Should I hire someone to implement Claude, or do it ourselves?
Most firms should do the first round themselves, and it takes about a week of part-time effort: pick one recurring written deliverable, write the prompt against your own documents, put it in a shared project, and have one person own it. Outside help earns its money in three situations, which are a rollout that already stalled once, a regulated workflow where a wrong answer creates real liability, and the case where the work lives in a system of record that no connector reaches. Ask anyone you are evaluating for a fixed-scope first engagement with a named deliverable rather than a monthly retainer, because a retainer before anything ships transfers all the risk to you. If a vendor leads with headcount reduction or a percentage saved, they have not seen your operation and are quoting a brochure.
Do this first, because it is free and it answers the question
Before you price anyone, run one workflow yourself. Not a pilot, not a committee, one deliverable that somebody on your team produces every week and dislikes producing.
The reason to start here is not thrift. It is that you cannot brief a vendor properly until you have felt where the work actually resists. People who have run one workflow ask completely different questions of a consultant than people who have read about AI, and they are much harder to sell nonsense to.
If that week goes well, you may find you need nothing else for a while. That is a real outcome and it happens often enough that any honest person in this market will tell you to try it.
- Pick one recurring written deliverable: the weekly client update, the intake summary, the monthly report nobody enjoys
- Write the prompt against three real examples of your own finished work, not a generic template
- Put it in a shared project so the next person inherits it instead of rebuilding it
- Name one person who owns it, because a workflow owned by everyone is owned by nobody
- Give it thirty days and judge it on one observable thing: does that deliverable now start from a draft every time
The three situations where help is genuinely worth paying for
The first is a rollout that already stalled. Seats were bought, an announcement went out, and usage is close to zero. The second attempt has to be different from the first, and the thing that makes it different is usually an outside constraint on scope rather than more enthusiasm. This is the most common reason people call.
The second is a regulated workflow where a wrong answer creates liability rather than embarrassment. Treatment write-ups, coverage questions, employment screening, anything a licensed professional signs. The work is not the prompt, it is the review step, the record of who checked what, and knowing which parts must never be automated at all. Getting that wrong is expensive in a way that a bad marketing email is not.
The third is the system-of-record problem. If the work you want to improve lives in a practice management system, an EHR, legal case management, a field service platform, or a mid-market ERP, the ready-made connectors do not reach it. Someone has to decide what gets exported, what stays put, and whether the thing is worth building at all. Often the honest answer is that it is not, and that answer is worth paying for too.
What good help is shaped like
Ask for a fixed-scope first engagement with a named deliverable and a date. Not a discovery retainer, not a monthly fee that starts before anything ships. A retainer signed before the first piece of work transfers all the risk to you, and it is the structure most likely to end with a slide deck and no change to anybody's Tuesday.
Expect the first engagement to produce something your team uses without the consultant in the room. A prompt library nobody opens is not a deliverable. A recurring report that now starts from a draft is.
Expect to be told no. Any operation has workflows where this genuinely does not help, and someone who finds none of them either has not looked or is not telling you.
- Fixed scope, named deliverable, a date, and a price before work starts
- Your team operates it at the end, not the vendor
- Written down where it lives after they leave, in your systems, not theirs
- A named list of what they recommended against, and why
Questions that separate operators from resellers
These are worth asking verbatim. The answers are quick to give if someone has actually done the work, and difficult to fake.
- Which of our workflows would you tell us not to automate, and why
- What does the review step look like when the output is wrong, and who is accountable for catching it
- Show me something you built that a client still runs without you
- Which plan do we actually need, and what specifically requires it rather than the tier above
- What happens to this if the person who owns it leaves
- What have you seen fail, and what did you change afterwards
Red flags
The clearest signal is a number attached to your business by someone who has not seen it. Headcount reductions, hours saved per week, percentage efficiency gains: these are quoted from a brochure, and they are the reason so many of these projects end in disappointment. Nobody can tell you what this is worth in your operation until they have watched one workflow run.
The next is a proposal that starts with a platform rather than a job. If the first artifact is an architecture diagram and the second is a licensing schedule, the engagement has already drifted away from anything that changes a workday.
The last is reluctance to say what plan you need and why. Plan selection is a small, checkable question with a right answer for your compliance requirements. Vagueness there usually means the vendor is optimising for their margin rather than your bill.
Who wrote this, and what I actually do
I am Travis Johnson. I co-founded AuraPath AI, where we put Claude to work inside real small and mid-sized businesses, and I am an Anthropic Claude Ambassador, one of fifty worldwide. I also co-founded and help run a concierge medical practice, so I am on the receiving end of this work as well as the delivery end.
I built the free tool on this site because most owners I meet have heard about AI a hundred times and still have no idea what to do with it on Monday morning. Every workflow and every answer here is free and has no gate in front of it beyond an email address, and the prompts work whether or not you ever speak to me.
If you want a second opinion on something you are already doing, reply to any email from this site. It comes to me directly and a straight answer costs you nothing. If it turns out you do not need help, I will say that, because the alternative is a client who resents the invoice.
Sources
Policies and product details change. Check the source rather than trusting this page indefinitely.
- Anthropic: getting started with Claude Cowork, including which plans it requires
- Anthropic: what the Team plan includes
- Anthropic: what the Enterprise plan adds
- Anthropic: projects, for keeping a workflow where the next person can inherit it
- Anthropic Usage Policy, including the human review requirement for high-risk uses
- Anthropic Trust Center, for the compliance reports a security reviewer will ask for
Need to send this to someone else?
I will email you this answer with every source linked, so it stands up when it lands in front of IT, legal, or finance. Plus the questions that usually come next. Unsubscribe any time.
Want to know what Claude can actually do in your business?
Four questions, about a minute, and Claude writes three automations for your specific situation with the exact prompts. No account.
Build my plan