Consultant Interview Guide — Eliciting Evidence for the Company Graph

This guide is for consultants who need to convert scattered memories, partial operational context, and cross-functional interviews into graph-quality evidence.

Related guides:

• [User guide](/user-guide) for end-to-end workflow
• [Builder guide](/builder-guide) for field-level model entry

Why this guide matters

Consultants have the hardest role in discovery: interviewees often remember incidents but not structured evidence. Your job is to ask leading/searching questions that move answers from vague narratives to model-ready facts.

For each session, prioritize evidence that changes ranking, not completeness.

Interview objective (what you should extract)

Aim to leave each session with:

• 1-2 high-value workflows mapped as processes
• Linked pains with severity/frequency (and rough cost if available)
• Data/asset clarity for those workflows (`accessible` vs not yet accessible)
• At least one clear bottleneck skill
• Missing-edge cleanup: `serves`, `uses`, `reuses`, `bottlenecked_by`

60-minute interview structure

Recommended opener: "Let's use real recent examples. I will ask for concrete instances, then convert them into the evidence map."

Memory-first funnel (vague -> concrete)

When interviewees say "it depends", "often", or "many", use this funnel:

1. Recent event trigger: "Tell me about the last time this happened."
2. Timeline anchor: "Was that this week, month-end, or quarter-end?"
3. Artifact anchor: "Which system/sheet/ticket/doc did you open?"
4. People anchor: "Who had to approve or intervene?"
5. Volume anchor: "How many items per day/week/month?"
6. Failure anchor: "Where did it stall, bounce back, or need rework?"

This reliably turns abstract responses into graph-ready evidence.

Leading/searching question bank by graph section

Use these prompts when responses are thin.

1) Offerings and segments (`serves` edges)

Primary:

• "What do you sell repeatedly, and what is still bespoke each time?"
• "Which customer types buy each offering most often?"

Searching prompts:

• "If I opened your last 10 won deals, which 2-3 offerings appear most?"
• "Which segment asks for exceptions most often?"
• "Which offering has the longest handoff from sales to delivery?"

Capture:

• Offering name, delivery mode/components (`manual`, `rule_based`, `judgment`)
• Segment names and explicit Offering -> `serves` -> Segment links

2) Processes (workflow backbone)

Primary:

• "Walk me through the process step-by-step from trigger to completion."
• "Where are delays, queues, or repeat loops most common?"

Searching prompts:

• "What happens at month-end / quarter-end / audit prep?"
• "Which step fails when your best person is unavailable?"
• "Where do people copy-paste between systems?"
• "Which inbox/queue/ticket list grows the fastest?"

Capture:

• Process name, rough volume (`low/med/high`), repetitiveness
• Stage where work waits, rework points, exception paths

3) Assets and data (`uses` edges, accessibility)

Primary:

• "What data or systems are used at decision time in this workflow?"
• "Can your team access that data today without special engineering?"

Searching prompts:

• "Which export/report does the team pull before making a decision?"
• "If I asked for last 90 days of this process, where would that come from?"
• "What data exists but is trusted by no one?"
• "What is in spreadsheets because core systems do not capture it?"

Capture:

• Asset type (`data`, `tech`, `physical`)
• Accessibility truth (`accessible` only if usable now)
• Process/Offering -> `uses` -> Asset links

4) Pains and impact evidence

Primary:

• "What is the cost of this issue in time, money, risk, or customer impact?"
• "How often does it happen?"

Searching prompts:

• "What did this delay block last week?"
• "How many escalations or SLA misses came from this?"
• "What gets worked after hours because of this step?"
• "What do you warn new hires about on day one?"

Quantification prompts:

• "Is this daily, weekly, or monthly?"
• "Is it 5 items, 50, or 500?"
• "Is rework minutes, hours, or days?"
• "Ballpark annualized cost: <$50k, $50k-$250k, or >$250k?"

Capture:

• Pain statement attached to a node, severity/frequency, optional annual cost

5) Skills and bottlenecks (`bottlenecked_by` edges)

Primary:

• "Which decisions only 1-2 people can make reliably?"
• "Where does quality drop when experts are unavailable?"

Searching prompts:

• "What work pauses during leave/holiday periods?"
• "What takes longest to train a new hire to do safely?"
• "Which judgment calls are hard to document?"

Capture:

• Skill entity, scarcity (`med/high` when true), Process -> `bottlenecked_by` -> Skill

6) Projects and abstractions (`reuses` edges)

Primary:

• "Which past delivery pattern has repeated across clients/use cases?"
• "What did you build once that keeps getting adapted?"

Searching prompts:

• "If you strip industry words, what is the reusable pattern?"
• "Which past project would you clone first under time pressure?"
• "Where do proposals repeatedly reuse the same approach?"

Capture:

• Project summary, abstraction (domain-neutral pattern), Offering -> `reuses` -> Project

7) Constraints (feasibility and trigger activation)

Primary:

• "What can kill this initiative even if the idea is good?"
• "Which compliance/security/risk constraints are non-negotiable?"

Searching prompts:

• "What is the biggest reason previous automation efforts stalled?"
• "Where does legal/compliance review add cycle time?"
• "What budget, integration, or governance gates usually block rollout?"

Capture:

• Constraint type (`regulatory`, `risk`, `stack`, `budget`, `talent`, etc.)
• Scope and practical effect on delivery

Edge-completion checklist (ask before ending)

Ask these explicitly to prevent under-linked graphs:

• "Which offerings serve which segments?" (`serves`)
• "Which processes or offerings use which assets?" (`uses`)
• "Which offerings reuse which past projects?" (`reuses`)
• "Which processes are bottlenecked by which skills?" (`bottlenecked_by`)
• "Which pains belong to which node?" (impact grounding)

If an edge is uncertain, mark it as follow-up instead of guessing.

High-yield probes for hard-to-remember evidence

Use these when recall is incomplete:

• Calendar probes: "What changed at month-end, quarter-end, annual audit?"
• Exception probes: "What happens when data is missing, late, or inconsistent?"
• Escalation probes: "Which issues get escalated to managers or experts?"
• Workaround probes: "Where are side spreadsheets, shadow tools, manual trackers?"
• Handoff probes: "Where does ownership change between teams?"
• Artifact probes: "Show me the report/template/checklist used before decisions."
• Top-N probes: "What are the top 3 reasons this workflow misses target?"
• Last-time probes: "Describe the most recent incident end-to-end."

Interview anti-patterns to avoid

• Asking "Do you have data?" without asking where, who can access it, and how often it is used
• Capturing pains without linking them to a process/offering/project
• Treating opinions as evidence when no incident, artifact, or metric is cited
• Marking assets `accessible` because they exist, not because teams can use them now
• Trying to complete the whole graph in one session instead of closing top gaps iteratively

Session output template

Use this checklist format in your notes after each call:

• Session focus (department/persona):
• Top workflows discussed:
• New/updated processes:
• New/updated pains (with severity/frequency):
• New/updated assets (and accessible status):
• New/updated skills and bottlenecks:
• New/updated projects and abstractions:
• Constraints captured:
• Edges confirmed (serves/uses/reuses/bottlenecked_by):
• Unknowns to resolve:
• Artifacts requested (report, export, ticket sample, SOP):

Use this output to update the builder, run Analyze, and generate the next interview agenda.