How consulting firms actually use Claude: four workflows
This page is four workflows you can set up in claude.ai this week: a findings memo built from your discovery call, a scope of work that says plainly what is not included, a client deliverable where every number came from you, and the method in your head written down so someone else can run it. Every prompt is here to copy. Nothing to install, no code, and nothing that asks you to trust a figure Claude produced on its own.
1 of 4 · about 20 min to set up
Turn a discovery call into a findings memo
Today
You get off a ninety minute call with four pages of notes and a strong feeling about what is wrong. Three days later you write it up, and half of what you send back is what you already believed walking in, because the actual quotes have gone cold.
With Claude
You paste the notes the same afternoon and get findings that each carry the line they came from, the contradictions in what the client told you sitting side by side, and the questions you still need answered before anyone prices the work.
Paste this into claude.ai
You are helping me turn a discovery call into a findings memo for a client engagement.
Call notes or transcript: [paste your raw notes or the transcript here]
Who was on the call and their roles: [list names or titles, for example COO, VP of Operations]
What the client asked us to look at: [one or two sentences]
What I already know about this client from prior work: [paste it, or write SEE PROJECT KNOWLEDGE, or NONE]
Rules you must follow:
1. Use only the material I gave you. If something is not in it, write NOT STATED rather than filling the gap.
2. Keep three categories strictly separate: what the client said, what I appear to have inferred, and what is still unknown.
3. Do not add industry benchmarks, market figures, or comparisons to other companies. Not one.
Produce the memo in this order:
1. The situation in five sentences, in language the client would recognize as their own.
2. Findings. For each one give the finding in a single sentence, the exact note or quote it rests on, and how strong that evidence is: stated directly, implied, or thin.
3. Root causes: where two or more findings are probably the same underlying problem, and why you think so.
4. Open questions, ordered by how much the answer would change our recommendation. Mark the three I should ask before writing any proposal.
5. Contradictions: anything the client said that does not square with something else they said, quoted side by side.
6. The words and phrases they used repeatedly that we should mirror back in our writing.
Then stop and ask me up to five questions that would materially improve the memo.
2 of 4 · about 35 min to set up
Scope the work before it becomes an argument
Today
The client says yes in principle, so the scope of work is now due. You copy the last one, change the names, and leave the assumptions thin because tightening them means a harder conversation this week. Month two is when you find out what that cost.
With Claude
You paste the findings and your staffing model and get phases with defensible hour ranges, an assumptions section that names who owns what, a blunt out of scope list, and the change order triggers written down before either side needs them.
Paste this into claude.ai
You are helping me turn an agreed set of findings into a scope of work I can send to a client.
The findings or discovery memo: [paste it here]
What the client said they want as an outcome: [paste their own words if you have them]
How we staff and price work: [paste your roles, rates or blended rate, and typical engagement shapes, or write SEE PROJECT KNOWLEDGE]
Two or three scopes of work we have signed before: [paste them, or write SEE PROJECT KNOWLEDGE]
Hard constraints: [budget ceiling, deadline, client team availability, anything the client has already ruled out]
Do this in order:
1. State the outcome this engagement is being bought for, in one sentence, from the client's point of view. If the findings support more than one plausible outcome, list them, recommend which to scope against, and say why.
2. Break the work into phases. For each phase give the deliverables by name, the client decision that closes the phase, the roles involved, and an hour range you can defend. Where you are guessing at effort, label that line ASSUMPTION.
3. Write the assumptions section. Every hour range above that depends on something the client must provide belongs here, named specifically, with who provides it and by when.
4. Write the out of scope section. Be blunt and concrete. Include the three things a client in this situation most often assumes is included.
5. List the change order triggers: the specific events that would make this scope wrong, and what happens when one of them occurs.
6. Write acceptance criteria for each deliverable, one sentence each, phrased so the client and I can both tell whether it was met.
7. Flag anything in my staffing or rate information that does not line up with the phases you built.
Write it in plain language a non-technical buyer can read in ten minutes. Do not invent a price. Give me the hour ranges and I will apply our rates myself.
3 of 4 · about 30 min to set up
Build the client deliverable from your own analysis
Today
The thinking is done and lives in your head, a spreadsheet, and twenty pages of notes. Turning that into something a board will read takes the weekend, and the section you rewrite four times is the one that has to land.
With Claude
You paste the analysis and your figures and get a storyline where every headline is a conclusion, evidence traced back to what you supplied, the hard questions the room will ask, and a downloadable deck you edit instead of build.
Paste this into claude.ai
You are helping me build a client-ready deliverable from analysis I have already done. Our deck structure and past examples are in the project knowledge.
Client and engagement: [client name or just CLIENT, and the question we were hired to answer]
Audience for this deliverable: [who is in the room, what they can approve, what they currently believe]
My analysis and working notes: [paste them here in whatever state they are in]
The data, pasted exactly as it came out of our own analysis: [paste the tables or figures, or write NONE]
What we are recommending: [one or two sentences, or write UNDECIDED]
Rules you must follow:
1. Every number in this deliverable must come from what I pasted. Never produce a figure I did not supply. Simple arithmetic on my figures is allowed, and when you do it, show the arithmetic.
2. Never add a benchmark, industry average, market size, or comparison company. If a claim needs outside evidence I did not give you, write EVIDENCE NEEDED in brackets where it belongs and move on.
3. Keep what the data shows separate from what we conclude from it.
Produce this in order:
1. The storyline as a numbered list of section headlines, each stating a conclusion rather than a topic, so the headlines read alone as the argument.
2. For each section: the point in one sentence, the two or three pieces of evidence behind it with where each came from, and what I should actually put on the slide or page.
3. The executive summary, one page, written so it holds up when someone reads it on a phone ten minutes before the meeting.
4. The three hardest questions this audience will ask, and how you would answer each.
5. Every place a reasonable reader could disagree with our logic.
When I confirm the storyline, build it as a PowerPoint file I can download and edit.
4 of 4 · about 45 min to set up
Get your method out of your head and written down
Today
You are the only person who can run the diagnostic properly, so every engagement routes back through you. A senior hire watches you do it twice and still asks what happens next, because the real method is a set of judgment calls you have never said out loud.
With Claude
You talk it through once in one messy paragraph, hand over the past deliverables, and get a staged playbook that names the judgment calls, the signals behind each one, and the questions you ask at every stage. It goes in the project knowledge and the next engagement starts from it.
Paste this into claude.ai
You are helping me document how my firm actually does a piece of work, so that someone other than me can run it.
The method I want captured: [name it, for example our operations diagnostic, or our vendor selection process]
Everything I have that shows how it goes: [paste past deliverables, working files, checklists, or emails, or write SEE PROJECT KNOWLEDGE]
How I would explain it out loud to a new hire: [talk it through in one messy paragraph, do not clean it up]
Who will run it next: [their role and experience level]
Where it has gone wrong before: [paste whatever you remember, or write NONE]
Produce a playbook with these parts:
1. What this method is for, and the two or three situations where it is the wrong choice.
2. The stages, in order. For each stage: the inputs needed, what the person actually does, the artifact it produces, and how they know the stage is finished.
3. The judgment calls, which is the part that matters most here. For every decision I appear to make on instinct, write the question being answered, the options, and the signals that point to each option. Where my material does not explain a call I clearly make, ask me about it instead of guessing.
4. The questions we ask clients at each stage, in the words we actually use.
5. The failure modes, drawn from what I told you plus what the stages themselves suggest, each with its early warning sign.
6. A one page checklist the next person can keep open while running the method.
When the draft is done, list every place where you had to guess at my reasoning, and ask me the smallest set of questions that would close those gaps.
What to skip for now
×Anything your MSA, NDA, or client data policy does not clearly allow. Professional services contracts routinely restrict where client material can go, and many now carry specific AI or subprocessor language, so read the agreement and get written approval before you paste a client's documents anywhere. Know what your plan does too: on Free, Pro, and Max accounts your chats are used to improve Anthropic's models only if you turn that on in Privacy Settings, and Incognito chats are never used for it, while Team and Enterprise sit under the Commercial Terms, where inputs and outputs are not used for model training by default. When in doubt, work from anonymized material and keep the client's name out of it.
×Any finding, benchmark, or figure in a client deliverable that Claude produced rather than you. No industry averages, no market sizes, no comparison companies, no percentages it filled in because the sentence wanted one. Paste your own analysis and let Claude write the argument around your numbers. A single invented benchmark that a client's team can disprove ends the engagement and the referral behind it.
×Anything that requires seeing your systems. In the browser Claude cannot open your project tool, time tracker, billing, or the client's data room, so it cannot tell you where a project stands, whether you are over on hours, or what changed in an account this week. It knows only what you put in the chat or the project knowledge, and it does not know what happened after its training data ends unless you paste it in or have it search the web.
Want all 4 workflows for consulting and professional services firms in your inbox?
Every prompt on this page plus the rest, so you still have them on Monday. One email, then four short ones on making it stick. Unsubscribe any time.
Questions people ask before they try any of this
Straight answers with the sources linked, so you can forward them to whoever has to sign off.
Want this worked out for your own consulting and professional services firm?
Answer three quick questions and Claude writes three automations for your specific situation, with the prompts.
How many people work there?
Common questions
Do I need a paid plan to do this?
You can start free. Projects, which is where you keep your method, your scope templates, and your past deliverables so you stop re-explaining your firm every time, are available on all accounts, and free accounts can create up to five of them. Creating downloadable Word, PowerPoint, Excel, and PDF files is available on every plan too, under Settings and then Capabilities. Paid plans add expanded project knowledge capacity through retrieval and higher usage limits, which starts to matter once several people run these workflows every day.
What do I say when a client asks whether AI touched their deliverable?
Answer it straight, because the two things they actually care about are simple: whether their confidential material was protected, and whether a named human is accountable for the conclusions. Put one short paragraph in your scope of work covering which tools you use, what you do and do not put into them, and who reviews every deliverable before it leaves the firm. Check your existing MSAs first, since some already contain AI or subprocessor terms that settle the question for you.
Can I use the same setup for the weekly client status update?
Yes, and it is the fastest of these to adopt. Keep one project per engagement, and each week paste in what shipped, what is blocked with the owner named, what changed in scope, and what is next, then have Claude write the update in your format. Do not ask it for a percent complete or a burn figure, because it cannot see your project tool or your time entries. Those numbers come from you, and Claude writes the narrative around them.
We are eighteen people. How do we avoid ending up with eighteen versions of every prompt?
Treat the prompts like any other firm template. Keep them in one shared document with a single named owner per workflow, and run one project per engagement, named the way your file folders are named, holding the scope, the findings memo, and the deliverables to date. On Team and Enterprise plans you can share a project with specific people or the whole organization, which is where this stops being one principal's private habit and becomes how the firm works.