Buyer's guide
How to choose a Claude implementation partner, and when not to.
Choose the partner who will do the work in your accounts on your real data, agree a measurable finish line before starting, and hand everything over when they leave. Ask who does the work, what you keep and how success is measured. Be wary of any partner who cannot answer those three plainly.
First, whether you need one
A partner is not always the right answer.
Claude is easy to start using. A person can sign up and get useful work done the same afternoon, and many teams get a long way with a few hours of shared practice. If you have a technical person with time and a clear first problem, you may not need outside help for that first step.
A partner earns its place when the work crosses people and systems: a process that touches three tools and four people, data that has to be pulled together before Claude can use it, a team that needs to change how it works and not only try a new tool. It also helps when nobody inside has the time. The honest test is whether you can name an owner and a number. If you can, outside help speeds you up. If you cannot, a partner will be spending your money on the finding out.
The criteria
Seven things worth checking before you sign.
- Who does the work
- Ask for the name of the person who will build it. In many firms the people who sell and the people who deliver are different, and the difference shows up in month two.
- What you keep
- The system should run in your own accounts, under your own vendor contracts, with documentation your team can follow. Ask what happens if you stop paying.
- How success is measured
- A number agreed before the work starts, measured against your own baseline. 'Improved efficiency' is not a number.
- Your real data
- Work built on a tidy sample fails on the first real input. Ask whether they will work on the messy version.
- What they turn down
- A partner who accepts every brief is selling, not advising. Good ones say no to strategy decks as the deliverable and to processes whose owner has not agreed to change.
- Proof you can check
- A named client who will say what happened, in their own words. Be careful with proof that is always anonymous and always spectacular.
- Fit with your size
- A firm built for enterprises will staff a 100 person company like an enterprise. Ask who their usual client is.
Questions to ask
Nine questions for the first call.
- Who, by name, will do the work, and how much of their time is mine?
- What exactly will be running when you finish, and where will it live?
- What number do we agree on, and how do we measure it against today?
- What do you need from my team, and what happens if they cannot give it?
- What will you say no to on this project?
- Which of your clients can I speak to, and what did they report?
- What happens to my data, and which vendors does it pass through?
- What do I own when you leave, and could my team run it without you?
- What would make you tell me this is not the right time?
Red flags
Signs to walk away or ask harder questions.
- The deliverable is a strategy document and a roadmap.
- They cannot name who will do the work.
- Fees are hourly with no cap, or the first phase is undefined.
- The system runs in their accounts, or needs their licence to keep working.
- Every case study is anonymous, and every result is a large percentage.
- They promise to replace a team, or a headcount number is the pitch.
- They resell model capacity at a markup and call it a platform.
- They agree with everything you say on the first call.
The comparison
Partner, in-house or freelancer: what each does well.
| Implementation partner | In-house person or team | Freelancer | |
|---|---|---|---|
| Strongest when | The work crosses several people and systems, and nobody inside has time. | AI is core to the business and will need constant attention for years. | The task is narrow, well defined and short. |
| Speed to first result | Fast, since the method is already worked out. | Slow at first. Hiring and learning come before building. | Fast on a narrow task. |
| What you keep | Depends on the partner. Ask directly. | Everything, including the knowledge in people's heads. | Depends on the person. Documentation varies. |
| Main risk | Dependence, or a strategy deck that goes nowhere. | A single point of failure if the person leaves. | Availability, and no one to hand it to. |
| Change management | Part of the work in good firms. | Easier, since the person is already trusted. | Usually out of scope. |
Where Setanta fits
Where we are a good fit, and where we are not.
Setanta is an Anthropic Partner based in Wilmington, NC. It serves companies of 50 to 500 people in the US Southeast with no in-house AI team and a real operating bottleneck. Our founder does the work. Everything is built in your accounts and your team runs it afterward.
It is a poor fit if you need a large team on call around the clock, if the project is mainly a strategy document, or if the process owner has not agreed anything should change. It is deliberately small, takes a limited number of engagements at once, and has one published client story: THG Insurance, an independent agency in Wilmington, NC. We would rather you check that story and ask THG than take our word.
Questions
Asked before every first call.
How do I choose a Claude implementation partner?
Pick the partner who names the person doing the work, agrees a measurable finish line before starting, builds in your own accounts and hands everything over. Ask how success is measured, what you keep and what they would turn down. Check proof with a named client you can speak to.
Do I need a partner to use Claude?
Not always. Claude is easy to start, and a technical person with time and a clear first problem can get far alone. A partner helps when the work crosses several people and systems, when data must be assembled first, or when your team needs to change how it works.
What red flags should I watch for?
A strategy document as the main deliverable, no named person doing the work, a system that runs in the partner's accounts, only anonymous case studies, and promises to replace a team. Also be wary of a partner who agrees with everything on the first call.
Partner, in-house hire or freelancer?
A partner suits work that crosses people and systems when nobody inside has time. An in-house hire suits AI that is core to the business for years. A freelancer suits a narrow, well-defined, short task. Many companies use a partner first, then bring the skill in-house.
What does Setanta do differently?
Our founder does the work, everything runs in your own accounts, and each package has a finish line agreed up front. Setanta takes no margin on model spend and turns down strategy documents as a deliverable. It is small on purpose.
Is Setanta a good fit for every company?
No. It suits companies of 50 to 500 people with no in-house AI team and a real bottleneck. It is a poor fit if you need a large team on call all day, want mainly a strategy document, or the process owner has not agreed anything should change.
What is the free Operating Diagnostic?
A free call (book a time) that looks at how your team works today. It finds where Claude would move your numbers first and the process worth fixing first, and says which package fits. If it is not the right time, we say so.
Want to test this checklist on us? Book a call.
Ask the nine questions. We will answer each one plainly.