The AI adoption framework: how to prepare before you build

Sixty percent of AI projects will be abandoned through 2026 without ever delivering value, according to Gartner, and the culprit is almost never the model itself. It's what happens, or doesn't happen, before anyone touches a keyboard.

At Geeks, we built a four-stage structure to close that gap: prepare, align, observe, engage. Our CTO and Executive Director, Matt Mehrjardi, talked through exactly how it works across two episodes of The Innovation Room, and this article draws on both conversations. If you'd rather listen than read, start with part one and part two.

This piece is about the part almost nobody gets excited about: preparation. Because the businesses that get real results from AI aren't the ones with the biggest budgets. They're the ones who did their homework before they built anything.

Adoption of AI isn't one decision you make and move on from. It behaves more like a relationship than a rollout. It starts with an idea, changes shape as your team gets used to it, and keeps evolving long after launch. 

So what is AI adoption, if not the ongoing work of teaching an organisation to collaborate with a genuinely new kind of colleague? One that's brilliant in some contexts and unreliable in others, and needs the right conditions to actually help rather than hinder. 

That's precisely why a structured framework for AI adoption exists. AI isn't plug and play, whatever the vendor demos suggest. It's an evolving asset that needs looking after, and a framework gives your business somewhere to return to whenever the ground shifts, which it will, repeatedly. 

Some people call this kind of structure an AI implementation framework. We think of it as closer to a compass than a checklist, because the four stages loop rather than end. 

Each stage feeds the next, and Engage loops straight back into Prepare, because Matt is candid that no AI system is ever really finished. You release it, watch what happens, and evolve it. The businesses that treat this like a one-off exercise are usually the ones Gartner is counting in that 60%. 

Every AI adoption roadmap starts here, and the businesses that rush this stage tend to pay for it later, usually somewhere between the pilot and the rollout. 

On the podcast, Matt was blunt about the biggest failure point: leadership knowledge. Not the team on the ground, the people making the call. He's seen business leaders who don't believe AI can do anything meaningful, and he's seen the opposite problem too, leaders who expect AI to behave like magic. Both groups end up disappointed, just for different reasons, and both end up dismissing a technology that might have genuinely helped them. 

"We need to think about AI as a creative entity that, like any other human, might make mistakes," Matt explained. Traditional software gives you consistency, even when it's buggy. AI trades some of that consistency for flexibility and creativity, and that trade comes with risk you have to actively manage, not ignore. 

Preparation isn't just a leadership exercise either. Matt is equally clear that it doesn't matter how impressive a piece of technology is if the people actually using it every day don't understand it or don't trust it. Spend all your preparation budget educating the board and none of it on the team who'll be typing into the tool each morning, and you'll have built something brilliant that nobody bothers to open. 

Real scenarios beat theory every time, so here's one Geeks has lived through. 

Search Acumen, a PropTech pioneer Geeks has partnered with since 2013, had a problem familiar to any conveyancer: over 340 local authorities across the UK, each sending property data in its own inconsistent format, buried inside lengthy PDF documents that someone had to read manually. It was slow, it was error-prone, and it was capping how many portfolios their clients could realistically handle. 

The Geeks team started building what became REI, short for Real Estate Intelligence, a tool that automates and standardises the analysis of local authority data, years before ChatGPT existed. As Matt put it on the podcast, "we tried different technologies and the technology didn't give us the results that we wanted", but that didn't mean the decision to start was wrong. Search Acumen had seen a genuine gap in the market and trusted that a workable solution would arrive, even if it wasn't fully formed on day one. 

That's the thinking behind one of Matt's favourite lines from the episode: "if it's hard for you, it's hard for everyone." Technology only ever gets cheaper and more capable, never the reverse, so businesses that wait for perfect conditions are usually just handing the advantage to a competitor willing to start messy and improve in public. Today, Search Acumen's clients get fast, standardised access to data that used to take days to interpret by hand. 

Read the full Search Acumen story 

If you're wondering how to adopt AI without waiting for the tools to be flawless, Search Acumen's story is your answer: you don't wait, you manage the risk instead. 

Matt frames it as a simple question. Why choose AI over traditional software at all? Because you want the flexibility and creativity it offers, not despite the unpredictability that comes with it, because of what that unpredictability makes possible. The trade-off is real. You can never be completely certain an AI system's output is accurate, and if you're running it at scale, any mistake gets amplified across every interaction it touches. 

That's why Matt is cautious about full automation. "I don't think we are at the stage at the moment to rely fully on AI for automation, because that is very, very risky," he said, especially anywhere a system touches live customer data with no human checkpoint. The fix isn't avoiding AI. It's designing touchpoints where a person reviews what AI produces, catches the mistakes, and feeds corrections back in. Pair AI with humans on purpose, rather than hoping it works itself out, and the risk becomes manageable rather than existential. 

Every one of Geeks' AI adoption strategies starts with the same question: what, exactly, are we applying AI to? Matt calls this your context, and it's arguably the single most important concept in the whole framework. 

A context can be as broad as your entire organisation or as narrow as one repetitive task. Imagine you run a customer service team. You could come to Geeks and ask what's possible for customer satisfaction across the whole department, and that becomes your context. Or you could narrow it right down to one workflow, such as how complaint emails get triaged. 

Each context has a current state and a potential state once AI is applied well, and that potential isn't always full automation. If the relationship between your team and your customers is part of what makes your business valuable, that human element might be exactly what you want to protect rather than replace. Writing routine marketing content, by contrast, can often be automated end to end with the right quality checks in place. Getting this distinction right, before you build anything, is what separates a considered approach from a generic template bolted onto the business. 

On episode 58, Lindsay asked Matt directly why data readiness is such a sticking point. His answer was simple: data is what gives your AI solution its intelligence. 

"Think of AI as a super smart entity that knows a lot of different concepts and has a lot of different skills," Matt said, "but you want to introduce your organisation, your context, to that intelligent person. That's where data comes in." Without it, even the most capable model has nothing specific to your business to work with. 

The most common gap Matt sees isn't a lack of data exactly, it's data that only exists in people's heads, never written down anywhere a system could use it. That's a preparation problem, not a technology problem, and it's usually fixable faster than businesses expect once someone actually goes looking for it. 

The fix, according to Matt, is resisting the urge to solve it all in one giant step. Build a phased roadmap instead, take a small piece of the data problem, document it, test whether it moves the needle, then repeat. Businesses that try to digitise everything before touching AI usually stall long before they get to the interesting part. 

If you're not sure how clean your own data actually is, that's precisely what our Digital Due Diligence service is designed to map out before you commit to anything bigger.

Treat this less like paperwork and more like the backbone of your AI implementation framework. Before you can honestly say you've finished the Prepare stage of the AI adoption framework, work through this list. 

None of this needs to take months. Matt is candid that a properly scoped preparation stage can be done in days to weeks, not the quarter most business leaders assume it needs, provided the context piece gets the attention it deserves.

Preparation only earns its value if the following stages follow through, so here's the short version of what comes next. 

Align is where adoption of AI meets governance. It means connecting the initiative to your organisation's north star, defining high-level constraints such as data residency, and keeping a risk register so everyone from developers to compliance can see what's happening and why. 

Observe means tracking your inputs, outputs and KPIs against the goals you set in Prepare, and taking small, evaluated steps rather than waiting a year to find out whether a project worked. Engage is about education at every level, secure and honest communication about what AI is actually for, and building feedback loops so your team feels like contributors, not bystanders. 

We've kept this section brief on purpose. Each stage deserves its own deep dive, and we've mapped the full detail, including the specific questions to ask at each one, on our AI Adoption Wheel page. 

By now, the answer to what is AI adoption should feel less abstract than it did at the top of this piece. It's a journey with a proper starting point, and that starting point is preparation done properly, not skipped in the rush to build something impressive. 

If you'd like a second pair of eyes on your AI adoption framework before committing budget, that's exactly what our AI Opportunity Discovery workshop is built for: a structured session to pin down your context, your potential, and a realistic roadmap, before any code gets written. 

And if you haven't yet, go and listen to Matt talk through all of this in his own words on The Innovation Room. Part one covers when to hit go and when to hold fire. Part two walks the full framework, stage by stage.