Setanta

Safe rollout

Half our staff use personal AI accounts for quotes and contracts. How do we roll out AI safely?

If your people already paste quotes and contracts into personal AI accounts, the rollout has started without you. Setanta is an Anthropic Partner that helps US manufacturers and industrial companies of 50 to 600 people roll out Claude safely: business accounts, clear rules on data, approvals for anything that changes price or inventory, and training on real work.

Gary O'Neill, founder and CEO of Setanta

Gary O'Neill, Founder & CEO, Setanta.

The short answer

How do you roll out AI safely across a manufacturing company?

Give people a better option than their personal accounts, then set the rules. Move everyone to a business plan you control, write down which data can go in, decide which actions need a person's approval, and train each team on its own work. Banning AI without an alternative pushes it further out of sight.

Safety in a 250-person manufacturer is mostly practical. Who has an account, what they can connect it to, what it is allowed to change and who checks the output. A short set of rules that people follow beats a long policy nobody reads.

Why it matters

People use AI because it helps. The risk is how.

Your estimators and inside sales reps found AI on their own because their days are full. They paste in an RFQ to save time, or a contract to find the clause that matters. That instinct is right. Doing it in a personal account, with your pricing and your customers' names, is the problem.

The hardest part for the person running operations is that nobody knows what is in use, what has been shared or who checks the output. Shutting it down loses the gain, and letting it run loses control.

Setanta makes your whole company as good as your best person. A safe rollout keeps the help people found, and puts it where you can see it.

How we do it

Controls first, then training on real work.

  1. 01

    Operating Diagnostic

    Free, and the first step

    We look at how AI is being used today, what data is involved, which teams gain most and what your systems can connect to. We tell you plainly where the real risks are and where they are not.

    Done when You know what to fix first and which package fits.

  2. 02

    Launch

    Claude rolled out to the whole team on a business plan in your name, with data rules people can follow, admin controls set up and training on the work each team already does.

    Done when Personal accounts are no longer needed, and the team would push back if you took Claude away.

  3. 03

    Build

    For the jobs where AI touches your systems, we build the workflow so Claude reads and drafts, and a named person approves anything that changes price, inventory, quality or a customer commitment.

    Done when A number you chose has moved, with the approvals working as designed.

  4. 04

    Scale

    The same controls carry across three or four processes in up to three departments, including one customer-facing workflow.

    Done when Someone in-house owns the rules and keeps them current.

  5. 05

    Run

    A retainer for monitoring, fixes and the next improvement once the work is live.

How long each step takes is scoped in the Operating Diagnostic.

What you get

What is in place when we are done.

  • Business accounts in your name, managed by your admins, with Claude licences billed to you by Anthropic.
  • One page of data rules: what can go in, what cannot, and why.
  • Approval steps on anything that changes price, inventory, quality or a customer commitment.
  • Every team trained on its own work, with someone in-house who owns the rules.
  • Everything in your own accounts and under your own vendor contracts.

Where this fits

What we are, and what we are not.

We are an Anthropic Partner, so we are not neutral about the model: we work with Claude. We are neutral about the rest, including whether you need outside help at all. If another tool fits a job better, we will say so.

We are not a compliance auditor or a legal adviser. If your contracts or regulators set specific rules for data, your counsel sets them and we build to them. We have no manufacturing client yet. See what to watch after a Claude rollout for the months after.

Questions

Asked before every first call.

Is it safe to put our pricing, contracts and customer data into AI tools? Do they train on what we type?

It depends on the plan. Consumer accounts follow consumer terms, which is why personal use is the risk. Anthropic does not use business and enterprise traffic to train its models by default. We set everything up in your own accounts, under your own contract, so you control the data.

Would I let an AI agent change my ERP, and how do I keep it from bypassing our controls?

Not on its own. We build it so Claude reads and drafts, and changes to price, inventory, quality or customer commitments wait for a named person. Access follows the same roles your people already have. Claude drafts. A person decides.

Should we pick ChatGPT Enterprise, Claude or Copilot?

Pick the one your people will use on real work, under a business plan you control. We are an Anthropic Partner and we work with Claude, so we are not neutral on that choice. We will tell you plainly when something else fits a job better.

Should we ban personal AI accounts?

Yes, once there is a business account that does the same job better. A ban with no alternative usually moves the use out of sight. Give people the better option first, train them, then close the personal accounts.

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.

See the Operating Diagnostic

AI already in use and nobody owns it? Tell us what you see.

We will say where we would start, or that it is not time yet.

Book a call