# AI governance: what it is and how to set it up in 90 days

> What AI governance means in practice, who should own it, and a 90-day plan to move from no rules to a working system with owners, approved tools and evidence.

Language: en · Category: governance · Author: Nicolas Corod (Founder and chief executive officer) · Published: 2026-11-19

## What is AI governance?

AI governance is the set of decision rights, rules and controls that decides what a company's AI may do, who answers for it and how anyone can check what it did. In practice it comes down to three questions: which AI is in use, who owns each use, and what evidence exists when something goes wrong.

Most definitions stop at "policies, processes and controls". Those words are accurate and they do not help a leadership team decide what to do on Monday. Without structure, governance tends to default to "no", because "no" is the safe answer when nobody knows who decides. The result is the opposite of control: in the Qualinsight study of 1,030 professionals in French-speaking Switzerland, [cited by PME in July 2026](https://www.pme.ch/strategie/ia-au-travail-72-des-romands-lutilisent-souvent-en-secret/rl801nw), 49% of respondents said their company had set no clear rules on AI.

## Why does AI governance matter for a mid-sized company?

AI governance matters because it is what lets a company scale AI beyond a few enthusiasts without losing track of data, decisions and accountability. Regulation adds a deadline: the [EU AI Act](https://eur-lex.europa.eu/eli/reg/2024/1689/oj) has required providers and deployers to ensure a sufficient level of AI literacy among their staff since 2 February 2025, and Swiss companies selling into the European Union are concerned too.

Practitioners in regulated sectors describe governance as an accelerator rather than a brake. At the AI & The Future of Financial Services conference held by EU Business School on 16 November 2026, Jean-Noël Ardouin, partner in financial services consulting at EY, was asked whether strong governance is a competitive advantage: "10 times yes. [...] governance is still not here to stop using AI, is just to make sure that the small agent you've implemented can be scaled in a safe manner."

The [overview of enterprise AI governance](/en/blog/enterprise-ai-governance/) covers the regulatory landscape in more depth. This article focuses on getting it in place.

## Who should own AI governance?

Leadership sponsors AI governance, a small cross-functional group runs it, and each AI use has a named business owner. IT, legal and compliance set the framework; the person who knows the job owns the use and answers for it. Ownership is the piece most frameworks leave vague.

Katerina Mita Moulin, founder of Faros Compliance, named it the biggest gap at the same November 2026 conference: "I would have to say ownership, ownership being the biggest gap. So AI cuts through the 3 lines of defense [...] So the answer on who controls is not always compliance. Right? And it cannot be AI."

| Role | Responsibility |
| --- | --- |
| Executive sponsor | Sets the direction, arbitrates priorities, reports to the board |
| Governance group (IT, legal, data protection, a business representative) | Writes the rules, approves new uses, reviews incidents |
| Business owner of each use | Knows the work, answers for results, decides on change requests |
| Employees | Use approved tools, declare new needs, flag problems |

## How do you set up AI governance in 90 days?

Ninety days is enough to move from no rules to a working system, provided you start from what already exists rather than from a blank policy. Split the period into three phases of thirty days: see, decide, run. Each phase ends with something leadership can check.

![The 90-day plan in four stages: see what exists (days 1 to 30), decide the rules (days 31 to 60), run on real work (days 61 to 90) and a first review at day 90](https://cdn.cosmicjs.com/e2fd11c0-e501-11f1-8fed-2b1155636572-ai-governance-90-days-plan.gif)

### Days 1 to 30: see what is already there

1. **Name the sponsor and the governance group.** Five people at most, with one decision meeting every two weeks.
2. **Inventory current AI use.** Approved tools, personal accounts, browser extensions and agents connected to company systems. Announce an amnesty so people declare uses honestly.
3. **Classify by data, not by tool.** Public, internal, confidential, personal. A summary of a press release is not the same risk as a payroll file.

### Days 31 to 60: decide the rules

4. **Publish a first policy.** Allowed tools, forbidden data, who to ask. An [AI policy for employees](/en/blog/ai-policy-for-employees/) can be two pages.
5. **Adopt a framework template** rather than writing one from scratch. The [AI governance framework template](/en/blog/ai-governance-framework-template/) lists the sections a European company needs.
6. **Name an owner for each of the first five uses**, and agree on one measure per use before anything changes: time, errors, deadlines or quality.

### Days 61 to 90: run it on real work

7. **Move two or three uses into an approved setup** with access rights and an execution trace.
8. **Train the people concerned**, using the uses you approved rather than generic AI courses. This also covers the literacy obligation.
9. **Hold the first review.** What changed, what was used, the gap to the baseline, the requests waiting. Then set the rhythm for the next quarter.

## What evidence should AI governance produce?

Governance that leaves no evidence is a policy on paper. By day 90, you should be able to show an inventory of AI uses with an owner for each, the approved tools and their data limits, a log of who ran what and when and a review record. Auditors, clients and regulators ask for the same four things.

The log is the hardest part, because general assistants keep their history per user, not per business process. The guide to [the AI audit trail compliance teams ask for](/en/blog/ai-audit-trail-execution-logs/) explains what such a log should contain.

The Swiss Software Industry Survey 2026 by the University of Bern and Swico ([report, October 2026](https://www.swico.ch/media/filer_public/3d/57/3d57aa05-9a2c-4d8e-955d-0332241a0655/report_2026_final.pdf)) shows the pattern among 104 Swiss software companies: usage guidelines and dedicated competence teams are the most common governance measures, while control of AI budgets and success criteria for AI projects are the least common. Governance in practice is still under construction.

## Where does skilder fit in AI governance?

skilder covers the run phase: each approved use becomes a role, with its skills, its scoped MCP tools from an internal, curated registry and its access rights, configured from a blueprint and used from Claude, ChatGPT, Copilot or any MCP-compatible agent. Each role has an owner, every change is versioned, each run leaves an execution trace attributed to the user and the audit log can be exported. Guardrails that check each tool call are in beta.

An AI role is never finished: it is authored, tested, deployed, controlled, monitored and improved, version after version. That cycle is what keeps governance alive after day 90. To plan your own first ninety days, a [free 30-minute exchange with an expert](https://koalendar.com/e/meet-with-nicolas-9?utm_source=skilder.ai&utm_medium=blog_ai_governance_90_days) helps structure AI adoption from where you are today.

**Takeaway:** give every AI use an owner who knows the job before you write the long policy, and let the evidence of the first ninety days shape the next version.

To map your first ninety days, [book a free 30-minute exchange with an expert](https://koalendar.com/e/meet-with-nicolas-9?utm_source=skilder.ai&utm_medium=blog_ai_governance_90_days).
