Every day, I see posts on LinkedIn and in the media proclaiming that “the AI bubble is about to burst”. Brutally, it’s nonsense. It’s nonsense because bubbles are the completely wrong analogy.

I asked ChatGPT what it means for something to be a ‘tech bubble’. It suggested three signals:

1. Overvaluation: Technology companies (especially startups) are valued far higher than their real revenues, profits, or fundamentals justify.
2. Speculative investment: Investors pour money into tech simply because they expect prices to keep rising, not because the businesses are solid.
3. Fragile market: Like a soap bubble, the market looks big and shiny, but it can “burst” suddenly if confidence collapses.

Every time there is a new technological innovation we get excited. We tend to look to them as a panacea to solve our problems, whether it be the internet 20 years ago, data-science and machine learning 10 years ago, or now generative AI.

What happens is that CxOs get told to adopt these technologies by research reports, tech-consultancies, or fellow CxOs wanting to build careers around a technology, which in turn creates a deafening echo-chamber of 101 rhetoric and misinformation. Technologies DO become overvalued, there IS speculative investment, and it DOES create fragility in the market.

The incorrect conclusion that is often drawn then is that the technology is ‘snake-oil’ or that it can’t deliver the expected business benefits – that it is ‘hot air packaged in a thin shiny veneer’. AI is not a bubble, it’s a mountain that has hidden riches. You just need to know how to find them.

I know that it’s a blindingly obvious thing to say, but always start with the problem (or the answer/outcome you want) and work backwards. Don’t start with data or technology. Failed AI projects are not the fault of the data or technology, they’re the fault of people misunderstanding how to apply the right technologies to solve the right problems.

Having built and scaled AI solutions in production for two decades for some of the world’s biggest companies, here are my guiding principles for success and red-flags for failure:

To succeed, you have to be willing and able

Willing

• Does this solve a real/tangible/realisable pain or create a gain?
• Is there an appropriate budget to build, support, maintain and innovate?
• Do you have top-to-bottom stakeholder alignment and agreement?
• Are there incentivised internal champions, sponsors, and key stakeholders?
• Is procurement savvy enough to take calculated risks?

Able

• Have you got a track record of building/deploying scaled software?
• Is the infrastructure (not necessarily data) ready (or readable in time)?
• Are there no competing or heavily dependent projects?
• Are there no complexities around existing vendor renewals or lock-in?
• Do you have a dedicated and appropriately skilled team?

Red flags to watch out for and avoid

• You have a strategy (on a page): We’ve all seen the data/AI/digital ‘strategy-on-a-page’. Some pay an extortionate amount of money for them. The hard part is detailed execution. Show me the core functions of my business, and for each function show me what projects/levers I can pull with technologies that are going to drive value. Don’t over-complicate things with overly detailed business-cases, there are simple and lightweight methodologies to start executing and driving immediate value.

• You have never built, deployed, scaled, and maintained software: I can’t emphasise this enough, if you haven’t built and scaled software then you’re unlikely to build and scale AI. AI projects typically fail and succeed for the same reason software projects fail and succeed; you have to be fully willing and able (see above).

• You are focused on quick wins and low-hanging fruit: I’ll loosely defer to Newton’s second law of motion; you get out what you put in. Yes, sometimes you get lucky and there’s a disproportionate return on investment, but focusing on “quick wins” often means trying to solve problems that are easy and actually not strategic or differentiating for your business. Yes, you can build AIs or agents that do expenses, onboarding, and other back-office tasks, but the reality is that there are (or soon will be) third-party providers that provide you with those solutions at a fraction of the cost. Focus on building solutions that differentiate you from your competitors; these are not typically low-hanging or quick solutions.

• You are waiting for data to be ready: Ten years ago we were told to build data-lakes, put an analytics tool on top and hire data scientists (or make analytics self-serve) to extract insights that will drive value. Ten years later, is your data ready? No. Have you found insights that drive the value you expected? I doubt it. In my experience, giving humans more or better insights typically doesn’t lead to better decisions. Your data will never be ready. Companies don’t have insight problems, they have decision problems. Don’t start with data, start with the desired solution to the problem and work backwards to the data.

• You are hiring an AI team: The reality is that most companies can’t attract bleeding-edge AI talent. If you want to build differentiated innovations then you need differentiated talent. If your CxO is wanting to hire an AI team, they are likely trying to build a career around the technology and have no idea what it takes to attract, nourish, and retain top AI talent. Hire some smart technologists that deeply understand your domain who can work with a third-party provider that have a proven track record of delivering scaled AI innovations.

AI is a mountain

Many mountains contain precious stones, minerals and metals. Finding these valuable assets requires a combination of luck, expertise, effort and having the right skills, tools and methodologies. Just like mining for precious materials, if you point your resources to the wrong places or don’t get there first then you won’t reap the desired outcomes. It’s not the fault of the mountain, it’s the fault of leadership being misinformed about what is required to be successful. Ask yourself, am I trying to apply GenAI to problems it is incapable of solving? Think of GenAI as an intoxicated graduate. If an intoxicated graduate can’t solve your problem then GenAI will also likely fail. There may be other flavours of AI that are much more applicable to solving your problem, from basic RPA to machine learning, advanced analytics to discrete optimisation.

There really is ‘gold in them there AI hills’. Riches beyond your wildest dreams. But you’ll have a much better chance of placing the right bets if you work with a true expert to help you on your journey. The million-dollar question is “how do you know who’s a genuine expert from people who have rebranded themselves as AI experts over the past 10 years?”. One easy thing to ask is, can they prove – genuinely prove – that they have a track record of consistently striking gold.