Practice — CCAO-F Domain 3: Product and Model Selection (12%)
12 scenario-based MCQs. Answer key + explanations at the bottom.
Q1
An operations team tags several thousand inbound support emails per day with a category and an urgency label. Each tag is one or two words, turnaround must feel instant, and the sheer volume makes speed and cost the dominant constraints. Which model choice best fits this task?
A. Opus, because the most capable tier gives the most reliable labels B. Sonnet with extended thinking enabled on every email C. Haiku, the tier built for high-volume, latency-sensitive classification D. Opus with extended thinking so each label is double-checked
Q2
A strategy lead is preparing a board-level competitive analysis: synthesizing several long documents, weighing conflicting evidence, and building a multi-part argument across many working turns. Latency doesn't matter and the work happens only a few times per quarter. Which model choice is the best match?
A. Opus, because the task demands deep reasoning and long-horizon planning B. Haiku, because infrequent tasks should default to the cheapest tier C. Sonnet, because the balanced default covers every business scenario D. Haiku run several times, with the outputs merged by hand
Q3
A marketing manager uses Claude all day for campaign briefs, email drafts, and meeting-note summaries. Nothing is high-volume classification and nothing needs long-horizon strategic reasoning, but the team does watch its usage. Which model should be the daily default?
A. Haiku, since the team is watching usage B. Opus, since writing quality is paramount in marketing C. Alternating between Opus and Haiku message by message to average out the cost D. Sonnet, the balanced tier built for day-to-day work
Q4
A consultant pasted a 90-page PDF into a chat, and after a dozen follow-up questions Claude's answers have turned vague and now miss details from the document. A colleague suggests switching the conversation to Opus. Which approach best resolves the problem?
A. Switch the conversation to Opus, since a more capable model retains documents better B. Start a fresh conversation and persist the PDF as knowledge in a Project C. Switch to Haiku so answers are generated before the context runs out D. Re-paste the full 90-page PDF into the same chat every few follow-up questions
Q5
An educator built a classroom activity where students submit a topic and Claude generates a short quiz question live while the class waits. She chose Opus "for the best quality" and enabled extended thinking, and now every response feels painfully slow. What is the best change?
A. Switch to a faster tier and turn off extended thinking for this task B. Keep Opus and disable thinking only, since the model tier itself doesn't affect speed C. Keep the current setup and write more detailed prompts to speed up generation D. Enable extended thinking on Haiku to get Opus-level quality at Haiku speed
Q6
A proposal writer answers customer RFPs every week using the same six reference documents and the same tone guidelines. Today she starts a new chat each time, re-uploads all six files, and re-explains the tone. Which approach best streamlines this recurring work?
A. Keep one single never-ending conversation for all RFPs so the files and tone never need repeating B. Ask Claude to remember the files and tone so future chats can recall them automatically C. Create a Project with the tone guidelines as instructions and the six documents as knowledge D. Paste the six documents into an artifact so they can be reused across chats
Q7
A team built a "Company Brain" Project and uploaded 200+ files covering every department. Since then, Claude often ignores the Project's custom instructions and gives shallow answers about any single document. What is the most likely cause?
A. Projects silently drop any files beyond a small fixed limit, so most of the uploads were never stored B. Project knowledge counts against the context window, so the oversized library crowds out the instructions C. Instruction-following is only reliable on Opus, and the Project is running a smaller tier D. Custom instructions apply only to the first message of each new conversation
Q8
A marketing analyst needs a briefing on how competitors have repositioned over the past three months, drawing on many external sources — news coverage, product announcements, analyst commentary. Which approach best fits this need?
A. Ask in a plain chat and rely on Claude's built-in knowledge of the market B. Switch the chat to Opus so its training data reaches more recent events C. Create a Project and list the competitors' names in its custom instructions D. Use research mode so Claude investigates across multiple current sources
Q9
A consultant has two recurring needs: (1) client deliverables that must always follow the firm's proprietary methodology documents, and (2) fresh scans of what has changed in a client's industry since last month. Which pairing of features fits best?
A. A Project holding the methodology docs for the deliverables; research mode for the industry scans B. Research mode for the deliverables; a Project holding last month's scan for the industry updates C. One Project holding the methodology docs, used to answer both kinds of request D. Research mode for both, re-uploading the methodology docs at the start of every session
Q10
An operations coordinator wants Claude to open the team's web-based inventory dashboard, read the current stock numbers, and enter reorder quantities into a form in another browser tab. Which surface is designed for exactly this?
A. claude.ai chat, pasting fresh screenshots of the dashboard and retyping the numbers by hand B. The Claude in Chrome extension, which navigates pages, extracts data, and fills forms C. Claude for Microsoft 365, which already covers the team's day-to-day business applications D. A Project with the dashboard's URL stored in its knowledge files for reference
Q11
A program manager must work through a shared folder of roughly 80 intake files: rename each to a standard convention, sort them into subfolders, and log every one in a tracking spreadsheet. The deliverable is the changed files themselves, not a written answer. Which is the best fit?
A. A plain chat that outputs the new file names for the manager to apply by hand B. An artifact containing a table of proposed names and folder assignments C. Research mode, so Claude can investigate each file before renaming it D. A Cowork session with access to the folder, run delegate-and-review
Q12
An operations lead wants every ticket from the company's in-house, proprietary ticketing system automatically triaged by Claude each night. No existing connector supports the system. What is the best move for a Claude associate in this situation?
A. Describe the ticketing system's API in a chat and ask Claude to establish the connection from there B. Paste each night's tickets into a fresh chat by hand and adopt that as the standing workflow C. Scope the requirements and hand the integration build to the developer team D. Switch the conversation to Opus, which can reach internal systems directly
Answers
Q1: C. High-volume, latency-sensitive classification is the textbook Haiku case — first stop in the model decision tree. Opus (A, D) buys capability the task doesn't need while raising both cost and latency, exactly the wrong direction when volume dominates. Extended thinking (B, D) trades latency for accuracy on hard problems and should never be enabled for trivial calls like two-word labels.
Q2: A. Deep reasoning over conflicting evidence plus long-horizon planning across many turns is the Opus profile, and the low frequency plus latency tolerance means its cost premium is acceptable. Sonnet (C) is the everyday default, but this scenario explicitly hits the "deep reasoning / long-horizon planning" branch that routes to Opus. Haiku (B, D) optimizes for speed and cost that this task doesn't prioritize, and merging repeated cheap runs by hand (D) doesn't add the missing reasoning depth.
Q3: D. Mixed day-to-day drafting and summarizing that is neither high-volume classification nor deep strategic reasoning is precisely what the balanced tier is for — Sonnet is the default when neither extreme applies. Haiku (A) is tempting on cost but is aimed at classification, routing, and high-volume flows, sacrificing quality this writing work needs. Opus (B) adds cost and latency without a deep-reasoning justification, and per-message alternation (C) is unmanageable and solves nothing.
Q4: B. Degradation after many follow-ups on a huge paste is context-window exhaustion — the classic wrong-layer distractor is "switch to Opus." The right response is restart/persist: a fresh conversation with the document moved into a Project so it's available as curated knowledge. A model swap (A, C) doesn't recover crowded-out working memory, and re-pasting the full PDF repeatedly (D) balloons the same conversation and accelerates the very problem being fixed.
Q5: A. The task is simple, latency-sensitive generation, so the fix is to move down the capability ladder and drop thinking — capability increases bring cost and latency increases, and extended thinking deliberately trades latency for accuracy the task doesn't require. B is half right about disabling thinking but rests on a false premise, since tier does affect speed. C doesn't address either latency source, and D misstates what extended thinking does — it slows any model down rather than granting a higher tier's quality.
Q6: C. A recurring task with stable reference material and standing style rules is the defining Project use case: instructions carry the tone, knowledge files carry the documents, and every new chat starts pre-briefed. One endless conversation (A) degrades as the context fills. Asking Claude to "remember" (B) fails because nothing persists between separate conversations unless deliberately carried over, and artifacts (D) are deliverables Claude generates, not a knowledge store.
Q7: B. Project knowledge counts against the context window, so dumping an entire document repository crowds out the custom instructions and dilutes attention on any single file — the fix is to curate. A and D invent behaviors Projects don't have, and C is the wrong-layer "bigger model" reflex: the root cause is an overloaded context, which a tier upgrade doesn't cure.
Q8: D. Multi-source investigation of recent, external developments is the defining research-mode case. Plain chat (A) runs into the knowledge cutoff and invites confident interpolation about events the model never saw. Upgrading to Opus (B) is the misconception that capability equals freshness — every tier shares the cutoff limitation. A Project (C) grounds Claude in materials you upload; naming competitors in instructions fetches nothing new.
Q9: A. Stable internal reference material that every deliverable must follow belongs in Project knowledge; gathering current information across external sources is research mode's job — each need maps to a different feature. B inverts the mapping and would freeze the "fresh" scan at last month's snapshot. C leaves the industry-update need unmet since Project files can't surface new external developments, and D wastes effort re-uploading stable docs while misapplying an investigation tool to internal material.
Q10: B. Reading a live web dashboard and filling a form in another tab is Claude in Chrome's exact design brief: operating on live web apps by navigating, extracting data, and filling forms across tabs. Screenshot-pasting (A) gets Claude describing the page, not acting on it — the human still does the data entry. Claude for Microsoft 365 (C) targets the Outlook/Word/Excel ecosystem, not arbitrary browser apps, and a URL in Project knowledge (D) is inert reference text, not browser access.
Q11: D. The deliverable is edited files and organized folders, which is the chat-vs-Cowork dividing line: chat ends with text you copy out, a Cowork session ends with the work actually done, reviewed on the delegate-and-review model. A chat (A) and an artifact (B) both stop at a proposal the manager must execute manually across 80 files. Research mode (C) is for multi-source external investigation, not file operations.
Q12: C. A custom integration with a proprietary system — custom code, an API connection, or an MCP server — is developer work, and the hallmark associate move is to scope the need and hand it off. Attempting the build from the chat surface (A) exceeds what the productivity surface can do. Manual nightly pasting (B) abandons the automation goal, and D encodes the misconception that a more capable model gains system access — no tier can reach internal systems without an integration.