The HK$1 Million AI Mistake: Hiring Before You Know What to Fix

We're going to hire someone internally to handle AI.

We hear this surprisingly often. On the surface, it makes perfect sense. You want more control. You want to protect your company's data. You don't want to depend on an outside provider forever. And if AI is going to become important to the business, why not hire an AI engineer or data analyst and build the capability internally?

There is nothing wrong with that decision. But there is one question you should be able to answer before you post the job advertisement: what exactly are you hiring this person to fix?

• Not "We need AI."
• Not "We need our own model."
• Not "Everyone is doing it."

What actual problem inside the business are you expecting this person to solve? Because a very good engineer working on the wrong problem is still an expensive mistake. And in Hong Kong, it can become a HK$1 million mistake very quickly.

If the business genuinely needs that capability, spend the money. But don't spend close to HK$1 million just to discover what problem you should have been solving in the first place.

The mistake isn't hiring an AI engineer

The mistake is starting with the technology instead of the business.

Imagine a trading company. The team is constantly chasing suppliers. Someone checks backorders manually. Stock shown in the ERP does not always match the stock the team can actually promise to a customer. Supplier documents arrive by email. Excel is used to rebuild information already sitting somewhere else. Every time the owner wants a clear picture of outstanding orders, someone has to prepare another report. The company is growing, and everybody is busy.

Management looks at the situation and thinks:

We need AI. Let's hire somebody.

But "we need AI" is not a business problem. The problems are:

• Why are people chasing every purchase order?
• Why does answering a customer question require checking three different places?
• Why are people typing information twice?
• Why do managers only discover supplier problems after they have already affected a customer?
• Why does more revenue always seem to require another person?

Those are business problems. Technology should come after you understand them.

You can hire a great engineer and still build the wrong thing

This is not just something we see in SMEs. RAND researchers interviewed experienced data scientists and engineers to understand why AI projects fail.

The engineers could work for months and successfully deliver what they had been asked to build. It simply didn't create much value for the business.

RAND also found another interesting problem: 16 of the 50 industry practitioners described situations where technical teams became more interested in using newer tools and technology than in finding the simplest solution to the user's real problem. And 30 of 50 discussed persistent problems with the quality of the company's data.

That should matter to any business owner thinking about hiring internally. The engineer may be excellent. The brief may be nonsense. And if nobody in the business has properly understood the operations, the engineer cannot magically know what matters.

The people doing the work usually know where the real problems are

An operations manager may tell you:

The ERP says we have the stock, but I still need to check before confirming the order.

Your procurement person may tell you:

I have to chase every supplier because otherwise I don't know if the delivery date has changed.

Customer service may say:

I need to ask operations every time a customer wants an update.

Your finance person may have another Excel file because the figures they need are not available in the way they actually use them.

Those details may look boring compared with a management presentation about an "AI strategy." But that is where the business actually operates. Operations are not a PowerPoint diagram. They are hundreds of small checks, decisions, exceptions, emails, spreadsheets, documents and conversations happening every day. And if you don't understand those first, there is a very good chance you will automate something that looks impressive in a demo while the team continues doing the real work exactly as before.

A technically impressive tool nobody uses is still a failed business project

This is where many AI projects go wrong. A company builds an assistant. Or a new dashboard. Or a forecasting tool. Management sees the demonstration and everybody is impressed. Six months later, the operations team is still using Excel.

Why? Because the system was added on top of the business instead of changing the way the work actually gets done.

McKinsey examined 25 different factors associated with companies getting financial value from generative AI. The factor with the strongest relationship to improved financial results was not simply having better technology. It was redesigning how the work itself gets done.

In other words: companies often spend most of their attention on the smallest part of the problem.

AI investment is moving faster than business clarity

There is another reason this happens. Nobody wants to be left behind. AI is everywhere. Competitors are talking about it. Employees are already using it. Boards are asking questions. Management feels it needs to act.

More interestingly, 64% of CEOs admitted that fear of falling behind can push them to invest in technology before they clearly understand the value it is supposed to create.

That is the trap. The conversation becomes:

What should we build with AI?

when it should be:

What is stopping this business from handling more customers, orders and work efficiently?

Those are very different questions.

"But we want to keep control of our data"

That concern is completely legitimate.

So yes, businesses should care about where their information goes. But there is an important distinction: hiring an employee does not automatically give you control of your data. And: owning your data does not mean you need to build your own AI model.

What matters is much more practical.

• Where is your information stored?
• Who can access it?
• What information is allowed to leave your company?
• What gets recorded?
• Who has permission to do what?
• How do you remove access when someone leaves?
• Can you see what the system has done?

Those are decisions about how the system is set up and controlled. They are not solved simply by putting an engineer on your payroll. A company can work with an external specialist while keeping tight control over its information. And a company can employ its own technical team while still having terrible controls.

Data control is a business requirement. Internal hiring is only one possible way of delivering it.

We have seen the opposite approach work

Start with the work. Then choose the technology.

The latest research is increasingly pointing in the same direction. In July 2026, BCG reported that although nearly nine in ten CEOs were seeing some AI benefit somewhere in their companies, only 26% had embedded AI into a broader business transformation.

That is a much more useful lesson for an SME than trying to predict which AI model will be best next year.

Understand the business. Understand the work. Understand where capacity is being lost. Then decide what should be automated, what should stay human, what should be connected, what should be bought and what genuinely deserves to be built internally.

So when does hiring internally make sense?

There are absolutely situations where it does.

• If technology is part of what makes your company genuinely different, developing an internal capability can be a very smart investment
• If you have a permanent pipeline of work that justifies a technical team, hiring may make sense
• If you already understand your priority workflows, know what success looks like and need someone continuously improving them, an internal resource can be valuable

McKinsey recently studied AI-focused companies and found a useful principle emerging among them: build what genuinely makes the company distinctive. For internal operational tools, many still use existing solutions rather than building everything themselves. McKinsey also warns that something inexpensive to build today can become expensive to maintain over time.

That is a much better build-versus-buy test than:

We want control, so let's build everything ourselves.

Five questions to answer before you hire

Before committing hundreds of thousands of dollars to an AI or data hire, answer these questions:

• Which operational problem is costing us the most time, money or capacity today?
• What does the team actually do manually to manage that problem?
• What measurable business result should change if we fix it?
• Which systems, spreadsheets, documents and people are involved?
• Do we genuinely need a permanent internal technical capability to solve this, or do we first need to fix and connect the way the business operates?

If you cannot answer the first three clearly, you probably aren't ready to hire yet. Not because AI is wrong for your company. Because you haven't defined what you are buying.

The HK$1 million question

Before you commit close to HK$1 million to internal AI capability, ask yourself: if this person does an excellent job over the next 12 months, what will be visibly better in my business?

• Will the same team handle more orders?
• Will supplier chasing disappear?
• Will reporting stop consuming half a day?
• Will customers get answers faster?
• Will you stop adding admin headcount every time revenue grows?
• Will margins improve?
• Will management regain time?

If you can't describe the answer in business terms, don't start by hiring the engineer. Start by understanding where the business is losing capacity.

AI does not make an unclear business problem clearer. It can simply make it more expensive.