputs.

The questions

They are designed from the goal, and they come in three types.

• An open question calls for a narrative or a sequence of steps and cannot be answered yes or no. BABOK gives the justification worth keeping: it allows the interviewee to supply information whose existence the interviewer did not suspect. That is why the technique beats a survey, where one only ever harvests what one knew to ask.
• A closed question calls for a single value: yes, no, a number, a date. BABOK assigns it a precise use, clarifying or confirming a previous answer. The closed question is the instrument of confirmation, and a guide that contains only closed questions will harvest nothing beyond what was already written in it.
• A probe is built on the answer just given. The most effective form reuses the interviewee's own words: "you said the receptionist checks the session, what does she check and on which screen?". The rule used by the UK public service's user research teams applies as it stands: aim for real stories and examples, avoid generalities. A generality is undone by asking for the most recent concrete case.

To those three types is added the most profitable and the hardest question, the cross-check question: putting two of the interviewee's own answers side by side. "You report 1'400 sessions a month and around 160 invoices, does one invoice cover several sessions?" It demands that the interviewer hold the whole interview in mind while listening, and it is the question that surfaces the processes nobody thought to mention.

The order

BABOK notes that questions are often organised by priority and importance, and it names the possible sequences: general to specific, start to finish of a process, detailed to summary or by the interviewee's level of knowledge and by subject. The practitioner's default is general to specific, and it is what gives the interview its funnel shape: an open question opens the subject, one or more probes narrow it onto the concrete case, a closed question confirms what has just been heard and the next funnel opens on the next subject.

The interview guide

That is BABOK's term and it is the technique's real artifact. It gathers the questions, the proposed timing and the planned probes, and it follows from the interview type, the goal, the mode of communication and the time granted. Two properties BABOK states explicitly make it an instrument of work. The guide is a document in which the interviewee's answers are easily recorded: the right-hand column starts empty and fills up during the session, which saves a transcription. And the guide identifies in advance the questions that may be dropped if time runs short: the day the interviewee announces thirty minutes instead of sixty, the triage is already done, and the analyst does not sacrifice the central question because it came last. To these elements is added the introduction script, which Swiss law makes mandatory (FADP art. 19) and which BABOK leaves implicit: who the analyst is, what the interview covers, how the notes will be handled, who will see them.

The logistics

Four decisions carry consequences. The location and mode (on site, by telephone, by video conference) adapt to the interviewee's diary. Recording, if refused or impossible, imposes a scribe. Sending the questions in advance is a practice BABOK restricts with remarkable sharpness: it is indicated only where the interviewee must gather information in order to prepare their answer. Sent out of courtesy, a complete question set trades a spontaneous answer for a prepared one, and a prepared answer describes the official process. Confidentiality last: if the results are confidential, how they will be summarised without any individual being identifiable is decided before the first interview, because it changes what people will say and because it constrains the write-up.

What Swiss law imposes on an interview

The interview touches two legal regimes, and both have teeth. Recording a non-public conversation in which one takes part oneself, without the consent of the other participants, is a criminal offence: art. 179ter of the Criminal Code punishes it, on complaint, with a custodial sentence of up to one year or a monetary penalty. An elicitation interview is a non-public conversation in which the analyst takes part, so that article is indeed the one that catches it. The Federal Supreme Court has settled the scope of the notion: a conversation is non-public where its participants speak with the legitimate expectation that their statements will not be accessible to everyone (BGE / ATF 146 IV 126, 2020), which brings the professional exchange within the provision. The FDPIC gives the operational rule: the consent of all participants, obtained before the recording starts and informed as to its nature and purpose. Asking for agreement once the video-conference tool is already recording does not meet that requirement.

Interview notes are personal data, and the Federal Act on Data Protection applies to them. Its art. 19 requires that the data subject be informed of the collection, at minimum with the identity and contact details of the controller, the purpose of the processing and, where applicable, the recipients of the data: the introduction script BABOK already describes is, in Switzerland, the discharge of a legal obligation. Its art. 6 sets proportionality and purpose limitation, which has a direct consequence: notes taken to specify a billing process are not available for appraising the person who supplied them. The interviewee retains a right of access (art. 25) and a right to have inaccurate data rectified (art. 32), and the confirmation loop is the cheapest way of satisfying that right in advance. Where the interviewee is an employee, art. 328b of the Code of Obligations tightens the frame further: the data processed must relate to their suitability for their job or be necessary for the performance of the employment contract. Finally, handing the recording to a cloud transcription service is a disclosure to a processor and possibly a disclosure abroad, with the rules that attach to it.

The practical rule fits in one sentence: announce the recording, obtain consent before starting, state the purpose and the recipients, offer the interview without recording as a real option (a second analyst takes the notes) and delete the raw recording once the notes are confirmed.

Conduct

1. The opening
   It is short. The analyst states the purpose of the interview and why this person's time is needed, confirms their role, deals immediately with any concern they raise and explains how the information will be recorded and with whom it will be shared. Half of the technique's purpose is settled here.
2. The body of the interview
   It takes up most of the session. The interviewer holds course on the goal and on the prepared questions while adapting to what is said and to the non-verbal signals. They gauge continuously the interviewee's readiness to participate and to give the information, and they accept that a single session is not always enough. They practise active listening and confirm their understanding in the room, by restating what they have just heard. They take notes, or their colleague takes them. Concerns raised are answered or recorded for follow-up.
3. The close
   It is short as well and carries one question BABOK singles out, because it regularly yields the best finding of the session: what have I not asked you about?. The analyst then recaps the session, indicates how the results will be used, leaves their contact details for the additions the interviewee will remember that same evening, then thanks them.

Confirm

BABOK makes follow-up a full element of the technique: the information is organised, then the results are confirmed with the interviewee as soon as possible, that sharing allowing them to point out anything missed or incorrectly recorded. The step practice skips is the one that decides the value of all the rest. Two reasons make it mandatory. The first is methodological: BABOK itself acknowledges that the documentation remains subject to the interviewer's interpretation, and the interviewee's review is the only control that closes that gap. The second is legal: the interviewee holds a right of rectification in any case, and the confirmation loop serves it before they have to claim it. The notes are written up while they are fresh, they go to the interviewee, their corrections are integrated and the finding changes state. An unconfirmed note remains a hypothesis. Skipping this step is expensive: one correction from an interviewee can split a whole process in two.

Choosing between the interview and its neighbours

Five elicitation techniques look alike from a distance and are chosen on a single criterion: who is in the room and what is wanted out of it. The most frequent confusion concerns the focus group, which some texts call a group interview. The two separate on the role of whoever leads: the moderator of a focus group facilitates a discussion between the participants, the interviewer addresses questions to an individual. A group interview remains an interview as long as the questions stay addressed to people one at a time. And neither of the two makes the group converge: the focus group collects attitudes and perspectives, the workshop is the technique that produces a decision.

Observation deserves one more sentence, because it exists precisely to catch the interview's structural weakness. What people say they do diverges from what they do, and three well-identified mechanisms produce that gap: people bend their answer toward what they believe is expected of them, they report what they remember rather than what happened and they rationalise their own behaviour after the fact. The consequence for the analyst is operational: when the question is "how does the work actually get done", the interview produces a hypothesis and observation verifies it. The two techniques chain, the second correcting the first.

What makes an interview fail

The leading question

BABOK gives it as a limitation of the technique, the risk of unintentionally inducing the interviewee's answer. The best documented experiment on this point is Loftus and Palmer (1974): by changing a single verb in the question put to witnesses of a filmed accident, "how fast were the cars going when they smashed into each other" rather than "when they hit each other", the speed estimates shift; and a week later, the group given the more violent verb reported more often having seen broken glass, which appeared in no frame of the film. A leading question does not merely bias the answer, it can install it. The remedy is mechanical: ask "how does the invoice reach the client?", never "the invoice goes out automatically, is that right?".

The interviewer who talks

Silence after an answer is a tool, and it is the one that produces the second half of the answer. The instruction given to the UK public service's user research teams is to stay deliberately silent so the participant speaks more.

Letting one voice take over the group interview

BABOK describes an interviewer who is careful to elicit a response from each participant. Without a probe by name, the three other people present are recorded as agreeing with the first, and the finding goes into the specification under their names without their having said it.

Taking the declared process for the real process

The gap between saying and doing is structural, and no question wording closes it. The remedy is another technique, observation (BABOK 10.31).

Interviewing only the volunteers

The ones who answer the email, the ones the sponsor named, the ones who like projects. The sceptics are missing, and so is the night shift whose workaround is the finding. No interviewing skill repairs a biased sample, and a set of interviews drawn from volunteers is not presented as the view of the organisation.

Never having the notes confirmed

The most frequent and the most expensive error. The analyst's interpretation goes into the specification under the stakeholder's name, and the correction arrives at acceptance testing, where it costs fifty times more.

Sending the questions in advance out of courtesy

What is harvested is the official, prepared version of the process. BABOK judges sending them ahead advisable in one case only, where the interviewee must gather information in order to answer.

Reading the guide out word for word

The guide is a coverage list for the analyst, and it is read before the interview. Recited aloud, it produces an interrogation, and the interrogation destroys the rapport that is half the technique's purpose.

Taking one interview for the truth

An interview gives a perspective. The finding is triangulated against the other interviews, against the documents and against observation before it is treated as a fact.

Recording without consent

In Switzerland this is a criminal offence (art. 179ter of the Criminal Code), and consent is obtained before the recording starts.

AI considerations

AI acts on the technique's dominant cost driver, documentation. Automatic transcription turns an hour of interview into searchable text in minutes, where writing it up by hand cost one to two hours per hour of interview. The gain is real and it is the largest the technique has seen in a long time. Drafting the question set is the second useful application: a language model given the interview's goal and the findings from document analysis produces a first question set quickly, and it is good at the mechanical part of the work, turning a closed question into an open one, proposing probes, checking that each goal is covered by at least one question. Synthesis across a series of interviews is the third, and it is where the machine beats the human: proposing themes across ten transcripts, marking every passage that mentions a given step of the process and above all flagging the places where two interviewees contradict each other. The list of contradictions is exactly what an analyst produces badly across ten documents, and it is the raw material of the next round of probes. Preparation last: summarising the domain documentation so the analyst arrives with the vocabulary, which answers the first success factor BABOK states.

The limits are hard, and the first commands all the others. An automatic transcript is not evidence, because speech recognition fabricates. Koenecke and co-authors (FAccT 2024) measured that roughly 1% of the transcriptions produced by Whisper contained entire hallucinated sentences, absent from the audio, and that 38% of those hallucinations carried an explicit harm, in three categories: invented violence, inaccurate associations and false authority attributed to a speaker. The third is the one that hits the analyst head on. In an interview, a sentence invented in a transcript becomes an invented requirement, and it carries the name of a real stakeholder, who will then be cited downstream as its source. The confirmation loop with the interviewee is what catches this error, and using automatic transcription is precisely the reason never to skip that step.

Automatic summarisation flattens the exception. "We validate the same day, except at quarter end when volume triples and we process in batches" is summarised as "validated the same day". The exception, though, is very often the requirement, and it is the very reason an interview was run instead of a survey. The analyst rereads the transcript looking for the caveats, the "generally", the "except when". Generated questions drift toward the leading question. A language model has no memory of what the previous interviewee said, no reading of the room and a strong tendency to confirm the premise it was given. Every generated question is pruned by the analyst, and the leading ones are rewritten. The probe remains a real-time judgement: deciding which answer deserves pushing, hearing the hesitation before the "yes, that is the process", is the technique itself, and nothing in the toolchain does it. The relationship cannot be delegated: an agent that attends the meeting in the analyst's place, a summary the analyst has not read, a question they cannot explain, spend the capital of trust the interview should be building.

The Swiss constraint finally bears on the toolchain itself. A meeting assistant that joins the video conference is a recording (consent of all participants, before it starts) and a disclosure of personal data to a processor, possibly abroad. The service is therefore chosen on where the audio is processed and stored and on the quality of the processing contract, before it is chosen on the transcription error rate. The FDPIC states that data protection law applies directly to AI systems, with no exemption.

Examples

A group of three physiotherapy practices, headquartered in Yverdon-les-Bains (postcode 1400, business identification number CHE-107.893.452), is replacing its scheduling and billing software. The analyst conducts a 60-minute interview with the billing manager, on site, with a second analyst taking notes and no recording. The goal communicated to the interviewee: understand how a treatment becomes a paid invoice and where invoices are rejected. Two framing inputs do not belong to the interview and are stated here so that the calculations are reproducible: the working day counts 8 hours, and management costs an administrative post's internal rate at CHF 75 an hour. The guide is the artifact produced before the session.

The notes are written up the next day and sent to the interviewee. The log of findings is what they became after her review. Two columns stay separate by design, what was said and what the analyst concludes from it: the first is verified against the person, the second commits the analyst.

The correction of finding C-02 is what the confirmation loop returned, and it pays for the step on its own. During the interview the interviewee described the case she handles most often, the invoice addressed to the health insurer, and the analyst recorded it as the process. The review surfaced a second, whole flow: under tiers payant, the practice invoices the insurer, which pays the provider; under tiers garant, the patient receives the invoice, settles it, then claims reimbursement from their insurer. These are two distinct processes, with a patient reminder, a payment delay, a bad-debt risk and a collections path the first flow does not have. Without the review, the specification for the new software would have gone out with a single billing flow, and the second would have surfaced at acceptance testing or in production, at a client who would then have paid twice: for the analysis and for redoing it.

Visualisations

The technique produces three objects, and each has the form that follows from its nature. The run of the interview is a spatial object: three phases, a funnel of questions nested inside the middle one and above all a return edge, the interviewee's correction coming back into the notes before those notes become a finding. It is that edge which closes the interview into a loop: the correction returns into the notes before any finding exists, and a three-row table would destroy it. It is drawn. The interview guide and the log of findings are the reverse: rows and columns, with columns that must stay separate and a status column that is read at a glance. They are tables, selectable, searchable and resizable, and printing them as an image would lose everything and gain nothing. The choice between the five neighbouring techniques is likewise a grid.

The cycle has a second use, more useful still than the first: placed alongside a real interview, it works as an audit instrument, and three checks can be read off it. Where the questions came from, that is, whether a written goal exists from which they follow. Where the funnel is, that is, whether the closed questions serve to confirm what has just been said or make up the entire guide. And above all, whether the return edge was travelled, that is, whether the notes went to the interviewee and came back. An interview whose notes never went back has stopped two thirds of the way through the technique.

Cost

This rating holds for one interview. The technique's real profile is that it multiplies: ten interviews cost ten times the execution and ten times the documentation, with no economy of scale appearing anywhere except in preparing the guide. That is exactly the point at which the survey (BABOK 10.45) becomes the better use of the budget and the point at which three interviews are run in order to know what to ask the other three hundred people.

Tooling

The technique's minimum tooling is a notebook and a pen: the interview is conducted in a warehouse, on a building site or in a physiotherapy practice that owns no software at all, and it depends on no platform. What it requires is a guide template, document or spreadsheet, with the columns that make the guide operative: block, question, type, timing, planned probe, droppable question and the answer column left empty. BABOK explicitly wants the guide to double as the recording form, where the technique's tooling really lives.

Video conferencing (Teams, Zoom, Webex, Google Meet) carries the interview remotely better than it carries most other elicitation techniques, because a two-person dialogue does not suffer from a group's turn-taking. It costs non-verbal signal, and it makes the question of consent to recording immediate, since the button is right there. Recording and transcription are chosen on where the audio is processed, where it is stored, under what processing contract and on what deletion schedule. A service hosted in Switzerland or in the European Union, contracted as a processor, settles those four questions in one decision.

Two note-taking practices are worth tooling. A document shared on screen during a remote interview lets the interviewee correct the notes as they are written, which moves part of the confirmation loop inside the session and cuts the write-up accordingly. A mind map whose branches are the guide's blocks suits analysts who cannot listen and write linearly at the same time.

For analysing a series of interviews, a spreadsheet with one row per finding and a column of tags is enough up to about ten interviews. Beyond that and especially where several analysts code the same corpus, qualitative analysis tools (Dovetail, MAXQDA, NVivo, Atlas.ti) earn their licence: they hold the link between the excerpt, its tag and the interview it came from, which the spreadsheet loses as soon as it grows. Downstream, each confirmed finding keeps a link back to the interview that produced it, in the requirements or issue management tool (Jira and Confluence, Polarion, DOORS). Without that link, the log of findings becomes a document nobody reads again, and the traceability of a requirement back to its source disappears on the day it would be useful.

Sources

• IIBA, A Guide to the Business Analysis Body of Knowledge (BABOK Guide) v3, §10.25 Interviews.
• IIBA, A Guide to the Business Analysis Body of Knowledge (BABOK Guide) v3, §10.21 Focus Groups, §10.31 Observation, §10.45 Survey or Questionnaire, §10.50 Workshops: the definitions on which the boundary between the five techniques rests.
• ABPMP, Guide to the Business Process Management Common Body of Knowledge (BPM CBOK), §4.6 Gathering Information (Interviewing): the population to interview on a process subject, from the process owner to internal and external stakeholders and to those who pass inputs to the process or receive its outputs.
• GOV.UK Service Manual, Using in-depth interviews, Government Digital Service: the discussion guide, the introduction script, probing for real stories and the instruction to stay silent.
• Elizabeth F. Loftus and John C. Palmer, Reconstruction of Automobile Destruction: An Example of the Interaction Between Language and Memory, Journal of Verbal Learning and Verbal Behavior, 13(5), 1974: the experimental proof that a leading question shifts the answer and can install a memory that was never there.
• Jakob Nielsen, First Rule of Usability? Don't Listen to Users, Nielsen Norman Group, 2001: the gap between what people say they do and what they do and why observation verifies what the interview advances.
• Allison Koenecke et al., Careless Whisper: Speech-to-Text Hallucination Harms, ACM FAccT 2024: the measurement of entire sentences hallucinated by automatic speech recognition and the three categories of harm they carry.
• Swiss Criminal Code (CP / StGB), SR 311.0, art. 179ter: recording, without the consent of the other participants, a non-public conversation in which one takes part.
• BGE / ATF 146 IV 126 (2020), Swiss Federal Supreme Court: a conversation is non-public where its participants speak with the legitimate expectation that their statements will not be accessible to everyone, which brings a professional exchange within art. 179ter of the Criminal Code.
• Swiss Federal Act on Data Protection (FADP / LPD / DSG), SR 235.1, art. 6, 19, 25, 32: the processing principles, the duty to inform on collection, the right of access and the right to rectification.
• FDPIC, Recording conversations: the consent of all participants, informed and obtained before the recording starts.
• FDPIC, Artificial intelligence and data protection: the direct application of data protection law to AI systems.
• Swiss Code of Obligations (CO / OR), SR 220, art. 328b: the limitation of processing an employee's data to their suitability for the job or to the performance of the employment contract.
• Swiss Federal Act on Health Insurance (LAMal / KVG), SR 832.10, art. 42a: the insurance card and the AVS number it carries as the identifier of the insured person.
• Swiss Federal Statistical Office, Business identification number (IDE / UID): the CHE-NNN.NNN.NNN structure and the calculation of its check digit.