One of the biggest concerns I hear when we talk about AI governance in government is:
“Won’t all these controls slow us down?”
And I get it.
Nobody wants to create a giant approval maze where every AI idea needs six committees, three forms, two lawyers and a small miracle before anyone is allowed to test it.
But a recent McKinsey & Company article on AI adoption in government makes a really important point:
Strong controls can actually help public-sector organizations move faster.
Not because governance disappears.
Because governance is built in from the start.
And that changes everything.
The Problem With Bolting Governance On at the End
McKinsey describes a pattern that will feel very familiar to many public-sector organizations.
A team starts building something.
The technology looks promising.
The pilot works.
Everyone gets excited.
Then, right near the end, somebody asks:
What data is this using?
Who approved that?
What happens if the AI gets it wrong?
Can this decision be challenged?
Where are the logs?
Who owns the risk?
And suddenly the whole thing stops.
That is not really governance.
That is cleanup.
McKinsey’s argument is that controls should be designed into the AI use case from the beginning, carried through development, launch and ongoing monitoring.
That makes a lot of sense for local government.
Decide the Rules Before the Robot Arrives
Before building an AI workflow, municipalities should already be asking:
- Should AI even be used for this task?
- What information can it access?
- Which sources are approved?
- What is it allowed to do?
- When must it stop and escalate to a human?
- How will we test it?
- Who owns the outcome?
- What happens if something goes wrong?
Those questions are not there to kill the project.
They are there to make sure the team can keep moving.
Because once those rules are clear, developers, staff, privacy teams, IT and leadership are not constantly renegotiating the same questions.
Human Oversight Should Match the Consequence
This was probably one of my favourite parts of the article.
McKinsey argues that human oversight should be based on the consequence of the action, not simply on the fact that AI was involved.
Think about that in local government.
AI checking grammar in a draft email?
Probably not something that needs executive approval.
AI helping summarize a document?
A human should verify it, but we do not need to create a massive governance process around every comma.
Now compare that with:
A licence being revoked.
A benefit being denied.
A public safety decision.
A decision that materially affects a resident.
Very different situation.
The higher the consequence, the stronger the human control should be.
That is practical governance.
The Human Still Owns the Decision
The article also reinforces something I talk about constantly.
As the decisions become more complex and the consequences of getting them wrong become more serious, AI autonomy should decrease.
AI might handle simple, repetitive, low-risk work.
AI might recommend something in a medium-risk process while a human approves it.
And for complex, high-stakes decisions?
The human stays in charge.
There it is again.
Who is the human?
Because the AI can assist.
It can analyze.
It can recommend.
It can automate parts of the workflow.
But accountability does not magically move into the software.
Data Still Matters
McKinsey also brings the conversation straight back to data.
AI-enabled workflows are only as trustworthy as the information they use.
That means local governments still need to know:
Is the information current?
Is it legally permitted to be used?
Is personal information being handled appropriately?
Are the sources approved?
Is the AI pulling from a clean knowledge base or from the organizational equivalent of a junk drawer?
Because feeding AI a messy shared drive and hoping for excellent decisions is… optimistic. 😂
The article specifically emphasizes using clean, approved information and building privacy protections into AI systems from the beginning.
Stop Measuring Success by the Number of Pilots
This is another point I LOVE.
McKinsey argues that public-sector AI should not be measured by how many pilots were launched.
It should be measured by outcomes.
Did residents experience shorter wait times?
Were there fewer errors?
Did response times improve?
Were answers clearer?
Did staff gain capacity?
Did service quality improve?
That is the measure.
Not:
“We launched 14 AI pilots this year.”
Very cool.
But did anything get better?
That is the question.
AI Adoption Is Also a Change Management Project
The article also points out that technology is only one piece of the investment.
Organizations need training.
Adoption.
Capability building.
New ways of working.
Clear ownership.
And people who understand both the service and the risks.
That matters enormously in local government.
Because giving staff access to an AI tool does not mean they will suddenly redesign how they work.
They need time to learn it.
They need practical examples.
They need guidance.
They need to understand what they are allowed to do.
And most importantly, they need to understand why the new way is actually better than the old one.
Governance Should Help Us Move
The lesson I take from McKinsey’s article is not:
“Government needs more AI committees.”
It is almost the opposite.
We need better-designed governance.
Governance that is built into the workflow.
Governance that matches the level of risk.
Governance that gives staff clear boundaries.
Governance that allows low-risk experimentation.
Governance that keeps humans firmly in control of consequential decisions.
Governance that measures what actually improved.
Because the goal is not to move recklessly fast.
And it is not to move painfully slow.
The goal is to move safely, deliberately and continuously.
That is where the guardrail analogy works so well for me.
A guardrail is not there to stop the car from moving.
It is there so the car can move with less chance of ending up somewhere it really should not be.
Need Help Building Practical AI Governance?
If your municipality is trying to introduce AI without creating either a free-for-all or a giant approval bottleneck, I would love to help.
I work with local governments to build practical AI governance frameworks, policies, staff guidance and implementation processes that keep humans accountable while still allowing teams to actually use the technology.
Because governance should not be the brake.
Done properly, it is the guardrail that lets us move.
Credit to McKinsey & Company, and authors Hrishika Vuppala, Roger Roberts, Tim Fountaine and Tim Ward, for the public-sector research and framing that informed this piece.
