Current state
Walk the real path from trigger to outcome, not the official process diagram.
Library / 01 Business & Workflow Discovery
Reviewed Oct 2026AI Product Discovery Questions
Learn the real workflow before choosing AI, automation, or another product intervention.
01 / flow
From workflow to recommendation
Current workflowPain + baselineConstraints + exceptionsOptionsRecommendation + unknowns
02 / cards
Three starting lenses
Pain
Find delay, errors, rework, handoffs, workarounds, and expert judgment.
Baseline
Quantify volume, elapsed time, effort, escalation, and throughput.
03 / table
Questions that change the decision
| Ask | Why it matters |
|---|---|
| How often does this happen? | Scale, priority, and likely automation value. |
| Who handles it, and how do roles differ? | Different workflows hidden under one name. |
| What happens when normal steps fail? | Exceptions and handoff burden. |
| What information must people have? | Retrieval, data readiness, and permissions. |
| What would better look like? | An outcome measurable against the baseline. |
04 / cards
Stakeholders to hear
Business owner
Goal, cost, success criteria.
Frontline expert
Actual work, exceptions, judgment.
Newer user
Learning curve and hidden knowledge.
IT / operations
Integrations, maintenance, and access limits.
05 / table
Decision evidence
| Move forward when | Slow down when |
|---|---|
| Pain is recurring and measurable | Pain is rare or upstream process is unclear |
| Needed data exists and is permissioned | Critical information is missing or stale |
| A bounded pilot can prove value | Unmanaged exceptions dominate |
06 / list
Discovery note template
- Workflow: trigger → steps → handoffs → outcome
- Pain: where, for whom, how often, at what cost
- Baseline: time, volume, quality, rework
- Constraints: data, systems, permissions, policy
- Unknowns and smallest next evidence-gathering action
GOING DEEPER
Find the unofficial workflow
Questions that reveal hidden work
| Ask | What it reveals |
|---|---|
| Where do experienced people handle this differently from newer people? | Tacit expertise, hidden decisions, training gaps and variations in judgment |
| What happens when the normal path fails? | Actual exception volume, rework, escalations and manual fallbacks |
| Which side channels, spreadsheets or memories do people rely on? | The real workflow, not only the official process map |
| What information is required at each handoff? | Missing inputs, permissions, freshness and system-of-record boundaries |
Discovery warning: Stakeholders may describe an ideal process while the work actually runs through shared spreadsheets, informal chats, personal judgment and exceptions. Observe real cases from trigger through outcome, including how newer users differ from experts.
PRACTICAL FRAMEWORK / AI PRODUCT WORK
When faster AI moves the bottleneck
| Workflow | Before AI | After a faster drafting step |
|---|---|---|
| Support response | Agents spend time drafting replies. | Drafting shrinks, but correction, source verification or supervisor approval queues grow. |
| What to measure | Drafting time and response throughput. | Total resolution time, correction burden, escalations, repeat contacts and customer outcome. |
Discovery question: If this step became ten times faster, where would work pile up next? Test the end-to-end workflow, not just local speed.