Research Planning

Research Planning

> Part of the discovery-phase skill pack · discovery group · reads discovery-context.md (run profile-builder first if missing).

The skill that decides how the BA will learn — informed by access reality, not aspiration. In outsourcing, this almost always means a mix of methods, not pure interviews.

Step 1 — Read discovery context

Read discovery-context.md (section 4. Access & Data — drives the entire plan) and problem-canvas.md (anchors research questions to a hypothesis).

If discovery-context.md is missing, ask the BA inline: "(a) end-user access (direct / client-mediated / proxy-only / none); (b) which proxy data sources are available (SMEs / support tickets / analytics / none)?" — tag the output [ASSUMED ACCESS]. If problem-canvas.md is missing, recommend running problem-framing first; if BA overrides, tag research questions [NO-HYPOTHESIS]. Never block; recommend profile-builder for high-stakes work.

Step 2 — Match access level → method mix

| Access level (from context) | Primary method | Secondary | Skip | |---|---|---|---| | direct (BA can interview end users) | user-interviews (5-8 sessions to saturation) | competitive-scan, support-data-analysis if available | — | | client-mediated (interviews via client introductions) | user-interviews (3-5 sessions, accept slower recruitment) + sme-workshops | competitive-scan, secondary-research | — | | proxy-only (no interviews, but data + SMEs available) | sme-workshops + support-data-analysis | secondary-research, competitive-scan | user-interviews | | none (no users, no data, no SMEs) | secondary-research + competitive-scan | analogous-domain interviews if BA can recruit independently | user-interviews, support-data-analysis, sme-workshops |

State the chosen mix explicitly in the plan. If the BA wants to override (e.g., "I know we said proxy-only but I have a friendly client contact who agreed to one interview"), record it AND note the confidence-loss caveat.

Step 3 — Translate hypothesis to research questions

From problem-canvas.md Open questions, derive 3-5 research questions. Rule:

• Each must be answerable from the chosen method mix
• Each must be falsifiable (a "no" answer is possible)
• Drop quantitative questions ("what % of users do X?") — those are surveys/analytics, not interviews

Step 4 — If interviews are in the mix

Build ./discovery/interview-guide.md (separate artifact). Behavioral, not hypothetical:

• ✅ "Walk me through the last time you…"
• ❌ "Would you use X if we built it?"

Apply user-interviews skill conventions (see that skill for the full guide template). Pre-flight at least 1 pilot interview before booking 5+.

If interviews are NOT in the mix, skip this step and route to sme-workshops, secondary-research, or support-data-analysis skills directly when execution time comes.

Step 5 — Recruitment / data-request plan

Per chosen method:

• Interviews: target N, recruitment channels (client list / LinkedIn / community / Respondent), screener, stop rule (saturation)
• SME workshops: workshop count, format (1:1 vs group), duration, who facilitates, what artifacts they bring
• Secondary research: topic areas, sources (industry reports, competitor blogs, academic if relevant), time budget
• Support-data analysis: data sources to request from client, time window (e.g. last 90 days), volume estimate

Step 6 — Logistics

• Window: what calendar week(s) the data lands
• Interview/workshop length: 45-60 min (not 30)
• Recording + consent: decide before first session
• Note-taker: ideally a second person, especially for interviews
• Synthesis (next phase) start date

Step 7 — Success criteria

Pre-commit: "We'll know the research worked if we can update the problem-canvas current-workaround and target-user sections, and cite which evidence supported each change."

Output

• ./discovery/research-plan.md per ./template.md
• ./discovery/interview-guide.md IF interviews are in the mix (use the dedicated user-interviews template)

Append to _log.md: [research-planning | YYYY-MM-DD] mix: ; target_sessions: ; window: .

Anti-patterns

• Defaulting to interviews when access says proxy-only. "We'll figure out interviews later" usually means never. Plan for the access you actually have.
• Mixing segments without tracking. If you interview 3 admins and 2 end-users, label every note. Do not merge.
• Surveys disguised as interviews. "How often do you do X?" is a survey question. Reformulate to behavior: "Walk me through the last time…"
• No stop rule. Without saturation criteria, research expands to fill all available time.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: stanislavnianko
• Source: stanislavnianko/product-discovery-claude-skills
• License: MIT

Install and usage instructions live in the source repository linked above.