What Makes an Organisation AI-Ready?

An AI-ready organisation has five foundations in place: a clearly defined business problem for AI to solve, data it can trust, processes it understands, people who have been trained and supported to use AI, and human expertise to judge the output. Most organisations experimenting with AI have the tools. Far fewer have the foundations.

On Humans Behind the Tech, our podcast about why people build what they build, we keep asking founders, CTOs and AI consultants the same question: “What separates the organisations getting real value from AI from the ones quietly burning budget on it?”

Over the past six years we have also built more than 60 products at Old.St Labs, around a dozen of them with AI at their core, so we have seen the difference up close. The answer that keeps coming back has very little to do with which tools you buy.

Here are the five foundations, each framed as the question people actually ask us, because that is how the gaps show up in practice. The answers are in the words of the people we have interviewed.

‍

Why do AI projects fail without a business strategy?

AI projects fail without a strategy because they start with a tool rather than a problem, and that is the first foundation: a business problem you can name before AI enters the conversation. The organisation buys AI because it feels like it should, then hunts for somewhere to point it.

Richard Geary, founder of Flo HQ, spent 25 years in corporate tech, including senior roles at Salesforce, before founding an AI consultancy for businesses the big players ignore. He put it like this on the podcast:

"Often people will buy a tool, then go looking for a problem to use that tool on. We are proponents of the exact opposite of that. Understand your processes, understand your pain points, what your problems are, what your opportunities are, then go and buy the appropriate tool to actually deliver a value against that need."

The organisations getting real value run it the other way round. They start with a business challenge they can name, an inefficiency they can measure, an outcome they can define, and then ask whether AI is the right way to get there. Sometimes it is not, and finding that out early is cheap. Finding it out after the rollout is not.

Jason Margolin, CTO of InTouchNow.ai and formerly of Meta, gave the same advice from the delivery side:

"Make sure that you've got a business goal in mind, and make sure that you don't forget that key goal or that key metric that you're trying to work towards. Keep that goal in mind at all times. Let that be your North Star, because it's easy to go off in lots of different directions, especially when you're building a new product. Especially with AI now, it's even easier."

Before any AI conversation, you should be able to answer three questions. What problem are we solving? What does success look like? How will we measure it?

‍

What data do you need before adopting AI?

You need data that is accurate, accessible, consistent and governed in the specific area you want AI to work on: the second foundation. AI output is only as good as the data underneath it, and no amount of model sophistication compensates for data you cannot trust.

Richard Geary reaches for a framework older than the hype:

"The three Vs of data: volume, veracity and variety. Where does it sit? How much of it is there? What's actually in it? Is it labelled well, and how quickly does it move, and where does it move to and from? That last piece for me is where most people trip up when it comes to AI."

Craig Brown, founder of Troubadour, the UK's first dedicated product marketing consultancy for B2B tech startups, drew the line that matters most in practice:

"You get more from an LLM when you have the data already, and you're not asking the LLM to go out and acquire data for you to then analyse."

He learned it the hard way, asking an LLM to build competitor intelligence from scratch:

"The hallucinations just start coming through and making claims that aren't true. 'Look at this feature set for this competitor.' And I said, well, actually, they don't have these features at all. Or you're missing out certain capabilities. Unfortunately, it requires so much change that it renders LLM usage for proprietary competitor intelligence somewhat useless."

When he fed data he had gathered and trusted, though, the same tools analysed it quickly and accurately. That is the pattern. If your data is fragmented across systems, inconsistent between teams, or nobody is sure which version is right, AI will confidently produce answers built on all of that, and you will not know whether to trust them. You do not need perfect data across the whole business. You need trustworthy data in the specific area you want AI to work on. Assess that honestly before you spend.

‍

Can AI fix broken processes?

No. AI accelerates whatever process you point it at, and that is why sound processes are the third foundation: it exposes weaknesses, it does not repair them.

Richard Geary again, with the warning worth pinning to the wall:

"If you are taking a turbocharging element like AI and slapping it onto a process that is broken or ill-defined, you'll find out really quickly at scale what bad and chaos looks like. So it's truly understanding your process, understanding the happy path of that process and the unhappy path of that process, then figuring out where there are issues, why those issues exist."

Automating an inefficient process gets you inefficiency at speed. So before AI touches a workflow, you need to understand how it actually operates today, where the bottlenecks are, and whether the process itself needs fixing first. In our experience building AI products for clients, the projects that work start with a process problem someone can describe precisely, not with an ambition to "use AI". 

‍

Why is AI adoption a people problem?

AI adoption is a people problem because tools do not adopt themselves, which makes enabled people the fourth foundation. The most common failure mode we hear about is not the technology underperforming; it is the organisation assuming people will just start using it.

Scott Evans, founder of Wonderland AI, spent eight years delivering digital transformation in financial services before going out on his own. He was blunt about where things go wrong:

"People are just running for the shiny tool, but the biggest part of AI adoption is enablement. There's a huge lack of understanding about what AI is, what it can do. And there's also a lack of urgency around AI literacy and upskilling people in the business."

His favourite illustration is from his own kitchen table:

"My wife's a teacher. I have an AI business. In fact, I have two AI businesses. She has not even clicked on Copilot. When I asked her if she'd used Copilot, she said, is it that rainbow-looking button on Outlook?"

If the person married to an AI consultant has not opened the tool sitting in her inbox, a launch email to your staff is not going to do it. Richard Geary described what the rollout data actually looks like:

"Everybody gets really excited in week one and by week two, it's like gym usage in February. It's absolutely tanked."

And he named the group most rollouts forget:

"Often the most diligent workers in a business are the last to adopt a new technology because they don't want to get it wrong or they don't want to make a mistake or they don't want to leak something."

Adoption needs leadership that visibly uses the tools, education about what AI can and cannot do, proper training rather than a launch email, clear ownership, and the psychological safety for careful people to experiment. None of that is glamorous. All of it decides whether the investment pays off. And as Scott put it: the industry is moving fast, but you have time.

‍

Will AI replace human expertise?

The opposite, mostly, which is why expertise is the fifth foundation: AI raises the value of knowing what you are doing, because someone has to judge the output.

Jason Margolin, who uses AI tooling daily across the startups he advises, described both sides of it:

"If you know what you're doing, it is a superpower. For me, it is a genuine superpower. I churn through work so much quicker because of it that I would never be able to take on as much business as I can take on without it. But someone who doesn't have experience and is using it, it's just spitting out stuff that you don't really know or understand. And how do you trust it? It is a problem."

Craig Brown made the same point about marketing work:

"In order to get the most out of it, you need to know what a good answer looks like, what a good output looks like."

I feel this one personally. I co-founded a tech business, but I don’t know how to code, and when I build something with AI tools it is like building a car and not knowing how the engine works. AI can accelerate research, drafting, analysis and build work dramatically. But it cannot tell you when its answer is subtly wrong, missing context, or right in general and wrong for your business. As Jason put it, these tools are "not at a mature enough state yet that it requires no human or no technical oversight". The organisations that benefit most are not replacing expertise. They are amplifying it.

‍

The AI readiness checklist

‍

If you can tick all five, you are more AI-ready than most organisations experimenting with AI right now. If you cannot, the gaps tell you exactly where to start, and none of them starts with buying a tool.

‍

Every organisation we speak to is somewhere on this journey, and most are further along on some foundations than others. If you want to get started, get in touch with  us today!

‍