Every software vendor seems to be telling the same story at the moment. Connect AI to your ERP. Point it at your data. Ask a question in plain English. Receive profound business insight moments later.
It's a compelling vision. In fact, as someone who has spent more than two decades helping organisations make sense of ERP systems, I genuinely believe we're heading in that direction. But my bullshit detector is clanging loudly when I think about this in the context of how most of today’s financial data is structured. I think we're skipping over a rather important detail.
Most ERP data is not a finished building. It's a construction site. And AI doesn't just need a ladder into your data. It needs scaffolding.
Over the years I've watched countless finance managers, analysts, consultants and executives wrestle with the same challenge. The issue is rarely access to information. Modern ERP systems are remarkably good at storing information. The issue is understanding what that information actually means.
Take a simple question:
"What was our margin last month?"
Most people would assume that should be easy. After all, the transactions are in the system. The invoices are there. The costs are there. The journals are there. Yet if you've worked in a ERP environment for any length of time, you'll know that this question is anything but simple.
Which margin? Gross Margin? Invoice Margin? Contribution Margin? Manufacturing Margin? EBITDA Margin?
And based on which revenue definition? Gross Sales? Net Sales? Revenue after rebates? Revenue after trade marketing allowances?
I've sat in enough boardrooms to know that two intelligent people can use the word "margin" while meaning entirely different things. Usually they discover this about twenty minutes into the meeting, right after a heated debate supported by three spreadsheets and a dashboard that everyone trusted yesterday but nobody trusts today.
Humans eventually sort this out because we understand context. We know the history of the business. We know which reports the board uses. We know which exceptions were agreed to six months ago after an uncomfortable conversation involving auditors, accountants and several cups of coffee. AI knows none of this.
AI sees transactions. It sees accounts. It sees classifications. It sees records. What it does not see is meaning. This is where much of the current AI discussion becomes interesting.
Many organisations are treating AI as if it were a highly intelligent employee waiting patiently outside the ERP, desperate to be given access. The reality is closer to hiring a graduate analyst and dropping them into the middle of a warehouse containing twenty years of financial records, operational transactions, custom segments, saved searches and inconsistent naming conventions. Then asking them why EBITDA fell in Sydney last month. The answer they provide might be correct. It might also be spectacularly wrong. The concerning part is that both answers can sound equally convincing. That isn't a criticism of AI. It's simply a reflection of the fact that intelligence without context is often just sophisticated guessing.
The more I work with reporting and analytics projects, the more convinced I become that the real challenge isn't building better dashboards or faster reports. The challenge is creating meaning.
Businesses spend enormous amounts of time discussing what metrics matter. They debate definitions. They establish rules. They create classifications. They decide which costs belong in one category and which belong in another. All of that work is valuable because it creates a shared understanding of how the organisation interprets itself. Unfortunately, much of that understanding remains trapped inside reports, spreadsheets and people's heads.
That's where scaffolding comes in. Scaffolding is not the data itself. It is the structure wrapped around the data. It provides context. It explains relationships. It establishes definitions. It creates a trusted framework within which information can be interpreted consistently.
A well-designed reporting model already knows what Net Sales means. It already knows how Trade Marketing Allowances should be treated. It already knows which costs belong in COGS, which belong in operating expenses and which should never be included in either. The model has already done the hard thinking. The transactions simply provide the evidence.
This is one of the reasons I became interested in building SpeedLedger through ClearSuite Reporting & Analytics. Originally, SpeedLedger was created to solve a practical reporting problem. Many organisations had invested heavily in ERP systems, business intelligence tools and reporting platforms, yet still struggled to answer basic questions consistently and quickly.
The issue wasn't that they lacked data. They were drowning in data. The issue was that the business meaning behind the data remained fragmented.
A finance team might have one definition of Net Sales. Operations might have another. Sales might have a third. By the time someone produced a board report, everyone agreed on the answer but nobody could quite remember how they got there.
The interesting thing is that the largest ERP vendors appear to have reached the same conclusion. For years, the industry focused on collecting data, storing data and visualising data. Today the conversation has shifted toward something much more important: understanding data.
Companies such as SAP have invested heavily in business data fabrics, semantic layers and AI-ready business contexts because they recognise a fundamental truth. AI cannot reliably reason over enterprise data unless it understands the business meaning behind that data.
The challenge is not teaching AI how to read transactions. The challenge is teaching AI what the transactions represent. An invoice is not simply an invoice. A customer is not simply a customer. Revenue is not simply revenue.
Every organisation applies its own classifications, definitions, reporting structures and performance measures. These definitions become part of the organisation's operating language. Without that language, AI is left to infer meaning. With it, AI can reason within the same framework that executives, finance teams and operational managers use every day.
What we are witnessing is a subtle but important shift in enterprise software. The focus is moving away from data storage and toward business context. Away from reporting tools and toward semantic models. Away from simply asking, "Where is the data?" and toward asking, "What does the data mean?"
In many ways, the enterprise software industry is building scaffolding around data because it has realised that AI requires far more than access. It requires context.
SpeedLedger was designed to address that problem by creating a curated financial model that sits between raw ERP transactions and reporting outcomes. It transforms thousands or millions of transactions into trusted business concepts that can be reused consistently across dashboards, reports, analyses and planning processes.
At the time, I viewed this primarily as a reporting solution. Today, I think it may be something more important.
As AI becomes increasingly embedded within ERP, analytics and planning platforms, curated semantic models become incredibly valuable. They provide the context AI needs to navigate business data safely. In effect, they become scaffolding for artificial intelligence.
Instead of asking an AI model to figure out what EBITDA means, the model already knows. Instead of asking AI to infer how rebates affect revenue, the model already knows. Instead of asking AI to determine which classifications matter, the model already knows.
The AI is no longer wandering around a construction site trying to work out what is being built. It's walking through a structure that already has walls, signs, staircases and a blueprint. That's a very different proposition.
The organisations that gain the most value from AI over the next decade may not be the ones with the largest language models or the most impressive demonstrations. They may simply be the organisations that have invested the most effort in defining how their business thinks. Because ultimately, AI does not need more access to data. It needs better context. And context comes from meaning that has already been defined, tested and trusted.
Giving AI a ladder into your ERP allows it to climb into your data.
Giving it scaffolding allows it to understand what it is looking at.
In my view, that's where the next generation of ERP reporting is heading.
Not simply connecting AI to our systems.
But teaching AI how our businesses actually think.
If you made it this far perhaps you can answer a Question: If you connected an AI agent to your ERP today, would it understand how your business defines Revenue, Margin and EBITDA, or would it simply be guessing?