AI governance for SMEs: ownership, data and review rules
AI governance sets boundaries for tool use, assigns responsibility and defines reviews according to the risk of each workflow. This guide shows SME owners and operations leads how to set it up with an inventory, three risk tiers, a short policy and a named owner for each workflow.
When tool use is unclear, staff may put confidential data in unmanaged apps or use inconsistent outputs. Managers, for their part, need to know which AI workflows affect customers.
Governance for an SME answers four questions: what staff can use, what data is off limits, when outputs need review and who is responsible for each workflow. The answers should be explicit, and they can be brief.
Start with an inventory of AI use
List where AI is already in use before writing any rule. Ask each team which tools they use, for which tasks, with what data, and whether the output reaches customers. Include the AI features inside software the company already licenses.
Ask without blame, because people leave things off a list they expect to be criticised for. Unmanaged tools tend to appear where the official process is slower than the workaround. Our article on connecting AI to the software your team uses describes how an approved, convenient workflow reduces that.
Classify workflows by risk
Not every AI use needs the same control. Drafting an internal meeting summary is different from generating a customer-facing legal clause or approving a refund. A simple model with low, medium and high risk helps teams move quickly without treating every task as dangerous.
- Low risk: internal drafts, summaries, brainstorming and non-sensitive research.
- Medium risk: customer-facing drafts, operational recommendations and data transformation.
- High risk: regulated content, employment decisions, financial approvals and sensitive personal data.
Attach a review rule to each tier. Low-risk output can go out with spot checks. Customer-facing drafts are approved by a named person until quality is established, and sampled after that. Anything that touches money, contracts or compliance keeps a mandatory review. Our guide to designing human review for AI workflows explains how to set up those checks.
Place each workflow from the inventory in a tier and record the reason in one line. When a workflow fits two tiers, use the higher one.
Take a workflow that drafts replies to customer complaints. The drafts reach customers, so it starts as medium risk. If a reply can also grant a refund, that financial approval moves the workflow to high risk, unless the refund decision is split off and kept with a person.
Write rules staff can remember
A useful AI policy is short enough to be read and specific enough to guide behaviour. It should cover approved tools, confidential data, customer-facing output, fact checking and escalation paths.
- Approved tools: which tools staff may use for work, and how to request a new one.
- Confidential data: the kinds of data that must not be entered, named in the company's own terms, such as client files or salary data.
- Customer-facing output: who approves an AI draft before it is sent.
- Fact checking: the person using the output checks names, figures and references against the source.
- Escalation: who to ask when a case is unclear, and how to report a mistake.
Write the policy in the words staff use, with examples from their own work. Test it on a new colleague, who should be able to say after one reading what may be entered into which tool. Swiss companies also need to account for the nDSG, which our article on AI readiness for Swiss SMEs covers.
Answer requests for new tools quickly. If a request sits unanswered, people go back to the unmanaged tool and the inventory is out of date again.
Assign owners
Every AI workflow needs an owner who can answer whether it is still accurate, useful and safe. Without ownership, prompts decay, data changes and no one notices until output quality becomes a problem.
Two roles are involved. One person, usually in operations, management or IT, owns the policy and the list of approved tools. Each workflow also has a business owner who understands the process, and whose name stands beside it in the inventory.
The business owner looks at a sample of outputs at a frequency that matches the tier: periodically for low-risk workflows, more often for customer-facing or regulated ones. They also check again after a change to the model, the source documents or the connected software, because each of these can change how the workflow behaves.
The owner also decides who may change prompts, sources and settings, and keeps a short record of changes. When output quality drops, that record is the first place to look.
Keep AI governance current
Some signs show that governance has slipped: a workflow whose owner has changed role, a tool in daily use that is missing from the approved list, or a review that is approved within seconds every time. Go through the inventory on a fixed date and look for them.
When a mistake reaches a customer, the escalation path in the policy applies. The person who notices reports it to the workflow owner, who decides whether to pause the workflow, correct the source or tighten the review.
Block Consult's AI readiness and workflow audit examines governance for each candidate workflow, alongside data, system connections, process consistency, people and economics, and lists the ownership issues to resolve before a build. For systems already in use, ongoing AI care includes reviews of usage, costs, sources and access.
Frequently asked questions
- Do SMEs need an AI policy?
- Yes, but it can be short. A practical policy defines approved tools, data boundaries, review rules and escalation points.
- Who should own AI governance in an SME?
- Ownership usually sits with operations, management or IT, but each workflow should also have a business owner who understands the process.
- How often should AI workflows be reviewed?
- Review frequency should match risk. Low-risk workflows can be checked periodically. Customer-facing or regulated workflows need more frequent review.
- How do we stop staff using unapproved AI tools?
- Give teams an approved workflow that is convenient to use. Unmanaged tools tend to appear when the official process is slower than the workaround.
Related reading
Let's build something durable.
Tell us about your project. We reply within two business days.

