TechHouse diagram showing data, security, systems, and people and culture as the foundation supporting AI adoption.

Could AI Bring Your Business to a Ground Stop?

When we bring powerful new technology into our business we own what happens thereafter.

That’s what I was thinking about as I was watching the press conference on the nationwide FAA ground stop. I was sitting in a hotel room I hadn’t intended to be in, having had my connecting flight cancelled.

Officials have made it clear that the AI system wasn’t the system that actually turned things off. They’ve also talked about the new technology placing new demands on the old infrastructure which it simply wasn’t designed to cope with.

For me the whole debate about whether AI technically caused the problem is rather pointless. What we really need to learn is something else.

AI adoption can bring everything to a halt, even when AI systems are not the direct cause.

AI can move far more information, trigger far more activity and connect far more parts of a business at a speed and scale that we’ve never had to deal with before.

Put that capability on top of poor data, outmoded systems, excessive access, poor recovery practices or an organisation that doesn’t embrace different viewpoints and a fairly minor weakness can very quickly become a massive problem.

The good news is we know a lot about how to prevent that.

We know how to deploy new tech. We need to follow those learned lessons with AI.

I’ve worked in technology for more than 30 years, and I don’t remember another major technology that has been this easy to adopt this quickly.

Someone can buy a license and be up and running today. An employee can connect the AI to the company information. A department can automate the process. An AI agent can start interacting with other apps and workflows.

That’s exciting.

It’s also how AI adoption can outstrip the foundations upon which it’s built.

We’ve spent decades learning the basics of reliable IT: understand and protect your data, control access, have your critical systems up and running, monitor what’s going on, know what the technology is allowed to do, have good backups and test your recovery.

We’ve also learned that good systems need people with different bits of knowledge and different perspectives, people who are prepared to challenge our assumptions honestly and leaders who are prepared to change course when the evidence tells them that’s what they need to do.

Those are all elements of the foundation.

AI doesn’t replace any of them. It simply increases the load we’re asking that foundation to carry.

Sometimes ‘shoring up the foundation’ means replacing the foundation.

We’ve got a customer now who wants better dashboards, more automation and access to the new tech. One of their key business systems, however, is over 15 years old.

They’re in a tough spot.

Do they build exciting new capabilities on top of what they’ve got, or do they spend months and a good chunk of their technology budget on modernizing the underlying system first?

I completely understand why the first option is so tempting.

No-one is going to be excited about spending money to get back to a solid base.

You want the dashboard. You want the automation. You want AI. Your competitors are already on the move.

But think of it as a house.

Suppose you’ve got a one-story house and someone builds a two story house next door which then blocks your view. You’d of course want to go up.

But if your foundation was designed to support one story, wanting two more stories doesn’t mean you can have them.

You’ve got to work out what will actually work. Can your existing foundation be strengthened? Does part of it need rebuilding? Does the whole thing need replacing? Is there some other way of achieving the outcome you’re after?

Technology is no different.

Sometimes “shoring up the foundation” means cleaning the data or tightening the security. Sometimes it means improving monitoring, backups orrecovery. Sometimes it means changing the way people work. And sometimes it just means it’s time to finally replace that 15-year-old system which you’ve been running successfully for all those years.

That’s a far harder call to make, particularly when the alternative is to spend that same sum on AI, dashboards and automation at the very moment. And it can be difficult to wait for the foundation when we want to see that view right now.

But getting the exciting bits built first can be expensive in ways that aren’t at all obvious at the outset.

If you’re spending your hard-earned tech budget on AI and automation on a foundation that you already know needs to be changed then some of that investment may well end up having to be re-designed, re-built or even thrown away when the underlying systems do change.

And that’s only the first cost.

The second is what happens when you’re running some very powerful new technology on top of a very fragile foundation: bad data moving faster, security issues reaching further, processes simply overwhelming the older systems, or automated decisions having consequences at a scale that you weren’t at all prepared for.

So, bolstering the foundation isn’t money you’re spending in place of moving on. It’s how you protect the money you’re going to spend.

1. Data: Bad data gets faster too

‘Garbage in, garbage out’ is probably one of the oldest sayings in the technology world.

AI hasn’t repealed it.

If anything, it raises the stakes.

A human working from wrong information can make a bad decision. An AI-enabled process working from the same information could go on to make that same bad decision across hundreds or thousands of transactions before anyone ever realizes what’s happened.

AI can do something wrong extremely well, very quickly and on a massive scale.

Before you connect AI to company information, know what you’re giving it access to.

Is the data good enough for the AI to be doing what you want it to do? Who owns it? Who should be able to see it? What about the information that the AI itself generates? Can the systems that are receiving that information cope with the additional load?

We’re not tossing out decades of good data management practice just because AI has arrived.

We need them more than ever.

2. Security: Don’t give AI more reach than it needs

Security pros have been on the least privilege bandwagon for years now. People and systems should only have the access they absolutely need to do their job.

AI makes that discipline more important, not less.

If an employee by accident has access to information they shouldn’t have, that’s a security issue.

Give an AI agent excessive access, connect it to other systems and allow it to take actions and the potential impact becomes massively greater.

Start with the fundamentals.

Know who and what has access. Protect sensitive information. Limit permissions. Keep the systems that AI connects to secure. Watch for any odd activity. Have a human approval step in place where the potential consequences are significant.

And don’t just give a system more authority because it’s technically capable of using it.

The aim isn’t to prevent everything going wrong. It’s also to ensure that one thing going wrong doesn’t lead to everything going wrong.

3. People and culture: Can someone challenge the plan before it becomes the problem?

The best time to stop a bad outcome is not when the warning lights are on.

It’s a lot earlier really, when you’re still asking whether you’re building the right thing at all.

The people who designed the system will understand the technology. The people doing the work will have an understanding of where the process breaks down. Someone in security will have spotted a risk that the rest of the team may not have. Someone responsible for the data will understand the fact that the data isn’t good enough for what we’re asking AI to do.

And someone with a different background or different perspective might well be the one to ask the fundamental question that everyone else doesn’t see.

We need sufficient diversity of perspective to get those issues to the fore and sufficient openness for the answers to be able to change the plan.

People need to be able to say:

  • “This isn’t the right project.”
  • “We aren’t ready for this yet.”
  • ‘The data won’t support the outcome we’re promising.’
  • “This system can’t safely carry what we’re planning to put on it.”
  • “There’s a better way to deal with this.”

And leaders have to listen.

That’s why transparency isn’t just a nice cultural value.

It’s part of good systems design.

If everyone is out there trying to prove that management’s preferred solution is the right one, then you don’t actually have a review process. You just have confirmation.

Anyone’s loyalty has to be to the good outcome rather than to the project, the technology, the person who put it forward or management’s preferred answer.

With AI that’s even more important as technology can move at an absolutely phenomenal pace once it’s got going.

Your best protection might well be the question that someone was prepared to ask before you even turned it on.

4. Backup and recovery: Plan for the things you didn’t prevent

Good IT has never been built on the basis that nothing will fail.

Hardware fails. Networks go down. People make mistakes. Software has bugs. Cyberattacks happen.

Good planning is all about trying to prevent problems. Good IT also assumes that eventually something is going to get through.

So, we’re ready to go and start the recovery.

If an AI-enabled process becomes important to your business, ask some basic questions:

  • Can we stop it?
  • Can we recover the data?
  • Can we roll back an action?
  • Can we tell what actually happened?
  • Can the business operate in some other way for a period of time?
  • Who has the power to intervene?
  • Have we actually tested any of this?
  • What systems, processes and people could be affected?

There is one more question that becomes particularly important when technology is operating at AI speed:

What indicators will tell us that something is going wrong while the problem is still very small?

You shouldn’t be discovering the limits of your recovery plan for the first time when you actually need it.

Find your own aging infrastructure.

Every business has its own version of the old copper lines.

Possibly the customer information is all over various spreadsheets.

Possibly permissions have built up over the years, and no-one now knows who can actually do what.

Perhaps there’s a critical application that no one wants to touch as the person who really understood it has been away for years.

Possibly backups run every night, but no-one can remember the last time a backup had actually been restored.

Perhaps everyone at the table is thinking the same thing and no one is actually asking whether the project in itself is the right answer.

Perhaps people know there’s a problem but just don’t think that leadership wants to hear about it.

Perhaps your infrastructure just wasn’t built to cope with the volume of data, automation and connectivity you’re about to throw at it.

Or perhaps it’s the 15-year-old business system that’s still running but has now effectively decided what the rest of the business can and can’t do.

Those are foundation problems.

You don’t need to modernize everything before you start using AI.

You do need to understand which of the weaknesses could turn AI’s speed and scale into the speed and scale of a failure.

Then you have to be honest about what can be reinforced, what needs to change and what actually does need to be replaced.

Powerful technology needs a solid foundation.

There’s another analogy I’ve occasionally used for AI: dynamite.

Dynamite has helped us build the roads, the tunnels, the railways and all the other infrastructure that would have been very difficult to build in any other way.

Its power is what makes it useful.

Its power is also why we don’t use it casually.

I’m absolutely convinced of what AI can do for business. It gives small businesses the ability to do things that previously would have required a much larger team, a whole range of specialist skills and a budget that simply wasn’t there.

But ‘easy to adopt’ is not the same as ‘easy to support’.

Fortunately, we don’t have to throw away 30 years of IT experience and start again.

Protect the data. Control access. Keep the systems running. Monitor what’s going on. Minimize the potential impact if something does go wrong. Have good backups and test the recovery.

Bring different perspectives to the key decisions. Get people to challenge the premise rather than just the way it’s going to be done. And be prepared to listen when someone tells you something you didn’t really want to hear.

When the technology underneath all of that can no longer support what you’re trying to build then you need to be ready to have that tough conversation about replacing it.

Sometimes the most important investment you can make in your AI strategy isn’t AI at all.

It’s about solidifying the foundation so your business is ready to carry what you’re building on top of it.

So absolutely ask:

What could AI do for us?

Then ask the harder questions:

Is the foundation beneath us up to it?

And have we created an organization that’s going to tell us if the answer is no?

Because the best ground stop is the one you never need.