Task statement 4.2 · Workflow integration and solution design
4.2 · Research, Planning and Process Optimization
In Wrenfield Housing, the course example, the operations team has agreed the requirements for handling repair requests (4.1), and now prepares the change with Claude: a look at what other providers do, a plan for a pilot on the Millbrook estate, and a map of where today's requests wait. If Claude works from general knowledge instead of Wrenfield's facts, all three look finished and fit nobody.
share
The tempting habit is to ask Claude for the answer in one line, such as "How can we make our repairs process faster?" Claude knows how repairs work in general but nothing about your staff, your times or your budget, so it returns advice that sounds right for any housing provider and fits none in particular. Preparatory work pays off only when Claude has your own facts and limits, and people check what it finds before anyone acts.
KnowledgeClaude as a partner for preparatory work
Preparatory work comes before a change: finding out what others do, planning the change and studying the current process. Claude is a partner in it, not the decision-maker. It is quick at structure: grouping findings, laying out steps, spotting a pattern in a list of times. It cannot know your budget, who is away in March or why a step takes nine days unless you tell it, so it finds gaps only among the facts it has. This rules out acting on Claude's plan or process advice before your sources and staff have checked it. The pattern has four moves, and the last belongs to people:
- Give your facts and limits: the goal, figures, dates and what cannot change.
- Ask for a structure you can check: options side by side, a plan with owners, or a map with times.
- Ask what it misses: gaps, risks, and what may not fit your organisation.
- Check with sources and people: open the citations, and ask the staff who do the work.
Interactive · what Claude finds in a process map, and what the check keeps
Choose what Claude looks for in the map, then what the staff kept.
Waits: in steps 2, 3, 6 and 7 nobody works on the request. They hold 38 of the 41 days.
Repeated step: the surveyor's report from step 4 is typed again in step 5.
Bottleneck: step 3 is the longest wait, because only two surveyors inspect every request.
After the staff check: step 2 goes to the pilot, step 5 goes to IT, and step 6 runs twice a week.
KnowledgeThree kinds of preparatory work
1.4 covered the research task type and Research, a mode in the Claude apps that investigates in depth and returns a cited report. The team asked Research how other housing providers triage repair requests, with sources from 2024 on. It found three approaches: sorting requests by urgency on the first call, photos from tenants before an inspection, and a dedicated damp and mould team. A finding is not yet an option, so the next prompt asked Claude to set the three side by side: what each needs, costs and rests on. The last prompt gave Wrenfield's limits (only two surveyors, and no new IT system unless the IT team builds it) and asked which assumptions do not hold. Photos needed an upload route in the repairs system and a dedicated team needed more surveyors, so sorting on the first call went to a pilot.
Planning turns a goal into steps, each with an owner, a date, its dependencies (the steps that must finish first) and its risks. Claude Academy's examples ask Claude for a pilot plan's milestones, success metrics and potential challenges, then for what you are missing. The goal was a six-week pilot on the Millbrook estate, 9 February to 20 March. The team gave Claude the roles, the dates people are away and the other work planned on Millbrook, then asked: "What does this plan miss? List the dependencies and risks we have not named." Claude flagged three gaps. There was no baseline, the current wait measured before the pilot starts. Boiler servicing on the estate from 3 to 14 March would send rebooking calls to the same repairs line. Nobody owned the call handlers' script. Claude saw the boiler clash only because the prompt listed the other work. The team checked each gap, then added a six-week baseline, extra repairs line cover and an owner.
Process optimization means finding where a process loses time or repeats work, then changing it. Claude needs a process map, a list of every step in order with who does it and how long it takes, as Claude Academy's process template asks. A bottleneck is the step where work piles up because it has less capacity than the steps around it. A duplicated step handles the same information twice, and a wait is time when a request sits and nobody works on it. Claude sees only the map, so its suggestions stay hypotheses until the people who do each step confirm them.
Give your limits; ask for options side by side and what does not hold for you.
Then checkThe sources behind each option kept.
Give roles, dates and other work; ask for owners, dependencies and what the plan misses.
Then checkEach flagged gap, with its owner.
Give every step with its time; ask for waits, duplicated steps and the bottleneck.
Then checkEach suggestion, with the staff who do that step.
SkillWorked example: the process question that got generic advice
The team wanted to know where the 41-day median wait for damp and mould goes. The first prompt asked about a process Claude had never seen:
# before How can we make our repairs process faster? → "Consider an online portal, more staff and a chatbot for reporting." # after Below is how a damp and mould request moves today, with typical days. 1 Call handler logs the report (day 0) 2 Triage queue (6) 3 Officer triages, waits for a surveyor (14) 4 Surveyor's report (2) 5 Officer re-keys the report (1) 6 Manager approves over £500 (9) 7 Contractor books the visit (9) Mark each wait, duplicated step and bottleneck. Suggest changes that need no new IT system, and say which step each change shortens. → "Steps 2, 3, 6 and 7 are waits (38 days). Step 5 repeats step 4. Step 3 is the bottleneck: the longest wait, for a surveyor."
The cause was a missing map. With only a goal, Claude could offer only what speeds up repairs in general, and two of its three ideas needed new IT. The new prompt gives the steps, the times and the limit on IT, so each suggestion names a step its staff can check. Sorting requests on the first call became the pilot, and the call handlers needed a written guide to do it (4.3). The repairs manager kept the £500 rule, which finance sets, but now approves twice a week. Ending the re-keying needs a change to the repairs system, so it went to the IT team's developers.
- Findings become useful once they are options tested against your organisation's size, budget and systems.
- Every plan step needs an owner, a date and its dependencies; Claude spots missing risks only among facts you listed.
- A process map with times shows waits, repeated steps and the bottleneck; the staff decide which changes are real.