Study guides / CCAO-F / Domain 4

Workflow Integration and Solution Design · Lesson 2 of 6

4.2 — Using Claude for Research and Planning

Scope a research task into clear sub-questions and turn Claude's findings into an actionable plan, rather than asking one shallow question and taking the first answer.

Claude can be a genuinely useful research partner for business questions — competitive landscape scans, market sizing, vendor comparisons, policy summaries, internal-document synthesis — but the quality of that research depends heavily on how the question is scoped, not just on the fact that Claude is capable of research. claude.ai's research mode (where available on a plan) lets Claude search across multiple sources, follow up on what it finds, and synthesize a longer, cited answer, rather than answering purely from what it already knows. Even without research mode, the same scoping discipline applies to any research-style prompt.

Shallow Question vs. Scoped Research Task

A shallow, one-shot research question looks like: "What's the competitive landscape for project management software?" This is answerable, but it's so broad that Claude has to guess what you actually care about — pricing, feature gaps, target market, recent funding news, all of it are legitimate readings of that question, and the resulting answer will likely graze all of them shallowly.

A properly scoped research task breaks the same underlying interest into explicit sub-questions tied to an actual decision: "We're evaluating whether to add a Gantt-chart feature to our project management tool. Research: (1) which of our top 5 competitors already offer Gantt charts, (2) how they price that feature relative to their base plan, (3) any public complaints or praise about their Gantt implementations in reviews or forums, and (4) whether any competitor has removed or downplayed this feature recently, and if so why." Each sub-question is independently answerable, and together they point at a specific decision — whether to build the feature — instead of at a vague topic.

From Findings to a Plan

Research that stops at a findings summary is only half the value. The second step is asking Claude to turn findings into a plan: what should happen next, in what order, with what owner and what open question still needs a human decision. This is a distinct prompt or follow-up, not something to bundle into the original research request — asking for findings and a plan in the same breath tends to produce a plan built on a still-shallow set of findings, echoing the task-decomposition principle that sequentially dependent work benefits from being split into steps you can check before building on them.

A good planning follow-up names constraints explicitly: budget, timeline, who has final say, and what would change the recommendation. "Given this research, propose a 3-step plan for deciding whether to build the Gantt feature, including what data we'd still need to collect and who should be involved in a go/no-go decision" turns a research summary into something a team can actually act on, while still leaving the go/no-go call to a human.

Key Concept

Scope research into explicit sub-questions tied to a real decision, rather than one broad topic question. Treat "turn findings into a plan" as a distinct, later step — not something to request in the same prompt as the research itself — so the plan is built on findings you've actually reviewed.

Common Exam Distractor

Watch for an option that treats "ask Claude to research and produce a final recommendation in one step" as equivalent to a properly scoped research task. Bundling everything into one shot removes the checkpoint where a human reviews findings before they become the basis for a plan or decision — this is the same sequential-dependency problem task decomposition addresses elsewhere in the exam.

Exam traps

Practice question

A team wants to know whether to enter a new geographic market. Someone drafts this single prompt: 'Research the market for us and tell me whether we should expand there.' What's the most accurate critique of this approach?

  • A It's a well-scoped research task because it clearly states the business goal (deciding whether to expand)

    Stating a goal isn't the same as scoping the research. The prompt still doesn't break the question into specific, independently answerable sub-questions, and it asks for a final decision in the same step as the research.

  • B It should be split into explicit sub-questions (e.g. market size, competitor presence, regulatory barriers, entry cost) reviewed first, with the go/no-go recommendation requested as a separate follow-up once those findings are reviewed Correct

    This applies both parts of the lesson: scoping into concrete sub-questions tied to the decision, and treating the findings-to-plan step as separate so a human can review findings before they become the basis for a recommendation.

  • C It's fine as written, since research mode will automatically break the question into sub-questions internally

    Even where research mode improves sourcing, a vague prompt still risks Claude guessing at which angle (cost, competitors, regulation, demand) matters most to the team, and it still skips the human review checkpoint between findings and recommendation.

  • D It should be shortened further, dropping the expansion question entirely and just asking for a general market overview

    Removing the actual decision context makes the research less targeted, not more — the fix is to add specific sub-questions tied to the decision, not to strip context out.

Build exercise: Scope a Research Task and Turn It Into a Plan

Beginner · 25 minutes

You'll practice:

  1. In claude.ai, ask a broad, single-sentence research question on a topic relevant to your work or interests (e.g. 'research the landscape for [some tool/market/topic] and tell me what to do'). Note how broad or unfocused the resulting answer feels.

    This establishes the baseline shallow-question failure mode before you see the scoped version for comparison.

    You should see: An answer that covers several different angles briefly rather than going deep on any one, and that doesn't clearly connect to a specific decision.

    Hints
    1. Pick a topic broad enough that it could reasonably be interpreted several different ways.
    2. Notice whether the answer asks you any clarifying questions back, or just guesses at what you meant.
    3. Keep this response open to compare against the next step.
  2. Start a new conversation (using research mode if it's available on your plan). Rewrite the same underlying interest as 3-4 explicit, independently answerable sub-questions tied to a specific decision, following the pattern in this lesson.

    This is the core scoping skill the exam tests: turning a vague topic into sub-questions anchored to an actual decision.

    You should see: A noticeably more targeted, structured answer that addresses each of your sub-questions distinctly, ideally with sources or citations if research mode is available.

    Hints
    1. Make sure each sub-question could be answered on its own, without needing the others answered first.
    2. Tie the sub-questions explicitly to a decision your team or you would actually make.
    3. If using research mode, check whether it cites sources you could verify.
  3. Review the findings from step 2 for anything that looks off or worth double-checking. Then, in a separate follow-up message, ask Claude to turn the reviewed findings into a short plan: next steps, owner, timeline, and what open question still needs a human decision.

    This practices treating 'plan' as a distinct, later step built on findings you've actually reviewed, rather than bundling research and recommendation into one shot.

    You should see: A plan with concrete next steps and an explicit call-out of what still requires human judgment, clearly grounded in the specific findings from step 2 rather than generic advice.

    Hints
    1. If you skipped reviewing the findings, go back and actually read them before asking for the plan — that review is the point of this step.
    2. Ask Claude to reference specific findings by name in the plan, not just produce generic next steps.
    3. Check that the plan explicitly flags at least one decision point for a human, rather than presenting itself as a final answer.

Sources