McKinsey's $1.5M AI Contract Shows Everything Wrong With Government Tech

McKinsey scores $1.5 million for three months of AI consulting at Services Australia. Learn why this approach guarantees Australia falls further behind in government AI adoption.

$1.5 million. Three months. "Right building blocks."

Services Australia just hired McKinsey to tell them how to use AI. At $500,000 per month, that's roughly $23,000 per working day. For advice.

Let me tell you what $23,000 per day doesn't buy you: a single line of working code.

The Building Blocks Mythology

"Right building blocks" is consultant speak for "we'll tell you what you already know, but with frameworks."

You want to know the right building blocks for AI? Here they are: data, talent, and permission to fail. That'll be $1.5 million, please.

Except Services Australia already knows this. Every government department knows this. What they don't have is the courage to build without a consultant safety net.

Three Months to State the Obvious

I've seen these engagements. Month one: discovery. Month two: analysis. Month three: recommendations. The output? A roadmap that recommends hiring more consultants to implement the roadmap.

Estonia built their entire digital government for less than Australia spends on PowerPoints about digital government.

While McKinsey consultants interview stakeholders, Singapore's government AI teams ship actual products. While Australia buys frameworks, South Korea builds systems.

The Expertise Extraction Machine

Here's the McKinsey playbook: Interview your smartest people. Package their ideas. Sell them back to you. Charge premium rates for the privilege.

Services Australia has thousands of employees. Some of them deeply understand AI. They know what the organisation needs. But their voices don't carry the weight of a $1.5 million invoice.

External validation addiction is killing government innovation. Departments trust consultants more than their own people. So their best people leave for places that listen.

The Real Cost Mathematics

$1.5 million for three months equals:

• 10 data scientists for a year

• 15 developer salaries

• 20 AI training programmes for existing staff

• One actual working AI prototype with change left over

But prototypes are risky. Reports are safe. So we buy reports.

Taiwan's Digital Ministry built their entire COVID response system with less. It worked. It saved lives. No McKinsey required.

Why This Keeps Happening

Risk management has evolved into risk avoidance. Nobody gets promoted for building something new. Everyone gets promoted for following process.

Hiring McKinsey is process. It's defensible. It's what serious organisations do. Never mind that it doesn't work.

The successful government AI implementations share one trait: internal ownership. The UK's Government Digital Service built capability, not PowerPoints. Denmark's Agency for Digitisation writes code, not contracts.

The Alternative Path

Want to know how to use AI in government? Start using it. Build small things. Learn from failures. Scale what works.

Services Australia could create an AI lab tomorrow. Staff it with their own people. Give them permission to experiment. In three months, they'd have working prototypes instead of PDF recommendations.

But that requires admitting you don't need McKinsey to tell you what to do. And apparently, that admission costs more than $1.5 million.

The Questions Nobody Asks

Before signing another consulting contract, ask:

• What will we be able to do after this that we can't do now?

• Why can't our own people answer these questions?

• How many times have we paid for similar advice?

The answers are uncomfortable. Which is why we keep hiring consultants instead of facing them.

When did we decide building the future required permission from McKinsey?