Along with everyone else I’ve been thinking a lot about AI lately, and somehow I keep coming back to spreadsheets. I know. Possibly the most middle-aged tech comparison I’ve ever made. But bear with me.
When spreadsheets arrived on everyone’s desktop, something quite profound happened. People stopped having to wait for the IT Department. Suddenly, users could manipulate data, build models, answer questions and create their own solutions. It was hugely empowering. It was also, arguably, the beginning of what we now call shadow IT.
And then we gave people even more power.
I remember tools like BusinessObjects making business intelligence and reporting available to a much wider audience. Again, absolutely the right idea. Why should somebody have to raise a request with IT every time they want to answer a question about their business?
But there was a problem; before long, organisations could have dozens of reports apparently answering the same question — and giving slightly different answers. In retail, even something as seemingly straightforward as markdown could have four or five definitions depending on who had built the report. Which report was right? Which definition were we using? Where had the data come from?
The technology wasn’t really the problem. We’d democratised the tools without necessarily democratising a common version of the truth.
And now we have AI
AI feels strangely familiar. ChatGPT, Claude, Copilot and the rapidly growing collection of AI tools are putting extraordinary capability directly into the hands of users. Except this time, the capability being democratised is vastly more powerful. People can analyse data, write code, interrogate documents, build applications, automate processes and create their own agents, often without involving IT at all.
So if spreadsheets helped create shadow IT, AI has the potential to put it on steroids.
But I don’t think the answer is to try to stop it, quite the opposite.
Trying to restrict people from using these capabilities feels both unrealistic and counterproductive. People will use them, just as they used spreadsheets. And businesses should want them to.
The challenge is to enable that freedom without allowing the truth underneath it to fragment.
Perhaps we can do it better this time
There’s an important difference between the spreadsheet revolution and the AI revolution. With spreadsheets, users often assembled their own data from multiple sources. They downloaded something from one system, keyed something in from another, added their own assumptions and created a model. The spreadsheet became both the tool and the data store.
Once hundreds of people started doing that, controlling the underlying truth became almost impossible.
AI gives us an opportunity to do something different. We can give users extraordinary freedom at the intelligence layer while being much more disciplined about the data underneath it. The spreadsheet fragmented the truth. AI could actually reinforce it.
The prerequisite is that we treat foundational data as sacrosanct. Product. Stock. Customer. Supplier. Location. Transactions. The definitions, relationships and governance around that data aren’t suddenly less important because AI has arrived; they are more important. An AI capable of producing an answer in seconds is only useful if we can trust the information on which that answer is based.
From shadow IT to proliferated intelligence
That leads me to a slightly different way of thinking about shadow IT. Perhaps it doesn’t necessarily have to be something we fear. If we can establish a governed, trusted foundation, why wouldn’t we want people across an organisation experimenting with AI?
Let them build. Let them analyse. Let them ask questions we hadn’t anticipated. Let them automate things nobody had thought were worth automating. Let intelligence proliferate throughout the business. But give all of that intelligence a common foundation to work from.
That’s where I think the really interesting opportunity lies. For businesses, and particularly retailers, the AI conversation can very quickly become dominated by models, agents, copilots and whichever new technology has appeared that week. They’re exciting. But perhaps we’re starting the conversation one level too high.
The most important question might not be: “Which AI should we use?” It might be: “What truth are we going to give it?”
Because however intelligent the technology sitting on top becomes, accurate, governed and trusted foundational data remains rather stubbornly important.
Maybe the AI revolution starts with something considerably less glamorous than AI, getting the basics right.
