Discovery Method Session Templates

Three parallel session templates for qualitative process discovery, structured consistently as Preparing · Roles/Principles · Session flow · When NOT to use. Designed to be used as slide content or as session-planning checklists.

For the method-selection decision (which template to use), see syntheses/qualitative-discovery-method-selection-matrix. For the full operational interview guide, see syntheses/interview-structuring-for-process-models.

1. Depth Interview

Preparing the interview

Document analysis / informal chat first — never walk in blind.
Formulate working hypotheses — e.g.:
- Role X performs tasks J and K, after receiving data from Role Z.
- Role X sends the final output to Role Y.
Interview guide tailored to the role being interviewed.
Aim: confirm or reject the hypotheses.

The interviewer

Keeps the interview on track.
Listens; does NOT create a model during the interview.
Keeps detail at activity level — not step level (steps can be elicited later if needed for value-added / waste analysis).
Validates an existing draft model through hypotheses (iterations 2+).
Translates the model to natural language when validating — never shows raw BPMN (sources/2012-ottensooser-graphical-vs-textual: n=196, untrained readers gain no statistically significant understanding from BPMN alone).

Session flow (60 min)

Short intro (5 min) — welcome, purpose, recording consent, scope confirmation.

Structured (40 min)

Opening: "What tasks are you responsible for?"

Traversal — mix forward and backward across interviewees:

Forward: "how does your work start?" → "what happens next?"
Backward: "when is the job done?" → trace inputs needed to get there
Per step: inputs needed · decisions taken · output produced · resource forwarded to

Completeness: "what other processing options exist here?" · "after you do A and B, you do C — is that right?"

Rainy-day (derived from exception taxonomy — do not skip):

"How did you handle your most difficult case?"
"What happens if the customer doesn't reply on time?"
"What do you do when [system] is down?"

Frequency: "How often does this happen? When was the last time?"

Unstructured (15 min) — "Is there anything else I haven't thought of?"

When NOT to use

The process is well-defined and uncontroversial → document analysis + automated discovery cheaper.
The process has severe role-conflict → workshop resolves faster.
Multiple roles must be reconciled in real time → workshop.

Sources

sources/2018-dumas-fundamentals-of-bpm-ch5-discovery §5.2.2 · sources/2014-sharp-whats-wrong-with-this-process (stakeholder-lens question packs in syntheses/interview-structuring-for-process-models §1.3)

2. Workshop

Preparing the session

Scope first — sketch TRAC on flipchart (Trigger · Result · Activities · Cases) before the session starts.
Stakeholder roster covers customer-side, every participant role, and the owner.
≤10–12 participants per session.
Pre-read: policies, forms, existing docs.

The roles (dedicated people — not the analyst)

Facilitator — controls parole, neutral.
Process modeller — draws live.
Scribe — parks unresolved questions.

The facilitator

Starts with sticky-notes, no BPMN gateways yet.
Holds granularity at subprocess / activity level — lifts up from micro-steps.
Encourages introverts; limits talkative participants.
Balances criticism; surfaces disagreement as insight, not conflict.
Translates model to natural language when validating — never shows raw BPMN.

Session flow (~3 hours)

Short intro (5 min) — welcome, purpose, recording consent, ground rules.

Build (90 min)

"What triggers this process?"
"What result is expected — by whom?" (per stakeholder)
"What are the 5 ± 2 milestones in between?"

Sticky notes on the wall. No gateways yet. Subprocess-level granularity.

Assess (45 min) — walk the model one enabler at a time (see concepts/six-enablers-framework):

Workflow
Information systems
Motivation & measurement
HR & organization
Policies & rules
Facilities

Blanket "what's wrong?" yields defence ("it's worked for 30 years"). Single-enabler focus elicits dysfunction in every dimension.

Close (15 min) — confirm decisions, assign owners, set next session.

Cadence across multiple sessions

3–5 sessions of ~3 h each.
Session 1: sticky-notes build.
Session 2: BPMN primer + validate the consolidated model.
Subsequent: exceptions, data, handoffs.
Consolidate offline between sessions.

Alternative: Sharp's 90-minute facilitated scoping session

Produces TRAC + Process Summary Chart + issues/goals table in one sitting. For time-pressured executives or emergent/adaptive processes where swimlane mapping is overkill. See sources/2014-sharp-using-scope-models.

When NOT to run a workshop

Strict-hierarchy / non-open culture — subordinates won't speak freely.
Strong politics / hidden agendas.
Cannot get all roles in the room at once (geographical distribution, shift work).

→ Use serial interviews instead.

Sources

sources/2018-dumas-fundamentals-of-bpm-ch5-discovery §5.2.3 · sources/2014-sharp-using-scope-models · sources/2014-sharp-whats-wrong-with-this-process

3. Observation

Preparing the session

Get access + consent from site owner AND individual operator.
Decide mode:
- Passive — shadow the operator (Hawthorne risk, requires site access).
- Active — analyst plays the customer / initiator (only sees customer-facing steps).
Pre-read existing docs so you know what you expect to see — and what's missing.
Prepare HTA-style field-note template (below).
Plan multiple sessions across shifts, days, operators — one session = sunny-day only.

The field-note template (per observed activity)

Per observed step, capture these 8 fields (sources/2010-salmon-hta-vs-cwa after Kirwan & Ainsworth 1992):

Field	What to note
a) Initiating cue	What triggers this step
b) Information required	What the operator needs to know
c) Controls / displays used	Where the information is, how it is accessed
d) Actions performed	Physical/system actions
e) Decisions taken	Choices made
f) Output produced	Result of the step
g) Feedback observed	How the operator knew it worked
h) Complexity / difficulty / workarounds	Pain points, shortcuts, "officially X, actually Y"

The observer

Does NOT interrupt the flow.
Parks questions — does not ask mid-observation.
Notes workarounds: "officially X, actually Y".
Notes cognitive moments (pauses, checks, corrections) — invisible decisions.
Does not model live.
Expects Hawthorne effect early in session; plans later sessions after behaviour normalises.

Session flow (~3 hours total, including debrief)

Pre-brief (10 min) with operator — purpose, scope, "behave normally", confirm analyst won't interrupt, consent to note-taking.

Observe (2–4 hours) — fill the field-note template per activity. Do NOT interrupt; park questions for debrief.

Debrief (30 min) — walk the operator through your notes:

"Is this what you did?"
"Why this step here?"
"What did you decide at X?"
"Does this happen every time, or only in case Y?"

Iteration (HTA cycle)

Observe → draft hierarchy / process model → validate with SME → revise → observe again. Return after draft model to observe edge cases, other operators, and other shifts. Never trust a single observation session.

When NOT to use observation

Remote or dangerous sites (offshore, etc.).
Cognitive work with no physical artefacts.
Processes too rare to observe in reasonable time.
Strongly proprietary / confidential steps where presence is refused.

→ Use interviews + think-aloud walkthroughs instead.

Known biases

Hawthorne — people work more diligently when watched. Counter by comparing early vs late session notes; plan multi-session coverage so behaviour normalises.
Sunny-day — short windows don't contain exceptions. Plan coverage across conditions (different days, shifts, case types).
Customer-facing only (active mode) — back-office remains invisible unless passive access is granted.

Sources

sources/2018-dumas-fundamentals-of-bpm-ch5-discovery §5.2.1 · sources/2010-salmon-hta-vs-cwa · concepts/hierarchical-task-analysis

Cross-template notes

All three methods follow the same iteration cycle: observe / interview / consolidate → draft model → validate with domain expert → revise. No single pass is ever sufficient (Dumas §5.2.2 phase model; HTA procedure; Sharp's "consolidate offline between sessions").
All three methods share the sunny-day pitfall — the normal path is over-represented. Counter with: rainy-day questions (interview), enabler-walk (workshop), multi-session coverage across conditions (observation).
All three methods keep granularity at activity or subprocess level during discovery — never step level. Step-level detail is a later phase, for value-added or error analysis.
None of the three stand alone — Dumas §5.2.4 recommends mixing methods. Start with document analysis, then layer interviews, workshops, observation as context permits.

syntheses/interview-structuring-for-process-models — full operational interview guide.
syntheses/qualitative-discovery-method-selection-matrix — which method to pick for a given context.
methods/process-discovery-methods — parent method page.
concepts/six-enablers-framework · concepts/hierarchical-task-analysis