What questions should you ask SMEs before building training?

Who this answer is for

This answer is written for the person who has been handed a training request and a name. The name belongs to a subject-matter expert who is busy, generous with their time in principle, and available for perhaps ninety minutes before their calendar closes again. You have one or two sessions to extract enough to design something defensible.

It is aimed at internal L&D designers, agency-side instructional designers inheriting a client’s SME, operations and safety leads documenting a procedure for the first time, and product enablement teams working from an engineer’s head rather than a spec. It sits within our wider material on SME and stakeholder discovery, and it assumes you are the one responsible for what gets built afterwards.

If you are looking for a list of questions to paste into a meeting invite, the list below will serve. The more useful part is what to do when the answers come back thin, because they usually will.

Why the answer changes by context

The same question set produces very different value depending on four variables, and it is worth deciding where you sit before the first session.

How proceduralised the work is. A fixed compliance procedure with a single correct sequence needs verification questions: does the documented process match what people do. Judgment-heavy work — triage, diagnosis, negotiation, exception handling — needs elicitation questions, because the expertise is not written anywhere and the SME cannot simply recite it.

Whether the SME is the practitioner or the owner. A process owner knows what should happen. A senior practitioner knows what does happen. These are different interviews. Conflating them is, in our judgment, a frequent reason training ends up describing an idealised workflow nobody follows.

How much source material exists. When the SME arrives with a deck, the interview shifts from extraction to interrogation of the artefact: what is out of date, what was never true, what is missing. That is a different conversation from a blank-page session, and it changes what you can realistically do with the material afterwards — see our note on converting existing PowerPoint training into eLearning.

How consequential errors are. Where a mistake is expensive, irreversible, or unsafe, the failure-focused questions in Gate 4 stop being optional and become the centre of the interview.

The Six-Gate SME Discovery Sequence

This is a gated sequence rather than a checklist. A gate is passed when you have an answer you could hand to another designer without explanation. If a gate does not pass, you do not proceed to the next one on the assumption it will resolve itself — you escalate, re-scope, or write down the gap as a stated project assumption.

Each gate below gives its purpose, the exact wording to use, the signature of a weak answer, and the follow-up probe. The weak-answer diagnostics are the part that matters. Anyone can ask “who is the audience”. Recognising that “everyone in operations” is a failed answer, and knowing the next sentence out of your mouth, is the skill.

Gate 1 — Trigger: why this, why now

Purpose: establish whether training is the right response at all, and what will be judged as success.

• “What happened that made this a project this quarter rather than last year?”
• “If we build nothing, what goes wrong, and to whom?”
• “Six months after launch, what would you expect to be different that someone other than you would notice?”

A weak answer sounds like: “It’s an annual refresh,” “leadership asked for it,” or “people just need to be aware of the new policy.” Awareness is not a performance outcome, and a refresh cycle is a calendar fact, not a problem statement.

Ask next: “Walk me through the last incident this would have prevented.” If no incident exists, ask “Is the problem that people do not know how, or that something in the environment stops them?” That single question separates a knowledge gap from a tooling, staffing, or incentive gap — the distinction Robert Mager and Peter Pipe built Analyzing Performance Problems around. If it is an environment problem, say so in writing before you build a course that cannot fix it.

Gate 2 — Population: who is actually in the room

Purpose: define the real learner spread, not the org-chart abstraction.

• “Who is the least experienced person who will have to do this unsupervised?”
• “Who will find this insultingly basic, and how many of them are there?”
• “Where and on what device will they take it — desk, shop floor, vehicle, shared terminal?”
• “What prior training does the organisation assume they already have, and is that assumption true?”

A weak answer sounds like: “All staff.” “Everyone in operations.” Any answer that describes a distribution list rather than a job.

Ask next: “Name three real people who will take this and tell me how their day differs.” Then: “Which of those three would fail an assessment written for the other two?” If the spread is too wide to serve with one artefact, you have discovered a curriculum decision, not a content decision.

Capture device and environment constraints here too, because they set the accessibility requirements you will have to meet later. WCAG 2.2, a W3C Recommendation, adds success criteria including Target Size (Minimum) and Dragging Movements, both of which bear directly on whether something designed at a desk can be operated on a tablet in a vehicle. That is an established fact about the standard. The related cost claim — that conformance is cheaper designed in than retro-fitted — is our professional judgment, not a measured finding.

Gate 3 — Trace: walk the work, do not summarise it

Purpose: get the actual sequence of the task, with decision points, at the grain of real behaviour.

• “Take me from the moment the task lands in front of someone to the moment they are finished with it. Do not summarise — narrate one real instance.”
• “What is in front of them while they do this? Which screen, form, tool, or document?”
• “Where in that sequence does someone have to decide rather than follow?”
• “What does ‘done well’ look like, and who checks?”

A weak answer sounds like: a description in the passive voice with no artefacts named — “the request is reviewed and then approved.” Nouns without screens, and verbs without actors, mean you are receiving a policy summary rather than a task trace.

Ask next: “Can you show me?” Watching the work on a screen-share is, in our judgment, worth more than any verbal description of it. Failing that: “What is the first thing they click?” and “What would I have on my desk at that moment?” Keep pulling until the narrative contains objects. A trace that names artefacts converts directly into scenario steps when you reach the storyboard stage; a trace that does not will stall there.

Gate 4 — Failure library: where it goes wrong

Purpose: build the error inventory that will become your practice items, distractors, and assessment.

• “What are the most expensive mistakes made in this task recently?”
• “Which step gets skipped when someone is behind schedule?”
• “Which errors get caught immediately, and which stay hidden until much later?”
• “What does the rework or escalation queue mostly consist of?”
• “What mistake do you personally correct most often, and does correcting it ever stick?”

A weak answer sounds like: “People just need to read the procedure,” or “carelessness.” Attribution to attitude usually indicates, in our judgment, that the SME has witnessed the failure without analysing it.

Ask next: “Describe the last time a careful, conscientious person got this wrong anyway.” That reframe removes blame and usually produces the genuinely useful cases — ambiguous inputs, misleading defaults, two rules that conflict. Then: “What made it look right at the time?” The answer to that question is the material for your best distractors.

Gate 5 — Expert delta: what the SME cannot easily say

Purpose: recover the tacit judgment that separates a competent novice from an expert. In our judgment this is the gate most discovery interviews stop short of, and it is the reason the sequence is gated rather than listed.

• “If I followed the written procedure exactly and did nothing else, where would I still get a poor result?”
• “Show me two cases that would look identical to me but that you would handle differently. What tells them apart?”
• “What do you notice before anyone else notices it? What is the cue?”
• “What does a competent second-year person get wrong that a ten-year person never gets wrong?”
• “When do you stop following the process and start using judgment, and what triggers that switch?”
• “What part of this would you not hand to someone else yet, and why not?”

A weak answer sounds like: “You just get a feel for it.” “Experience.” “You’ll know.” These are not evasions; they are the honest and expected response. It is an established finding in the expertise literature — set out in The Cambridge Handbook of Expertise and Expert Performance and applied to practice in Crandall, Klein and Hoffman’s Working Minds — that skilled performance becomes automatic and therefore resistant to verbal report, and that experts routinely omit steps and cues when asked to describe their own work. The proportion omitted varies by study and by task, so treat any single headline percentage with caution.

Ask next: stop asking about the general case and anchor to one incident. “Think of a specific time when your read of the situation differed from what a colleague would have said. Start at the first moment you felt something was off.” Then walk it forward slowly, asking at each turn: “What did you see or hear at that point?” and “What else could it have been, and how did you rule that out?” This incident-anchored, cue-by-cue walk is the core of the Critical Decision Method set out by Klein, Calderwood and MacGregor; the cognitive task analysis literature reports that incident-anchored probing surfaces cues and decision points that direct questioning leaves unspoken.

Gate 6 — Authority: whose answer counts, and when does it expire

Purpose: protect the build from contradiction, late reversal, and silent obsolescence.

• “If you and another expert disagree on a step, who decides?”
• “Which document is the source of truth, and who owns it?”
• “What in this is scheduled to change in the next twelve months?”
• “Who has to approve the content, and who merely has opinions about it?”
• “How much of your time can you give to review, and by when?”

A weak answer sounds like: “We’ll circulate it to the team.” Circulation is not approval. In our professional judgment, an undefined reviewer group is among the more common causes of schedule slippage — a planning heuristic rather than a measured finding, and one of the reasons review cycles sit behind the ranges in how long it takes to build an eLearning course.

Ask next: “Name the person who can say yes on their own.” Get a name, not a function. Then ask what happens if that person is unavailable for two weeks.

How to run the sessions, step by step

1. Read everything first. Never spend SME time on facts a document already contains. Arrive with a list of contradictions you found in the existing material.
2. Send Gates 1 and 2 in advance; keep Gates 4 and 5 back. Trigger and population questions benefit from preparation. Failure and expert-delta questions do not — in our judgment, a prepared answer to “how do people fail” produces the sanitised version.
3. Split the sessions. Session one covers Gates 1 to 3. Session two covers Gates 4 to 6, after you have a draft task trace to react to. This is professional judgment rather than a measured result: two shorter sessions with a draft in between tend to outperform one long session, because SMEs correct more readily than they generate.
4. Record and transcribe, with permission. Note-taking during Gate 5 competes directly with listening, and Gate 5 is where the cues live.
5. Play back the trace as a numbered sequence. Read it aloud and stop at each step. “Is that right?” surfaces corrections that open questions do not.
6. Write the gaps down as stated assumptions. Anything a gate failed to produce becomes a written assumption in the design document, with the name of whoever must resolve it and a date. Unwritten gaps become rework.

A worked example

The following scenario is hypothetical and constructed for illustration. It does not describe a real organisation or a real engagement.

A regional water utility asks for a two-hour eLearning module on a revised hydrant inspection procedure. The named SME is a field supervisor with eighteen years of service.

Gate 1 produces “the procedure was updated in March” — a weak answer. The follow-up, “what goes wrong if we build nothing,” yields something better: inspection records have been failing audit because condition ratings vary between crews. The real problem is rating consistency, not procedural ignorance.

Gate 2 reveals two populations, not one: apprentices in their first year, and crews with a decade of habit who believe the new rating scale is a downgrade of their judgment. One module cannot serve both intentions.

Gate 3, pushed for artefacts, establishes that the rating is entered on a tablet form with a free-text notes field that nobody reviews.

Gate 4 produces the error inventory: borderline corrosion cases rated inconsistently, and the flow-test step skipped in cold weather.

Gate 5 is where the project changes. Asked to show two hydrants that would look identical to an outsider, the supervisor describes surface staining that is cosmetic versus staining that indicates a seeping seal — a distinction resting on the pattern and dryness of the deposit, mentioned in no document. Asked what a second-year technician gets wrong, the answer is that they rate what they see rather than what they see plus what the last two inspections recorded.

Gate 6 confirms the asset management lead, not the SME, owns the rating definitions.

The resulting design is not a two-hour procedural module. It is a short procedure refresher plus a photo-based calibration exercise on borderline cases, a job aid for the tablet form, and a request that the notes field be reviewed. That redesign came entirely out of Gate 5 — the gate a flat question list gives you no particular reason to reach.

Common mistakes

• Asking for content instead of performance. “What do they need to know” invites a syllabus. “What do they need to do, and where do they get it wrong” invites a design.
• Accepting the org chart as the audience. “All staff” is a distribution list.
• Treating “you just get a feel for it” as the end of the conversation. It is the beginning of Gate 5.
• Interviewing one SME. A single expert gives you one idiosyncratic model of the work. Where the task carries real consequence, a second practitioner is worth the scheduling cost, principally because the disagreements between two experts mark the genuine judgment points. That rationale is professional judgment.
• Letting the SME design. In our judgment, SMEs asked “how should we teach this” tend to propose the format they themselves experienced. Ask them what must be true of a competent performer; the format decision is yours.
• Failing to name an approver. Review by committee without a decider is a familiar route to a stalled build.
• Skipping the constraints conversation. Device, language, offline access, and accessibility requirements are discovery inputs, not production details. WCAG 2.2 is a published W3C Recommendation; our judgment, stated as such, is that conformance costs less when designed in from the first storyboard than when added to a finished module.

Gate decision table

The pass thresholds in the middle column — five error cases, two undocumented cues — are planning heuristics we offer as defaults, not measured thresholds. Adjust them to the consequence of the task.

Where discovery is itself the deliverable — because the knowledge is scattered, the SMEs disagree, or nobody has yet established what the programme should contain — that work sits in curriculum consulting rather than in a production engagement.

Sources and review note

Editorial owner: IETERNUS Learning Systems. Last updated: 8 August 2026.

Established fact. That expert performance becomes automated and consequently difficult for experts to articulate, and that experts omit steps and cues when describing their own work, is a well-documented finding in the expertise and cognitive task analysis literature, summarised in The Cambridge Handbook of Expertise and Expert Performance. The incident-anchored interviewing approach in Gate 5 derives from the Critical Decision Method, published by Klein, Calderwood and MacGregor and elaborated for practitioners in Working Minds. The distinction between knowledge gaps and environmental barriers is a long-standing principle of performance analysis set out by Mager and Pipe. WCAG 2.2 is a published W3C Recommendation, and Target Size (Minimum) and Dragging Movements are among the success criteria it adds.

Professional judgment. The six-gate sequence itself, the gate ordering, the gate names, the weak-answer diagnostics, the specific question wording, and the pass criteria in the decision table are IETERNUS’s structured practice guidance — reasoned design heuristics, not measured findings. So are the following claims made in the body: that conflating the process owner with the practitioner is a frequent cause of idealised training; that an undefined reviewer group is among the more common causes of schedule slippage; that Gate 5 is the gate most discovery interviews stop short of; that attribution of error to attitude usually signals unanalysed failure; that watching work on a screen-share yields more than a verbal description; that SMEs asked how to teach something tend to propose the format they experienced; that accessibility conformance costs less designed in than retro-fitted; that withholding Gate 4 and Gate 5 questions from advance circulation produces less sanitised answers; that two shorter sessions with a draft in between outperform one long session; and that a second practitioner is worth interviewing on consequential tasks. None of these is a measured result, and none should be cited as one.

Assumption. This answer assumes you have direct access to at least one practising SME and some authority over the design. Where SME access is mediated through a client account team, or where the format has already been fixed by procurement, the sequence still applies but Gates 5 and 6 will need sponsor support to complete.

Sources cited by name: Crandall, Klein and Hoffman, Working Minds: A Practitioner’s Guide to Cognitive Task Analysis, MIT Press; Klein, Calderwood and MacGregor, “Critical Decision Method for Eliciting Knowledge”, IEEE Transactions on Systems, Man, and Cybernetics; Mager and Pipe, Analyzing Performance Problems; The Cambridge Handbook of Expertise and Expert Performance, Cambridge University Press; WCAG 2.2, W3C Recommendation.

Learning topic: SME and Stakeholder Discovery