Is it safe to let AI software into your shop system?

It can be, when the software gets the access a new advisor would get and nothing more, and when nothing it writes can reach a customer without a person approving it. The Federal Trade Commission gives every business that handles personal information the same rule, which it calls the principle of least privilege. Before you connect any AI tool, find out what it can touch in your system and who at the vendor can see your data.

Key takeaways

• Your repair orders hold customer contact details and VINs, along with your pricing.
• An estimating tool needs drafting access, and never administrator access.
• Nothing should reach a customer until an advisor approves it.
• Give the tool its own login so you can shut it off the same way you would for an employee.

What does AI estimating software touch?

Everything on the repair orders it reads. An estimating tool works from the inspection findings, and those findings sit on a ticket with the customer attached. That means the tool sees:

• Customer contact details. Names and phone numbers, plus an email address on many tickets.
• Vehicle details. Year, make, model, mileage and the VIN.
• The inspection. Every finding the tech wrote, including the work the customer declined.
• Your pricing. Labor rate, parts costs, markups and canned jobs.

Most owners think about the first two and forget the last one. Your costs and your markups are competitive information, and they deserve the same care as the customer data.

How much access should it get?

The least that lets it do the job. The FTC's guide to protecting personal information puts it plainly: "Follow the 'principle of least privilege.' That means each employee should have access only to those resources needed to do their particular job."

The same logic applies to software. The FTC's Start with Security guide says "If vendors and contractors don't have to use consumers' sensitive personal information as part of their services, there's no reason for them to have access to it." An estimating tool has to read the findings and write the estimate. It has no reason to change your labor rates or reach your accounting.

Administrator access is the line to hold. The same guide says "Administrative access, which lets a user make system-wide changes to your system, should be limited to the employees tasked with that job." A tool that drafts estimates is not tasked with changing your system, so it should never have that level of access.

For Mechanica, the access is a login with the permissions you would give a new advisor, and you can set it lower. It drafts on the repair order, and that is the whole footprint.

Should it be able to contact your customers?

No. An estimate is a promise the shop makes, and the shop should decide when it goes out.

Mechanica sends nothing to a customer on its own. Every draft waits in a review queue until an advisor reads it and sends it. That approval step is also what keeps the shop in line with repair law in states that require a written estimate and customer authorization before work, which we covered in can AI write an estimate from the tech's inspection notes.

If a vendor tells you their tool can text or email customers on its own, ask how to switch that off before you connect it. A wrong number sent automatically is harder to walk back than a wrong number caught at the counter.

What about pasting repair orders into ChatGPT?

That is the riskier habit, and it is common. Anything pasted into a chat window leaves the building, with no login in your system to revoke afterward.

OpenAI publishes Data Controls settings that let an account holder decide whether conversations are used to improve its models. That setting helps. Names and VINs still have no place in a personal account, whatever the settings say. Our breakdown of whether ChatGPT can write an auto repair estimate covers what it is good for and how to use it without handing over customer data.

What should you ask a vendor before you connect it?

The FTC's advice for any outside provider is to "investigate the company's data security practices and compare their standards to yours," and then to put them in writing. Its Start with Security guide is blunter: "Security can't be a 'take our word for it' thing."

Six questions cover most of it:

1. What permissions does the login need, and will it run with fewer? The answer should be a regular user, and never an administrator.
2. Does it change anything besides drafting estimates? Anything it can edit, it can edit by mistake.
3. Can it send anything to a customer without a person approving it? The answer should be no.
4. Who at your company can see our repair orders? Named roles are a better answer than "the team."
5. Is our data used to train models, and can we opt out? Get the answer in writing.
6. How do we cut off access, and what happens to our data afterward? Ask what gets deleted and when.

A vendor who answers all six in writing is a vendor you can hold to the answers. A vendor who will not is telling you something too.

How do you shut it off if you need to?

Disable its login. That is the same step you take when an advisor leaves, and it should end the tool's access to your system on the spot.

It only works if the tool has its own user. Sharing an advisor's password with software means you cannot shut off one without the other, and you lose the record of which changes the tool made. The FTC's guidance on passwords applies here as much as anywhere, starting with its first lesson: "Insist on complex and unique passwords."

How does this line up with the FTC's five principles?

The FTC's guide to protecting personal information boils its advice down to five lines: "TAKE STOCK. Know what personal information you have in your files and on your computers. SCALE DOWN. Keep only what you need for your business. LOCK IT. Protect the information that you keep. PITCH IT. Properly dispose of what you no longer need. PLAN AHEAD. Create a plan to respond to security incidents."

Each one maps onto an AI tool without much stretch:

• Take stock. List what the tool can read. Repair orders, the inspection and your pricing belong on that list, and anything else it asks for needs a reason.
• Scale down. Give it the smallest set of permissions that still lets it draft an estimate.
• Lock it. Its own login and a unique password, with administrator access kept with your own people.
• Pitch it. Know what the vendor deletes when you stop using the tool, and get that answer in writing before you start.
• Plan ahead. Decide now who disables the login if something looks wrong, and make sure that person knows how.

None of this is special to AI. It is the same discipline a careful shop already applies to a new employee with a login, extended to software that works inside the same repair orders.

What does a sensible setup look like?

Most of this takes an afternoon, and none of it requires an IT department.

• Create a dedicated user for the tool in your shop system, named so anyone can tell what it is.
• Give it the permissions a new advisor gets, then take away anything estimate drafting does not need.
• Keep administrator access with the people who run the system. The tool never gets it.
• Review what it writes every day for the first couple of weeks, the same way you would check a new hire's estimates.
• Put the vendor's answers in the agreement. Permissions, customer contact, training on your data and deletion on exit.

A tool set up this way can be shut off in a minute and audited at any time. That is the standard the FTC applies to employees, and software working inside your repair orders deserves no looser a standard.

If you want to see what the drafting itself looks like, how an estimate gets priced walks through it line by line. More on putting AI to work at the counter is on the blog.