Home › Study guides › CCAO-F › Domain 4 › Lesson 4.2

CCAO-F · Domain 4 · 16% of the exam · Lesson 4.2 · 19 min read

Research, planning and process optimisation with Claude

How to use Claude to research options with sources, map a process to find its bottlenecks, and plan a measured pilot, while people make the decisions.

Written against objective 4.2 of the official CCAO-F exam guide (Version 1.0, effective July 2026). An independent resource, not affiliated with Anthropic; the practice questions are written from scratch.

4.2.1 Why a big goal gets you a list of generic tips

You run operations for a group of six neighbourhood restaurants. At the quarterly meeting, the owners point at one cost that keeps rising: food that is bought, prepped and then thrown away. They will fund a 90-day pilot at two of the six sites and want a plan by the end of the month. That evening you ask Claude: "How can our restaurants reduce food waste?"

The answer is tidy and confident: track waste daily, use the oldest stock first, prep in smaller batches, trim the menu. It adds that restaurants which measure their waste typically cut it by half. None of this is wrong, and none of it tells you what to do on Monday. What has worked for businesses like yours? Where in your kitchens is the waste made? And where did "by half" come from?

The request bundled three jobs, and Claude had the inputs for none of them. Research asks what others have done and how well it worked, so it runs on sources. Process optimisation asks where your own process loses food, time or money, so it runs on a picture of today's work. Planning turns a chosen change into dated steps with owners, risks and a measure. Claude can prepare all three, given the right input for each. The call itself stays with the owners and head chefs, who answer for the money, the staff and the plates.

4.2.2 Research that ends in options, not an answer

A research report with citations looks finished. But research that will shape a spending decision is a small workflow of its own. Scope the question, choose where the evidence comes from and ask for a source and date for every claim; then verify the ones that matter, and only then turn the findings into options.

Scoping turns "how do we reduce food waste" into a question evidence can answer. Which businesses count, and which waste: prep and spoilage, or what diners leave on the plate? How recent, and what counts as a result: a measured change, or a line in a sales brochure? Anthropic's guidance on research prompts agrees: define what a successful answer looks like, and include constraints such as budget or timeline. Each source route then suits a different kind of evidence.

Memorise the middle column; where each switch lives in the app will change. Research mode, on paid plans, runs many searches that build on each other and returns a cited report after a few minutes. In the brief below, look at the scope line, which says what counts as proof, and the last line, which asks for options, not a verdict.

Context: we run six neighbourhood restaurants with one shared supplier and a head chef at each site. The owners will fund a 90-day waste-reduction pilot at two sites.

Research question: what have restaurant groups of 3 to 20 sites done to cut kitchen food waste (prep waste and spoilage, not plate waste), and what results did they measure?

Scope: sources from the last five years. Count a result only if a source reports a measured before-and-after change. Label case studies from companies selling a waste product as vendor claims. Where you can, confirm each result in a second, independent source.

For every finding, give the source and its date, and say whether the result was measured or only claimed. Keep what the sources report apart from your own interpretation.

Output: a table of 4 to 6 options showing what each involves, strength of evidence, rough cost and effort for a six-site group, and the main risk. Do not pick one; we will decide after checking the sources. If the evidence for an option is thin, say so.

When the report comes back, open the sources behind any option you might choose. Check that each says what the summary says, is recent, and describes a business like yours: a hotel's banqueting kitchen says little about a sixty-seat bistro. Anthropic's help page on web search agrees: cross-reference cited sources, and use authoritative sources for critical decisions. Trace the "by half" figure too; if no source you trust reports it, it stays out of the owners' pack.

Think of a surveyor's report on a house you want to buy: it tells you about the damp and the roof, with evidence, but it doesn't buy the house. Research works the same way: the checked options go to the owners, who fund and run the change and choose between them.

4.2.3 Map the process before you fix it

The research leaves you with five options that worked somewhere. Which one fits depends on a question no outside source can answer: where in your own kitchens is food thrown away, and why? Choosing a fix before you know is like a doctor prescribing before examining the patient: the medicine may be sound, but whether it treats your illness is luck.

So process optimisation starts with a process map: every step in order, who does it, how long it takes, where it hurts, and each handoff, where work passes from one person to the next. Claude is good at turning rough notes into a structured map and at finding patterns in records you attach. Ask it to mark three kinds of loss: waiting, where work or information sits idle; rework, where the same thing is done twice; and the waste itself. The prompt's last line holds back fixes and asks what Claude could not work out, so gaps show up as questions.

Attached are my notes on how ordering and prep work each week at our six restaurants, our standard prep sheet, and eight weeks of nightly waste logs.

Build a table of the current process, one row per step, from the Sunday stock count to the nightly waste log. Columns: step, who does it, when, time taken, what it hands to whom, pain points.

Mark every step where food is thrown away, where work waits for someone, and where the same information is entered twice.

From the waste logs, total the weight thrown away by reason and by step, and show the calculation so I can check it.

Do not suggest improvements yet. List every step or number you could not work out from this material.

Here is what the map shows. On Monday the head chef retypes Sunday's paper stock count and orders up to fixed par levels, standing quantities set last spring. The kitchen preps the same list every morning, whatever the bookings say. Claude's totals put about 60 percent of the logged waste at the end of the night: prepped food nobody ordered. That makes daily prep the bottleneck, the step where most of the loss piles up; fix it and the whole result moves. And the logs are typed up only at month end, so a bad Tuesday reaches you weeks later.

The map is an intermediate finding, and everything after it will stand on it, so check it before it shapes a recommendation. Walk the head chefs through the map: at two sites, it turns out, the sous chef orders from a spreadsheet of her own. Spot-check one week's totals against the paper sheets. Note what the data cannot show: two sites skipped logging on busy Saturdays, so weekend waste is under-counted. The people who do the work know where a map is wrong.

4.2.4 Aim each fix at a bottleneck, then let the owner decide

With a checked map, you can lift the "no improvements yet" rule. Ask Claude for fixes aimed at the steps where the map found loss, with a first-pass label on each. Don't judge a fix by how much of the work Claude takes over; the aim is less waste, and the best fix for the retyped count needs no Claude at all.

The labels are quick to state. An automated step runs without a person doing it each time. In an augmented step, Claude prepares or drafts, and a person judges and signs off. A human-led step is done and decided by people, with Claude's analysis as input at most. For a first pass, ask what an error would cost, how quickly someone could check the result, and whether the change touches staff, policy, money or customers. Linking two systems, such as bookings flowing into the prep sheet, is a build for Claude Architects and Developers.

The human-led column holds the hardest questions. Suppose Claude's analysis, which you have checked, shows that starting prep two hours later would cut end-of-night waste. Correct numbers do not make the change approved: it alters people's working hours and could leave a busy lunch short of food. A recommendation that changes staffing, policy or customer experience goes to the accountable process owner, here the owners and site managers, who approve or reject it. Sound analysis earns a hearing, not the right to happen.

4.2.5 Plan the pilot and decide how you'll measure it

The owners approve one change to pilot, the one aimed at the bottleneck: prep quantities drafted each morning from the day's bookings, at two sites. Just one, so that if the waste figure moves, you know what moved it. Give Claude the goal, the checked map, the change, the dates and the people. Ask for milestones, dependencies (what must happen before something else can start), named owners, risks and assumptions, the things the plan takes as true without checking.

Then ask Claude to attack the plan. Asking whether the plan is good invites reassurance; a stress test asks for specific ways it could fail and for what is missing. The second line below is a pre-mortem: it asks Claude to imagine the pilot has already failed and explain why.

Here is our 90-day waste pilot plan for two restaurants. Stress-test it before we start.

Imagine it is day 90 and the pilot has failed. Give the five most likely reasons, starting with the one that would do most damage.

List any step, owner or dependency the plan is missing, and any assumption it makes without evidence.

For each risk, name one early warning sign we could watch in the first four weeks.

Point out anything that would make the two pilot sites hard to compare with the other four.

Claude's list is specific. Cooks might cut prep too far and run out of a dish on a busy Friday. The day's bookings must reach each head chef before prep starts, and no one owns that step. A menu change in week six would muddy the comparison. And the plan assumes bookings predict demand, though one pilot site serves mostly walk-ins. You fix some and accept others, but each is now a decision rather than a surprise.

The last piece is the easiest to skip: decide how you will know the pilot worked BEFORE anything changes. Step on the scales only at the end of a diet and you have a number with nothing to compare it to. Here the measure is kilograms of food waste per 100 covers (a cover is one diner served). The baseline, the "before" figure, comes from two weeks of weighing at all six sites, since the old logs under-count weekends. The four sites that keep the old way are the comparison.

Add a guardrail, a second number that must not get worse, such as how often a dish runs out. Then agree a decision rule with the owners: extend to all six sites if waste per 100 covers falls by a fifth or more, with no rise in dishes running out. Without these, any result can be read as success.

4.2.6 The exam traps

Every trap in this objective skips a check or a person.

• ✗ Acting on a research report because it is detailed and cited. ✓ Check the key sources for date and fit, and present options. A citation says where to look, not that the claim is right.
• ✗ Asking Claude for improvement ideas before mapping the current process. ✓ Map the process first, so the fix targets the real bottleneck.
• ✗ Letting the plan build on an unchecked map or unverified numbers. ✓ Check them with the people who do the work and against the data first.
• ✗ Judging an improvement by how much work Claude takes over. ✓ Aim each fix at a bottleneck and label it by the cost of an error, how easily it can be checked and who it affects. Some of the best fixes need no Claude.
• ✗ Rolling out a staffing or customer-facing change because the numbers check out. ✓ Send it to the accountable process owner. Correct numbers are not a decision.
• ✗ Starting the change and deciding afterwards how to judge it. ✓ Set the measure, baseline and decision rule first; the "before" cannot be collected later.

4.2.7 Put it together: take one improvement from question to pilot

You now have the whole path. Research ends in options, a checked map shows the bottleneck, each fix gets a first-pass label, and the pilot plan carries a stress test and a measure set before day one. People check each stage and make the call. Now run the method on a process of your own, then take the map away.

The rest of Domain 4 carries the improvement forward. Designing and prototyping it (4.3) might produce a prep-quantity template the head chefs can try. Fitting the new routine into each kitchen's daily work is integration (4.4), and explaining to the owners what Claude did and where its limits lie is 4.5.

Key takeaways

• ✓ Research, process optimisation and planning are separate jobs with separate inputs; Claude prepares each one, and people decide.
• ✓ Research for a decision means scoping the question, choosing a source route, getting a source and date per claim, verifying, then synthesising into options.
• ✓ Map the current process, with owners, timings, handoffs, waiting and rework, to find the bottleneck before proposing any fix.
• ✓ Check intermediate findings with the people who do the work and against the data before recommendations depend on them.
• ✓ Aim each fix at a bottleneck with a first-pass label (automate, augment or human-led), and get the process owner's approval for changes to staffing, policy or customers.
• ✓ Turn the goal into a plan with milestones, dependencies, owners, risks and assumptions, and ask Claude to stress-test it rather than approve it.
• ✓ Set the measure, baseline and decision rule before you change anything.

Check your understanding

4 questions written for this lesson, then one from the CCAO-F question bank on the same topic. Every answer option is explained, including the ones you did not pick. Nothing is stored.

Sources

• Use research on Claude
• When should I use web search, extended thinking, and research?
• Enable and use web search
• Research for deep dives (Claude 101)
• Prompting best practices
• The CCAO-F exam guide, summarised