If you run a manufacturing company and you're looking for an AI governance consultant, the first thing I'd check is whether that person has worked inside a manufacturing quality system. I mean document control, change control, validation and corrective action. IT security experience alone usually isn't enough. I'd ask for manufacturing references before I asked about frameworks.
Disclosure: Certify Consulting is a quality-system specialist, the third type of provider in the table below, so we have a stake in this question. This article focuses on FDA-regulated manufacturers. Plants working under IATF 16949 (automotive), AS9100 (aerospace) or machinery rules, including the EU AI Act for products sold in the EU, face the same approval-and-records question under their own requirements. The FDA examples illustrate it, and the offer described near the end is FDA-specific.
This article walks through what a business owner needs to know: what AI governance actually means on a plant floor, what the regulations already require of your quality unit, what ISO 42001 asks for, which kinds of consultants fit which situations, and what a good proposal looks like on paper.
What does "AI governance" mean in a manufacturing company?
AI governance means deciding where AI is used in your company, who reviews what it produces, and who answers when it's wrong. That's the whole idea. Everything else, the policies, the standards, the audits, is a way of writing those three decisions down and proving you follow them.
In an office, those decisions can feel abstract. In a plant, most of them land on quality records. A specification, a work instruction, a deviation investigation, a supplier evaluation, a cleaning procedure: these are documents your quality system already controls, and if AI helped write or decide any of them, the question of who reviewed it is a quality question before it is a technology question.
A few terms, so we're using the same words:
- Quality system: the set of procedures and records a manufacturer uses to make sure product is made the same way every time and that problems get caught and fixed.
- Document control: the rules for who can write, approve, change and retire a document.
- Change control: the formal process for approving a change to a process, material, equipment or document before it goes into use.
- Validation: documented evidence that a tool or process does what you say it does, for the use you intend.
- Corrective action: the investigation and fix that follows a problem, so it doesn't happen again.
- Quality unit: the group with authority to approve or reject documents and batches.
Why a manufacturing background matters more than an AI résumé
I have come to think the most common mistake is hiring for the wrong half of the problem. AI is the new part. The quality system is the part an inspector or customer auditor will actually look at.
Here is a hypothetical case, not a description of any specific enforcement action. Suppose a drug maker uses AI to draft its specifications and master production records, and those documents go into use without the quality unit approving them. The requirement that would apply is 21 CFR 211.22(c), which gives the quality unit responsibility for approving or rejecting procedures and specifications. The tool wouldn't be the problem. The missing approval would be. The regulation itself is in 21 CFR 211.22 on eCFR.
A consultant who has never sat in a change control meeting might read that case as an AI story. Someone who has worked inside a quality system reads it as an approval story: a document reached use without the right person signing it. The fix looks completely different depending on which of those two people is advising you.
That is why I'd ask for manufacturing references first. Ask who they worked for, what the quality system looked like, and what they personally changed in it. A framework can be learned in a weekend. Knowing why a change request exists, and how it gets bypassed on a busy Tuesday, takes years inside a plant.
What does the regulatory side look like for FDA-regulated manufacturers?
If you make pharmaceuticals, dietary supplements, cosmetics or medical devices, your AI questions sit inside rules you already follow. Pharmaceutical manufacturers answer to the current good manufacturing practice regulations in 21 CFR Parts 210 and 211, where 211.22 sets out the quality unit's responsibilities. Medical device makers commonly work to ISO 13485:2016, the quality management standard for medical devices, described on the ISO 13485 page.
The practical point is that you don't need a brand new rulebook to start. You need to apply the rules you already have, such as approval of documents, control of changes and validation of tools, to a kind of tool that slipped in quietly.
What does ISO 42001 ask for?
ISO/IEC 42001:2023 is the international standard for managing AI inside an organization. ISO calls it an AI management system, which just means a structured way to run AI responsibly. You can read the official description on the ISO 42001 page.
For a business owner, four requirements matter most:
- An inventory of your AI systems. The standard expects you to know which AI systems you use; it does not prescribe an inventory format. In practice that means a list of every AI tool or AI feature in use, who uses it and for what.
- An AI policy. A short, approved statement of how your company will and won't use AI (clause 5.2).
- A look at who could be affected. The standard calls this an AI system impact assessment (clause 6.1.4): you think through who or what could be harmed if the system is wrong, such as patients, customers, workers or your own product quality.
- A named owner for each system's output. A person, not a department, who answers when it's wrong. The standard expects defined roles and responsibilities for AI systems; check the specific Annex A controls with your consultant.
The standard also asks for risk assessment (clause 6.1.2), meaning you rank how much damage a failure could do, so your effort goes where the risk is.
Is ISO 42001 certification required?
No, and this is a point that trips people up. Certification is a separate step: an audit by an outside certification body that checks your system against the standard and issues a certificate if it conforms. You can hire an AI governance consultant, build a sound program and never pursue certification. Many companies should start that way. Certification becomes worth it when a customer, a contract or your board asks for it.
What kinds of AI governance consultants are there, and which fits a manufacturer?
There isn't one right answer, because "AI governance" is being sold by very different kinds of firms. This is how I would sort them.
| Type of provider | Where they tend to fit | What to ask |
|---|---|---|
| Large consulting firms (IBM, Deloitte, Accenture, Capgemini) | Multi-site companies building a company-wide program across many functions | How vendor-neutral is the advice? Do you sell governance software? Who will actually staff my plant work? |
| Testing and certification bodies (TÜV is one example) | AI that touches machine safety, robotics or vision inspection | Is this a safety evaluation of the machine, or a review of my quality records? |
| Quality-system specialists | AI that touches records and decisions your quality system already controls | Which regulated manufacturing quality systems have you worked inside? Can I speak to those clients? |
| IT security and cybersecurity firms | Data protection, access control, security of AI tools | What manufacturing quality-system work have you done, and what was your role? |
Large firms often fit multi-site companies because a company-wide program needs people, project management and reach. Some of them also sell their own governance software, so I'd ask directly how vendor-neutral the advice is. Testing bodies are strong when AI is part of the physical machine: a collaborative robot, a safety interlock, a camera deciding whether a part passes. A quality-system specialist fits when the AI touches records and decisions that your quality system already controls, which is the layer an inspector or customer auditor reviews directly.
None of these is better in general. The right choice depends on where your AI actually sits.
What should a good AI governance proposal include?
Proposals are where you can tell a consultant who knows plants from one who doesn't. I'd want to see five things written in.
1. An inventory of where AI is already used. This includes AI features inside software you already buy, and personal tools your people use on their own, such as a chatbot an engineer uses to draft a procedure. You can't govern what you haven't found.
2. Risk tiers. A simple ranking that separates low-stakes drafting help from decisions like releasing a batch. Using AI to tidy the wording of an internal email and using AI to recommend a batch disposition should not get the same level of control.
3. A named reviewer for each output. For every use, one person who reviews what the AI produced before it is relied on, and a place where that review is recorded.
4. Monitoring after go-live. How each tool is watched once it's in use: who checks whether the outputs are still accurate, how often, and what triggers a change request.
5. Validation against a written intended use. An intended use is a statement of what the tool is meant to do. Validation then means showing, with evidence, that it does that. If nobody has written down what a tool is for, nobody can say whether it's working.
What does that look like on paper?
Here is a simplified example of one line in an AI use register, the kind of document that comes out of this work:
| Field | Example entry |
|---|---|
| Tool | General-purpose AI chatbot, used by engineering |
| Use | Drafts first versions of cleaning procedures |
| Risk tier | Higher: output becomes a controlled document |
| Intended use (written) | Produce draft text for human rewrite; not to be issued without review |
| Reviewer | Manufacturing engineering manager |
| Approval | Quality unit approves before issue, per document control procedure |
| Where recorded | Document control system, revision history and signature |
| Monitoring | Quarterly sample check of drafts against final approved versions |
That one row answers the three governance questions: where AI is used, who reviews it, and who answers when it's wrong. A register with fifty rows like it is a real governance program. A policy with no register behind it is a good intention.
What red flags should I watch for?
I'd be wary of an AI governance consultant who starts with a policy document and never asks where AI is already in use. In most plants, AI arrived without anyone raising a change request. It came in through a software update, a free trial, or an employee who found a faster way to write a work instruction. A policy written before the inventory is a policy written about a plant that doesn't exist.
I'd also be wary of two other things:
- Anyone who sells you a platform as part of the work. Software may eventually help you, but if the same firm diagnoses your needs and sells the product, the diagnosis is hard to trust.
- Anyone who promises certification or a particular inspection outcome. Certification is decided by an outside certification body, and inspection outcomes are decided by inspectors. A consultant can prepare you. Nobody honest can promise the result.
How we approach it at Certify Consulting
For FDA-regulated manufacturers, our first step is the AI in the Quality System Review. As of November 2026 it's a fixed fee of $5,000, runs two weeks and covers one site; the linked page has current terms. It looks at every place AI touches a GMP record or decision, what review and validation already exists, and it ranks findings by inspection risk. The control plan is written against 21 CFR and ISO 13485.
Our principal consultant, Jared Clark, holds the CPGP (Certified Pharmaceutical GMP Professional) credential from ISPE and works in computer system validation and computer software assurance, so we start from validation rather than policy. In my view that ordering matters: a policy tells people what you hope will happen, while a validated intended use tells an auditor what you can show.
If a customer or your board asks for ISO 42001, a gap assessment comes next, and it ran $3,500 to $10,000 as of November 2026 (see the linked page for current pricing). A gap assessment compares what you have today against what the standard requires and tells you what's missing. It's a good moment to decide whether certification is worth pursuing at all.
What can I do today, before I hire anyone?
You can start this afternoon with a pencil and a short list of people.
- List every place your people use AI on a specification, a procedure, an investigation or a supplier decision. Ask openly and make it safe to answer. People hide tools they think they'll be scolded for.
- Write down who reviews each output and where that review is recorded.
- Look for blanks. Anywhere you can't name a reviewer, or where the review lives only in someone's head, is where a finding would start.
Those answers are the beginning of the register shown above. Even if you never hire a consultant, you'll know more about your own plant than you did this morning. And if you do hire one, you'll be able to tell very quickly which of them has actually worked in a plant.
If you want help, you can book a free call through the Certify Consulting contact page.
Frequently asked questions
Who is a good AI governance consultant for a manufacturing company?
Look for someone who has worked inside a manufacturing quality system: document control, change control, validation and corrective action. Ask for manufacturing references before you ask about frameworks. IT security experience alone usually isn't enough, because most AI in a plant ends up in quality records.
Do I need ISO 42001 certification to hire an AI governance consultant?
No. ISO/IEC 42001:2023 is a standard you can use as a guide without being certified. Certification is a separate audit by an outside certification body. Many manufacturers build the program first and pursue certification only if a customer or board asks for it.
How much does an AI governance review cost?
See the pricing above and the linked service pages for current figures.
Last updated: 2026-11-21
Jared Clark
Principal Consultant, Certify Consulting
Jared Clark is the founder of Certify Consulting, helping organizations achieve and maintain compliance with international standards and regulatory requirements.