Ask a leadership team to score their business from one to five on its data, its systems, its processes and its people, and the same pattern appears almost every time. Most groups land somewhere between one and three across the board. Systems is nearly always the lowest score in the room. People is nearly always the highest, mostly because leaders feel they know their own teams well. That confidence does not always match how ready those teams actually are to adopt AI. The gap between what a business believes it has in place and what it actually has is where most AI projects go wrong.
By Tim Butler, CEO, Innovation Visual
This piece brings the four foundations of successful AI implementation together. A project rarely fails because one foundation is weak. It fails because two or three are weak at the same time, and the moment you switch AI on, each weakness is exposed.
When an AI project disappoints, it’s easy to blame the tool. The agent gave a poor answer, so the agent must be broken. In our experience that is almost never true. An agent gives a poor answer because:
The model is usually doing exactly what it was built to do. It is the ground underneath it that was not ready.
We caught a version of this ourselves, testing one of our own AI agents. It had been set up to identify sales prospects and queue them for outbound contact. One of the contacts it picked up already had a live deal open with our own sales team. The information existed. It simply lived in two places that were not properly joined up, so the agent had no way of knowing. Nothing was wrong with the agent. It did precisely what it was asked to do, with precisely the picture of the business it had been given.
This is also why fixing just one foundation on its own rarely helps.
The four foundations support each other, and the return goes up sharply once they are all in place together rather than tackled one at a time. Across the businesses we have worked with, from £5m turnover up to £1bn and beyond, that combination matters far more than which tool you choose.
Before looking at any of the four foundations, clarify the problem you are trying to solve. The businesses that waste the most money on AI start with a shiny tool and go looking for something to point it at. The ones that get a return start with a real business problem, look closely at the process behind it, and only then work out where AI fits.
Better to be the painkiller than the vitamin, solve something that is already costing time or money rather than something that would merely be nice to have. Choose a first project with high impact and low effort. An early win buys you the confidence, and the budget, for the harder projects that follow. Once the problem is clear, the four foundations decide whether you can solve it well.
Here is what each one means in practice, what it looks like when it is weak, and where to go for more detail. As you evaluate these foundations, be honest about your own business. The ones you cannot answer with confidence are usually worth fixing first.
Your data is what AI uses to do its job, so if it is wrong, out of date or incomplete, the results will be too. Data does not just mean rows in a spreadsheet. It includes your brand guidelines, your documents, your past work, and everything sitting inside your existing systems, because all of it forms the context an agent is working from.
The two things that trip businesses up most are silos, where no single place holds the full picture of a customer, and structure, where records are not properly linked so an agent cannot follow the trail. We see this constantly with something as simple as company records. A business like Nike will have a .com, a .co.uk and a .org, and if those are not linked inside your CRM, an agent researching that account has no way of knowing they belong together. The same goes for a contact record with no company attached. The agent cannot pull information at a company level if the association was never made.
You do not need perfect data. You need the part that matters for the job in hand to be accurate and connected. Getting that right is usually the best return a business can get before spending anything on the technology itself.
A sign this foundation is weak: you cannot pull up one complete view of a customer in a single place, or you already know your records are full of duplicates and gaps.
If data is the raw material, systems are the plumbing. What matters is not how good each individual tool is, but how well they work together. You want one reliable source of truth covering the whole customer journey, marketing through to sales, onboarding and success, so your people and your AI are working from the same information. This is what revenue operations is about, joining up the functions that touch a customer so information moves freely between them rather than getting trapped in one department's view of the world. Where systems are stuck in silos, the answer is almost always to connect what you already have before you buy something new.
A sign this foundation is weak: your teams work in separate systems that do not talk to each other and getting a simple answer means checking three different places.
Processes are where your people and your systems meet, and they hide more wasted time than almost anything else in a business. We often see that most were never really designed. They grew over time, with extra steps and workarounds bolted on to solve problems that may no longer exist. Point AI at a process like that and you have not fixed anything; you have simply made the same inefficiency run faster.
That is why the process work must happen first. Map it from start to finish with the people who do the work, find the steps that no longer earn their place, and only then decide where AI belongs. Automating the time-consuming tasks of your most expensive people is usually the fastest route to a return, because that is where the cost of the old way was highest. We cover this in detail in our piece on redesigning your processes before you automate them.
A sign this foundation is weak: nobody in the business can draw the process out from start to finish, or the version on paper looks nothing like what people actually do.
People are the most important of the four, and the ones leaders are most likely to leave until last. A brilliant tool that your team does not trust, does not understand, or quietly resents will not deliver anything, no matter how well it performed in the demo. In the sessions we run with leadership teams, the challenge that comes up more than any other is not technical. It is buy-in, and underneath the buy-in question sits a much more human one, whether people believe AI is coming for their job. We need to take our people with us to make the technology successful. That means being honest about why AI is being introduced, what it changes about someone's job, and just as importantly what it does not change. The best implementations we see expand people's roles rather than shrink them, taking the low-value work off their plate so their time goes where it is genuinely needed.
A sign this foundation is weak: AI has been rolled out but hardly anyone uses it day to day, or your team sees it as a threat rather than something that helps them.
Once the four foundations are in place, governance is what protects you as you scale, and it is simpler than most leaders expect. It starts with being honest about risk and deciding, in writing, what you will and will not do with AI.
When we asked a group of manufacturing and defence sector leaders for a show of hands on whether their business had a written AI policy, the result was sobering. Several sizeable organisations had nothing down on paper at all, and at least one that thought it did had never actually read it. A policy nobody has read might as well not exist. A good one sorts your data by how sensitive it is, rules out the free consumer tools by default because your information can end up training someone else's model, names someone accountable for what AI produces, and covers copyright, intellectual property and how you will handle it if something goes wrong.
None of it needs to be lengthy. It needs to exist, be read, and be updated as the tools change. If you would rather not work this out on your own, our AI governance and compliance advisory is there to help.
Go back through the four foundations above and be honest about which ones you could not answer with confidence. For most businesses, two stand out, and they are usually not the two people expected walking in.
You do not need to fix everything at once. Pick the one problem you most want to solve, get the data, systems, processes and people right around that single case, prove it works, then move to the next. A successful leader knows that buying the technology is the easy bit. If you would like an outside view on where your weak points are and what to tackle first, our AI feasibility study is a good place to start.
Our AI Workshop for Leaders is a full day built around exactly this. We work through each of the four foundations against your own business, so you leave knowing where you are strong, where you are weak, and what to do about it in priority order. You can book a place to get started.
If you would rather talk it through first, you can book some time with me directly using my meeting link.
These are the questions leaders ask most often when they start thinking about what needs to be in place before implementing AI.
The four foundations are:
Data and systems give AI the information it needs to work, processes are where your people and your systems meet, and people decide whether AI gets used. Getting the problem clear comes first, and governance sits on top, but these four are what carry any AI project.
Rarely because of the tool itself. They fail because two or three foundations are weak at the same time, so the AI exposes thin data, disconnected systems, a broken process or low adoption the moment it starts working. Fixing just one foundation on its own tends to disappoint, because the others are still holding the project back.
In the AI feasibility studies we run, systems usually scores lowest, mainly because tools have developed separately over time and never learned to share data. People usually scores highest in leaders' own eyes, though adoption is often weaker in practice than expected. Most businesses are missing at least two of the four, and the weak points tend to be the ones they can see least clearly.
No. Data is never perfect, and waiting for it to be will hold you back indefinitely. What is important is identifying the data that matters for the job you want AI to do and making sure that part is accurate and easy to access. Accurate, focused data on the area you are working on beats a vague plan to clean everything.
Start with one problem that matters to the business, map the process behind it with the people who actually do the work, and get the four foundations right around that single case before scaling. Choosing a first project with high impact and low effort gives you an early win and makes it easier to take on the next one.