Claude for Consultants: How to Set Up Claude for Client Work

Many people search for "the consultant's guide to Claude". There is no official one. Anthropic publishes guides for some sectors, like financial services and legal, but not for consulting. The guides that show up in search come from independent publishers and tool vendors, and most of them focus on prompts and slides.

I have spent the last year running my own consulting and development work on Claude: client research, proposals, deliverables and meeting cycles. I also train teams on it. This guide covers what I set up, what a firm has to decide first, and where Claude gets consulting work wrong.

Why setup matters more than prompting

Most of a good answer comes from context and setup, not from the question you ask. If Claude's output is not useful, the problem is almost never the prompt. Claude does not know enough about your situation.

Think about onboarding a contractor. You would not hand them a task and walk away. You would give them client background, your standards, the terminology you use and what good looks like. Claude works the same way. The more you invest in setup, the less time you spend correcting outputs.

A consulting-focused guide I have referenced models a typical 35-hour client-facing week like this: 10 hours writing deliverables, 8 hours on research, 5 hours on admin and 5 hours on client communications. That leaves about 4 hours for the strategic thinking the client is actually paying for. With proper setup, practitioners report reclaiming around 5 hours a week. Not by typing faster, but by not starting from scratch every session.

Which Claude features matter for consultants?

Six features cover most consulting work.

Projects hold the background for one piece of work: the client brief, interview notes and past deliverables.

Skills hold your firm's methods and house style. Claude loads them only when they are relevant.[1]

Research runs a series of web searches and returns a report with sources.[2]

Excel and PowerPoint add-ins work inside your files. In Excel, Claude can trace formulas and show which cells an answer comes from.[3]

Cowork takes on a bigger task, splits it into steps and hands back finished files.[4]

Claude Code is mostly for building. Consultants use it to clean data or prototype a small tool for a client.

Features differ by plan and change often, so check Anthropic's help pages before you promise something to your team.

How should a firm set up Claude?

Give each client its own Project. Give the firm one shared set of Skills.

The client Project holds that client's brief, notes and data. Material from one client never sits next to another's. Three active clients means three Projects. Within each one you run everything, from research and drafting to meeting prep, and Claude carries the context across all of it.

The Skills hold what stays the same across clients: your issue-tree method, your market sizing approach, your proposal structure and your review checklist. When a partner improves a Skill, everyone gets the better version the next time they use it. Nobody copies prompts from a shared document anymore.

Start with the two or three methods your firm uses most. Write each one like you would brief a new analyst. I explain how in how to write AI skills for your job.

Keep each Project tight. Projects drift when you fill them with too many documents and conflicting instructions. Use rules files for permanent standards and connectors for live documents instead of uploading everything.

Write a client context document

Before you upload anything to a Project, write a context document. Eight fields, under 500 words:

1. Company name and what they do
2. Target customers
3. Core value proposition
4. Main competitors
5. Current engagement scope
6. Metrics and traction data
7. Stakeholders you deal with
8. Tone and terminology preferences

Put it at the top of every client Project. It anchors proposals, research summaries and client emails to your client's actual situation rather than generic industry patterns. When several people work on the same client, everyone uses the same document and the outputs stay consistent.

Set up CLAUDE.md for permanent rules

If you use Claude Code, CLAUDE.md is the file Claude reads at the start of every session. It holds permanent rules that override casual prompt wording.

A practical CLAUDE.md for a consulting engagement covers:

• Client name, industry and engagement summary
• Terminology, for example "proposal" or "SOW", "client" or "partner"
• Output standards, such as executive summary first and a one-page limit on recommendations
• Hard constraints, such as never invent data and never suggest work outside the engagement scope
• Links to reference documents

There are three levels: a global CLAUDE.md for your personal preferences, a project-level one for each engagement, and CLAUDE.local.md (gitignored) for private working notes. Claude loads all of them together, from the most general to the closest, so your project rules and private notes come last.[5] Keep the levels consistent so they never pull in different directions.

Turn your edits into rules

No setup is right on day one. You will keep rewriting the same things in Claude's output. That repetition is a signal.

When you keep changing the same type of output, Claude is following a pattern you have not written down yet. Notice the friction, turn it into a specific rule and add it to CLAUDE.md or the Project instructions. Not "be more concise", but something testable: "Lead with the recommendation, not the analysis. Clients read the first paragraph and skip to the appendix."

You can ask Claude to help: "Based on how I have been changing your outputs this session, which rules am I applying that are not in your instructions yet?" Review what it proposes and keep the rules that hold. Over a few weeks your edits get smaller, because your standards are written down in a form Claude can follow.

Connect Claude to the tools you already use

Connectors let Claude read from the source instead of you pasting content into a chat. In Claude you add them under Customize > Connectors.[6]

Google Workspace. Claude has native Gmail, Google Calendar and Google Drive connectors, so you sign in with your Google account and skip the local setup.[7] Meeting prep pulls the relevant files, and post-meeting summaries land in the right document.

Notion. Claude reads and writes your workspace, creates project pages and fills databases. I run the whole content pipeline behind aibl.to this way.

Slack. Claude can summarise threads and draft replies.

GitHub and Linear. For product and engineering teams, Claude can review pull requests with the linked specs and draft release notes.

Start with read-only access and one tool, the one you live in most. Do not connect anything that holds sensitive client data until you know exactly what Claude can read and write.

How do consultants use it day to day?

Claude handles the gathering and first drafts. You keep the framing and the judgment.

These are the uses I see working:

• Proposals. Paste messy discovery notes and your proposal structure, and Claude drafts a 4 to 6 page proposal. Customising and finishing it takes me 30 to 60 minutes instead of 4 to 6 hours from a blank page.
• Desk research. Upload the client's last report, a few competitor pages and a market brief, or use Research. A structured overview with competitor comparison and data gaps takes 30 to 40 minutes including my review. Then I check every source.
• Interview synthesis, with a quote behind every theme so you can check it.
• Model reviews in Excel, tracing formulas and testing assumptions.
• Deliverables. I supply the analysis and recommendations, and Claude structures them for the audience. On a 20-page strategy document that saves me 4 to 5 hours, and I keep ownership of the thinking.
• Meetings. A brief before (attendees, likely questions, decisions needed) and a summary after (follow-up email plus internal notes). 15 to 20 minutes per meeting instead of 45 to 60.
• Decks, with the storyline agreed in chat before any slide gets made.
• Monthly client reports in Cowork, when the inputs and structure stay the same.

Skip the framing and you get generic output. Skip the check and one day you send a client something wrong.

One more use is worth stealing: pressure-testing. At Jabreen Academy's hackathon, 14 teams tested every business case against AI playing investors and regulators before they pitched on stage. You can do the same with a recommendation or a proposal. Ask Claude to argue against it as the client's CFO or their regulator, then fix the weak spots before the client finds them.

Is it safe to put client data into Claude?

It can be, on the right plan and within your client contracts.

On Anthropic's business plans (Team, Enterprise and the API), Anthropic says it does not train its models on your inputs or outputs by default.[8] One exception: if someone rates a response, Anthropic may keep that conversation. Many firms switch ratings off.

Personal plans follow different rules. Client work does not belong there.

Your client contract matters just as much. Many contracts and NDAs now say something about AI tools or where data may go. Check before you upload anything, and agree with the client what can go in.

In the EU, the AI Act also asks firms to make sure staff who use AI understand it well enough. That has applied since February 2025.[9]

Where does Claude get consulting work wrong?

Claude goes wrong on tasks just outside what it can do, and it rarely tells you.

In a 2023 Harvard study with 758 BCG consultants using GPT-4, consultants with AI finished 12% more tasks, 25% faster and at 40% higher quality on tasks the AI handled well. On a task designed to sit just outside its abilities, they got the right answer less often than consultants without AI.[10]

Made-up sources are the most expensive version. In 2025, Deloitte Australia refunded part of a government fee after a report turned out to contain invented references and a fake quote.[11] That tool was not Claude, but the lesson holds for every model.

The rule is simple. No source, quote or number reaches a client until a person has opened and read the original.

Claude can draft the work. Your name still goes on it.

Claude or ChatGPT for consulting work?

I use both. They are different tools.

Claude suits long analytical work. It follows detailed instructions closely and holds up well when it has to absorb a 40-page RFP and pull out specific implications. ChatGPT has apps that connect it to other tools, plus image generation. If you are already deep in the OpenAI stack, switching may not be worth it.

For research, proposals, deliverables and meeting cycles, Claude is what I recommend. Not because of benchmarks, but because it is consistent enough to build repeatable workflows on, which is what matters when you bill time.

What are the big firms doing?

They are rolling it out at scale. Deloitte announced Claude access for more than 470,000 people and plans to certify 15,000 professionals.[12]

For a firm of 50 to 500 people, copy the order, not the size. Train people first, set clear data rules, turn your methods into shared Skills, then measure what changed.

How you train matters as much as when. In one of our early technical workshops, a participant told us that for an experienced audience we should reverse the order: show the working result first, then explain the parts. We changed our sessions after that. Consultants care about context windows and connectors once they have seen Claude produce a piece of their own work.

How can we help your team?

If your team has Claude but uses it unevenly, our AI workshops include hands-on Claude training built around your own proposals, interviews and models. We agree the client data rules before anyone uploads a file, and help you set up client Projects and shared Skills.

Want to automate a recurring piece of client work, like a monthly report? AI Workflows builds it with your team and trains someone to own it.