Elicitation
Elicitation is the activity through which the business analyst obtains the information the analysis needs: what stakeholders know, what the organisation's documents and data contain and what nobody knows yet. BABOK describes three types of it. Collaborative elicitation rests on direct interaction with stakeholders and on their experience, expertise and judgment. Research systematically discovers and studies the information held in materials or sources that those stakeholders do not know directly, the analysis of historical data included. Experiments produce, through a controlled test, information that can be drawn neither from people nor from documents, because nobody knows it yet. A real initiative mixes several of them, and the mix chosen is written down in an elicitation activity plan that sets the scope, the techniques, the logistics and the supporting materials.
Goal
Elicitation serves to obtain the information the analysis will use, and its first decision bears on where that information comes from: which origin will yield it and at what price. Three origins are available. They differ in cost, in lead time and in what they cannot obtain, and the choice is made again at every open question.
The deliverable is the elicitation activity plan: for each activity, its scope, the techniques selected, the logistics and the supporting materials. It carries three decisions that come expensive when they stay implicit: which questions are settled without engaging anybody, which ones require bringing people together and at what price, which ones have no answer until a test has been run. The plan then governs how elicitation is conducted.
Usage
When to use it
- Heterogeneous sources: people, documents and systems each hold part of the answer.
- Contradictory findings: a second type settles by cross-checking what a single one leaves open.
- Information that exists nowhere yet: a usage preference, the actual duration of a processing step, an abandonment rate.
- Stakeholders scattered or barely available: research advances without them and reduces what is left to ask them.
- Steep hierarchy, low psychological safety: research and observation obtain what a group session will not.
- Regulated domain: the applicable text is read before it is discussed in session.
- An elicitation budget to defend: the plan prices the logistics, the supporting materials and the stakeholders' time.
When not to use it
- Information already elicited, confirmed and still valid: running the activity again produces the same answer at full price, reuse the result and its traceability.
- Disagreement over a priority: no additional information will close the gap, move to Prioritization or to Negotiation.
Description
The three types
Collaborative elicitation draws its information from direct interaction with stakeholders, and BABOK also calls it facilitated elicitation. What it buys is judgment: a person explains why a rule exists, arbitrates between two needs, reacts to a proposal. What it costs is other people's time, which is booked weeks ahead. It is the type most sensitive to the culture of the organisation.
Research systematically discovers and studies the information held in materials or sources that the stakeholders involved in the change do not know directly. Those stakeholders may take part in it, by opening an access or by steering the reading. The analysis of historical data belongs here, to bring out trends or past results. Its most useful property for planning lies elsewhere: it is run alone. BABOK says so about preparation, eliciting through research or exploration may be a solo activity for the business analyst and require no preparation from anyone else.
Experiments identify, through a controlled test, information that neither people nor documents carry, because nobody knows it: a user's real preference between two screen layouts, the actual duration of a processing step, the abandonment rate of a form. BABOK places observational studies, proofs of concept and prototypes in this type. Its planning constraint bears on materials: where the test assumes unusual tools, equipment or techniques, additional planning may be required, which means an order to place and a lead time to hold.
| Type | What it obtains | What it demands in preparation | Where it stops |
|---|---|---|---|
| Collaborative | Judgment, arbitration and the reason a rule exists. | Availability of the people, a room or a video conference link, a facilitator and a scribe, participants told the purpose and what they must have read. | What people do not know and what they will not say in front of a third party. |
| Research | The documented state, the historical data and the trends they carry. | Access to the sources, the rights and the legal basis for processing, the scope of reading. No preparation by third parties. | What is written nowhere and the gap between the document and the practice. |
| Experiments | A behaviour observed or measured under conditions fixed in advance. | Supporting materials, tools or equipment to be ordered with their lead time, a hypothesis and a decision criterion written before the test. | What the test does not reproduce: the real scale, the real stakes and the real duration. |
Choosing the type
BABOK lists the parameters that govern the choice of an elicitation technique: the cost and time constraints, the type of information source and the access available, the culture of the organisation, the outcome expected from the activity, then the needs of the stakeholders, their availability and their location, whether they are gathered in one place or dispersed. In practice, three of them decide almost every case.
The first is access, a piece of work in its own right: rights on a document repository, an extract from a system, a legal basis when personal data is involved, the diary of an executive that fills up six weeks ahead. Many plans choose the workshop by default because access to the documents was never requested.
The second is the nature of the question. A question whose answer exists somewhere belongs to research. A question whose answer exists in somebody's head belongs to the collaborative type. A question whose answer does not exist yet belongs to experiments. This sorting, question by question, is the work of planning. Sorting it wrong makes an analyst infer from a document a behaviour that only a test would have established.
The third is culture. Where contradicting a superior in public has a cost, a collaborative session collects the opinion of the highest-ranking person and the silence of the others. Research, observation and the anonymous questionnaire then obtain what the room will not give, and the individual interview remains the only workable collaborative format.
Mixing the three
An elicitation activity uses one or more techniques to produce the outcome expected within its scope, and in most cases it uses several. Research comes first, because it occupies nobody and shortens everything that follows: one enters the session with questions already formulated and attached to their source. The collaborative type comes next, on what the written record leaves open, namely the reasons, the trade-offs and the cases that appear nowhere. Experiments come last or in parallel, on what neither people nor documents can deliver.
This order reverses just as well. A paper prototype shown in a workshop is an experiment nested inside a collaborative session, and a broad survey frames the interviews that will follow. What matters is that the mix be decided and written down, line by line.
Need, open question
Collaborative
direct interaction with stakeholders
requires: their time, booked and prepared
Research
systematic study of existing sources
requires: access to the sources. No preparation by others
Experiments
controlled test
requires: materials and their lead time
one or more, within a single initiative
Elicitation activity plan
scope · techniques · logistics · supporting materials
Results
unconfirmed
Results
confirmed
The mix serves cross-checking as well. Two types bearing on the same subject control each other: what a manager describes in an interview and what the regulation prescribes diverge regularly, and the divergence is itself a result, to be recorded and carried to the body that will settle it.
The elicitation activity plan
The logistics name who, when, where, for how long, who facilitates and who records. The scope is the open question the activity has to answer, written before the session. The techniques selected are named together with the type they belong to, which makes visible a plan that rests entirely on a single type. The supporting materials cover the documents to be read in advance, the data sets, the mock-ups and the test equipment. Where the test assumes unusual tools, this item carries a lead time that belongs to the project schedule. A single question put to a single source does without all this: asking it and recording the answer with its source costs less than setting up a planned activity.
Preparing the stakeholders closes the plan. Where a research activity requires preparing nobody, collaborative elicitation requires telling the participants the purpose of the session, what is expected of them and what they must have read. A session whose participants discover the subject as they walk in produces a first iteration of their thinking, which will have to be gone over again.
Confirming the results
Results leave the activity unconfirmed, and confirming them is a separate task with its own output. The elicited information is compared against its source and against the other results, to find and resolve problems before resources are committed on that basis. BABOK adds a detail: collaboration with stakeholders may be necessary to ensure that their inputs are correctly captured and that they agree with the results of non-facilitated elicitation. A result obtained by research or by experiment therefore calls for an explicit return to the people concerned.
What makes an elicitation fail
A single type for everything
The workshop becomes the universal instrument, and stakeholder time is consumed on questions whose answer was written down. The mirror case exists and is less visible: an analysis that never leaves the documents delivers the documented state of an organisation where the practised state diverges from it, and the gap between the two is regularly the most useful finding of the study.
An activity with no written open question
The session then produces a conversation and the research a bibliography. The written scope is what authorises closing a session and not reading a document.
The result taken for a requirement
What a stakeholder says is an elicitation result. It becomes a requirement after analysis and confirmation. Skipping that step lets solutions disguised as needs into the backlog, which a quarter later nobody can attribute to a requester or to a reason.
The test with no hypothesis and no decision criterion
A prototype shown "just to see" collects opinions, which is collaborative elicitation in an expensive disguise. A prototype shown to decide between two layouts, with the success criterion written before the test, produces information that nothing else would have produced.
Confirmation deferred to the end
The longer the delay, the less the stakeholder remembers the context of their answer and the more the correction costs. A confirmation that follows the activity by a few days is settled in one session; the same confirmation six months later reopens the elicitation.
The techniques of each type
The list BABOK gives of the techniques usable to conduct elicitation overflows the knowledge area: it names modelling techniques that belong elsewhere and serve to obtain the information before specifying it. It is taken up whole with the four elicitation techniques of the Agile Extension and distributed across the three types according to the mechanism that produces the information. The guide does not make this classification itself, except where it names a technique in the definition of a type.
Collaborative
| Technique | Reference | What places it in this type |
|---|---|---|
| Interviews | BABOK 10.25 | Direct one-to-one interaction, the textbook case of the type. |
| Workshops | BABOK 10.50 | Group elicitation, conducted in a collaborative and facilitated way. |
| Brainstorming | BABOK 10.5 | Production of ideas and search for consensus by the group itself. |
| Focus Groups | BABOK 10.21 | Brings out the ideas and attitudes of a group assembled for that purpose. |
| Mind Mapping | BABOK 10.29 | The same collaborative use as brainstorming, to structure what comes out of it. |
| Collaborative Games | BABOK 10.10 | A participative family by construction: the participants produce the information by handling a shared artifact. |
| Fishbowl | BABOK 10.10c | One circle discusses, one circle listens and records: the information comes from the exchange observed. |
| Process Modelling | BABOK 10.35 | The process is elicited with the stakeholders, by drawing it in front of them. |
| Concept Modelling | BABOK 10.11 | The vocabulary of the domain and the links between its terms are settled with the people who use it. |
| Data Modelling | BABOK 10.15 | Drawing the entities, the attributes and the cardinalities in front of people brings out their special cases and their exceptions. |
| Business Rules Analysis | BABOK 10.9 | A rule applied without ever having been written down can be obtained only from the person who applies it. |
| Story Elaboration | Agile Extension 7.19.2 | The user story workshop: collaborative discovery of user needs and requirements. |
| Behaviour Driven Development | Agile Extension 7.2.2 | The Three Amigos session elicits shared understanding between business, development and testing. |
| Story Mapping | Agile Extension 7.20 | Elicits workflow, user activities and product scope. |
| Backlog Refinement | Agile Extension 7.1 | Ongoing elicitation and clarification of requirements, session after session. |
The four Agile Extension techniques all fall into the collaborative type. Each is a session that brings people together around a shared artifact and rests on their judgment.
Research
| Technique | Reference | What places it in this type |
|---|---|---|
| Document Analysis | BABOK 10.18 | Studies the existing systems, contracts, procedures, policies, standards and regulations. |
| Interface Analysis | BABOK 10.24 | Studies an already documented interaction between two systems or two entities. |
| Benchmarking and Market Analysis | BABOK 10.4 | Compares the organisation against an external baseline, a peer or a published standard that the stakeholders of the change do not know directly. |
| Data Mining | BABOK 10.14 | The guide places the analysis of historical data in this type: the regularities come out of the data set itself. |
| Process Analysis | BABOK 10.34 | Examination of a process in place, of its measured performance and of its variances, conducted on existing data and descriptions. |
| Survey or Questionnaire | BABOK 10.45 | A structured instrument distributed at scale, administered without facilitation and analysed by the analyst. |
Attaching the survey to this type is our reading. It does not meet the criterion of research, since those who answer are the stakeholders themselves. What places it in this type has to do with planning: it is administered without facilitation, a distinction the guide makes itself when it speaks of the results of non-facilitated elicitation, and it occupies nobody's diary.
Experiments
| Technique | Reference | What places it in this type |
|---|---|---|
| Observation | BABOK 10.31 | BABOK names observational studies in this type: the information comes from actual behaviour. |
| Prototyping | BABOK 10.36 | BABOK names prototypes in this type: the mock-up provokes a reaction that the question alone does not obtain. |
The Elicitation and Collaboration knowledge area gathers others, because it also covers preparing for elicitation, confirming the results and collaborating with stakeholders, which are tasks distinct from conducting it. Conversely, six of the techniques in this list belong first to another knowledge area. The four Agile Extension techniques, for their part, belong to another framework.
AI considerations
Artificial intelligence shifts the relative cost of the three types, which redistributes the questions between them.
Research is the type language models change most. Sorting a large corpus, extracting the passages that bear on the scoped subjects and flagging contradictions between two documents used to cost days and now costs hours. Questions that would have been taken to a workshop for want of reading time can then be handled by research, and stakeholder time is kept for what they alone hold. In exchange, a model reads every document as if it were telling the truth, and a citation it produces is checked back against the source.
The collaborative type gains in preparation and in working up the output. A model drafts an interview guide from the file already read, transcribes a session, produces a structured record of it and groups the free-text answers of a survey by theme. What it does not do often decides the outcome: perceiving a participant's hesitation, noticing that the person who holds the answer has gone quiet, deciding not to ask a question in front of the line manager of the person who should answer it.
Experiments gain above all on materials. A clickable mock-up is produced in an hour, which lowers the threshold beyond which a test is worth setting up and moves questions from the collaborative type towards experiments. The trap is symmetrical: a model simulates users on demand and the result looks exactly like elicitation. A synthetic panel returns an average of what has been written elsewhere, whereas experiments exist to obtain what nobody has written yet.
Two Swiss constraints frame all of this. The documents worth researching and the recordings of a session contain personal data: the FADP applies as it stands, the act being technology-neutral, as the FDPIC has pointed out for processing carried out by means of artificial intelligence. Sending a file extract into a publicly accessible model discloses personal data to a third party. Recording a session falls under another text: art. 179ter of the Criminal Code punishes the recording of a non-public conversation by one of its participants without the consent of the others. Participants are therefore informed and consent before the recording starts, which also changes what they will say.
Examples
A pension fund in French-speaking Switzerland is opening a portal where members simulate their own retirement capital themselves. One open question governs the plan: on what data and under what rules must the simulation be based for the member to understand it and for the fund to stand behind it.
| Activity | Type | Technique | Open question | Logistics | Supporting materials and preparation |
|---|---|---|---|---|---|
| A1 | Research | Document analysis | Which calculation rules do the pension regulations set and which ones are left to the administrators' interpretation | Analyst alone, 3 days | Pension regulations in force, pension plan, specimen pension certificate. No preparation by third parties. |
| A2 | Research | Interface analysis | Which fields does the administration system deliver to the portal, at what frequency and in what format | Analyst and one IT specialist from the software vendor, 2 half-days | Documentation of the feed, anonymised sample export. Access to the test system to be requested 2 weeks ahead. |
| A3 | Collaborative | Workshop | Which cases do the administrators handle today outside the written regulations | Half a day, 6 administrators and 2 member representatives, room on site, one facilitator and one scribe | Summary of A1 sent 3 days ahead, with the 9 points the regulations leave to interpretation. |
| A4 | Experiment | Paper prototyping | Between two layouts of the simulation screen, which one lets a member find their projected capital at 65 unaided | 5 members, 45 minutes each, on site | Two paper mock-ups, written task scenario, decision criterion fixed in advance: task completed unaided in under 3 minutes. |
The imbalance is the striking fact of this plan. The two research activities consume four analyst days and one day of an IT specialist from the software vendor, without engaging a single administrator or member. The vendor's IT specialist opens an access and answers from the record, with no preparation to supply. What research takes from the calendar is a delay: two weeks to obtain access to the test system. The workshop consumes half a day for eight stakeholders, so 32 hours of internal time, around CHF 3'200 at a fully loaded cost of CHF 100 an hour, and it covers only the nine points A1 left open. Without A1 and A2, the same half-day would have served to read the regulations out loud. A4 is the only activity whose answer existed nowhere before the test: neither the reading nor the session would have established which of the two layouts lets a projected capital be found unaided.
The results leave these four activities unconfirmed, and confirmation matters most for A1, A2 and A4, which are not facilitated elicitation.
Visualisations
Three objects come out of this technique and they do not call for the same support. The choice and mixing of the types is a structure: a need, three parallel routes that one initiative takes together, a single point of convergence and a follow-on. What has to be seen is the convergence, which a three-item list would destroy by suggesting that one picks a type and sticks to it. It is drawn. The activity plan is one row per activity and six columns of text, so a table. The profile of the three types follows the same rule, three rows and four columns.
The structure is also a check on a plan being written. Three questions arise. Are the three routes taken, or does the plan rest entirely on sessions. Does each activity carry a written open question or only a subject. Do the activities converge on a single dated plan, or does each analyst keep their own.
Cost
| Phase | Level | Justification |
|---|---|---|
| Preparation | High | The dominant item. Scoping the open question of each activity, choosing the types and the techniques, obtaining access to the sources, booking people's time, assembling the supporting materials and preparing the participants. The lead time on a piece of test equipment belongs to preparation, separately from the session that will use it. |
| Execution | Medium | The cost of conducting belongs to the techniques selected and runs from solitary reading to an eight-person workshop. What falls to the activity itself is arbitration in session and the switch from one type to another when a question turns out to be misplaced. |
| Documentation | Medium | The plan is kept up to date at each activity, and each result is recorded with its source, its date and the activity that produced it. This traceability link is what makes confirmation possible: it costs little to maintain continuously and a great deal to reconstruct. |
The total cost is decided in preparation, in the phase that looks the least productive, and it is paid in stakeholder time, which is booked weeks ahead and cannot be made up.
Tooling
The plan itself fits in a spreadsheet or a page of a team wiki, one row per activity. A template reused from one project to the next is enough, and a specialised tool that discouraged keeping the plan would be a step backwards.
The requirements tools (Jira, Polarion, ReqView, DOORS) carry the output: each result attached to the activity and the source that produced it. That link is what makes it possible, a year later, to answer "where does this requirement come from".
The collaborative type at a distance relies on video conferencing and a shared whiteboard (Miro, Mural, the Teams whiteboard). The distributed format changes what is playable: the interview transposes well, the game with physical artifacts much less so.
The survey platforms (LimeSurvey, which can be hosted in Switzerland, or Microsoft Forms) serve distribution at scale, subject to the place of processing as soon as an answer is attributable to a person.
Research uses the search engine of the document repository, an optical character recognition engine for digitised archives and SQL or OpenRefine when the source is a data set rather than a text.
Experiments run from paper and a marker pen, which remain the fastest instrument for deciding between two screen layouts, to Figma or Balsamiq for a clickable mock-up, then to an A/B testing tool when the question bears on a whole population rather than on five people.
Automatic transcription and summarisation are run on an instance whose place of processing is known and which does not train its models on the inputs, as soon as personal data appears in the recording.
That leaves diaries and room booking, the most mundane item in the plan and the one that wrecks the most sessions.
Sources
- IIBA, A Guide to the Business Analysis Body of Knowledge (BABOK Guide) v3, ch. 4 Elicitation and Collaboration, §4.1 Prepare for Elicitation, §4.2.2 Conduct Elicitation (Description), §4.2.6 Conduct Elicitation (Techniques) and §4.3 Confirm Elicitation Results: the three types of elicitation and their definitions, the parameters that govern the choice of a technique, the list of techniques usable to conduct elicitation, the elicitation activity plan and its content, the preparation of stakeholders, the supporting materials and the confirmation of results.
- IIBA and Agile Alliance, Agile Extension to the BABOK Guide, v2 (2017), §7.1 Backlog Refinement, §7.2.2 Behaviour Driven Development (Description), §7.19.2 Story Elaboration (Description) and §7.20 Story Mapping: the four agile techniques listed in the collaborative type, among them the Three Amigos session, which the guide names and defines inside Behaviour Driven Development.
- Federal Act on Data Protection (FADP), SR 235.1: the framework for processing the personal data held in the sources studied and in the recordings of a session.
- Swiss Criminal Code (SCC), SR 311.0, art. 179ter: the recording, without the consent of the other participants, of a non-public conversation to which one is party.
- FDPIC, Artificial intelligence and data protection: the FADP, being technology-neutral, applies directly to processing carried out by means of artificial intelligence.