When people talk about “being good at Excel,” they almost always mean the same thing: how well a person can use it. Do they know formulas? Can they build a PivotTable? Are they fast?
That framing made sense when Excel was essentially a programmable calculator. It makes less sense now.
Modern Excel has accumulated its own forms of intelligence. It understands structure. It remembers procedures. It reasons probabilistically through AI assistance. And, just as importantly, it does not do certain things on its own.
If you want to work effectively with Excel today, it helps to stop asking, “How advanced is the user?” and start asking a different question:
What level of intelligence is Excel itself operating at in this workflow?
This post is an attempt to name those levels.
Excel as an intelligent application, not a neutral tool
Excel has never been a blank slate. Even in its earliest versions, it embedded assumptions about how work should happen: values live in cells, formulas propagate, recalculation is deterministic.
Over time, Microsoft layered additional forms of intelligence onto that foundation. Some of those layers are obvious. Others are so normalized that people stop noticing them. But together, they form a hierarchy.
Each level does not replace the one beneath it. Instead, Excel becomes capable of more kinds of reasoning, while still relying on the simpler logic underneath.
Seeing Excel this way makes its behavior feel less arbitrary and its limitations feel more intentional.
A hierarchy of Excel intelligence
At the base of the hierarchy is the form of intelligence Excel has always had: arithmetic execution. Excel calculates exactly what it is told, using explicit formulas that reference explicit cells. There is no interpretation here, no understanding of meaning. A formula either works or it doesn’t.
This level still matters. Every higher capability Excel has ultimately compiles down to this deterministic core. When something breaks, it is usually because the assumptions of higher levels have collapsed back into raw cell logic.
Once structure is introduced, Excel crosses an important threshold. Tables, headers, and data types allow Excel to treat ranges as collections rather than accidents of layout. At this point, Excel begins enforcing consistency on your behalf. Columns stay aligned. Formulas propagate predictably. References become semantic instead of positional.
This structural intelligence is often underestimated, but it is foundational. Without it, nothing else scales. AI cannot rescue unstructured data, because Excel itself cannot reason about chaos.
From there, Excel’s intelligence becomes analytical. PivotTables, relationships, and the data model allow Excel to summarize, group, and filter without rewriting logic each time. Excel starts answering questions like “by category,” “over time,” or “compared to last period.” This is no longer just computation; it is pattern-based reasoning over structured information.
The next shift is subtler but arguably more powerful. With Power Query and refreshable connections, Excel gains procedural memory. It doesn’t just hold results; it remembers how results were produced. Data lineage becomes explicit. Transformations become repeatable. Refresh replaces rework.
At this point, Excel is not just analyzing data. It is maintaining a process.
Finally, at the top of the hierarchy, Excel becomes assistive. With Copilot and Python in Excel, it can interpret natural language, draft logic, suggest analyses, and generate code. This is not intelligence in the human sense, but it is a form of probabilistic reasoning layered on top of everything below it.
Crucially, this layer does not replace the others. AI assistance depends on structure, analysis, and procedure already being in place. When those are missing, the AI feels unreliable. When they are present, it feels powerful.
Here’s a summary table that ties each layer to what Excel is actually doing, the core tools that express that intelligence, and the kinds of work it’s best suited for.
| Layer (low to high) | What Excel knows how to do | Representative tools/features | Typical use cases (what this layer is for) |
|---|---|---|---|
| Arithmetic execution | Execute explicit instructions deterministically | Formulas (SUM, IF, XLOOKUP), cell references, calc engine, named ranges | Tax/commission calculations, basic business rules, sanity checks, deterministic model logic (“if this, then that”) |
| Structural intelligence | Understand shape and enforce consistency | Excel Tables, structured references, data types, data validation, conditional formatting rules | Reliable inputs, preventing broken formulas, standardized datasets |
| Analytical intelligence | Aggregate, group, filter, and relate data | PivotTables/PivotCharts, slicers, Power Pivot data model, relationships, measures (DAX) | KPI reporting, cohort summaries, trend breakdowns, cross-tab analysis, multi-table analysis without fragile formulas |
| Procedural intelligence | Remember and reproduce how results are produced | Power Query (M), refreshable connections, parameters | Repeatable ETL, monthly close prep, cleaning messy exports, reshaping data for downstream analysis, building refreshable pipelines |
| Assisted reasoning | Accelerate exploration and drafting (probabilistic assistance) | Copilot in Excel, Python in Excel, “Analyze Data”/insights-style features, Agent Mode | First-pass analysis plans, quick visual prototypes, code/formula drafting, explanation and documentation, “help me explore this dataset” workflows |
How this hierarchy fits into the modern Excel AI stack
In a previous post, I described the modern Excel AI stack as something broader than Excel alone. Tools like Dataverse, Power Query and Dataflows, Power Automate, Copilot Studio, and Power BI form an ecosystem that Excel increasingly sits inside of, rather than standing apart from:
This hierarchy doesn’t contradict that view. It complements it.
The AI stack is a horizontal view: how data, automation, and AI services connect across the Microsoft platform. This hierarchy is a vertical view: what kind of intelligence Excel itself is expressing at a given moment, regardless of what it’s connected to.
Seen this way, Excel’s role in the stack becomes much clearer.
Excel is not trying to replace Dataverse as a system of record, Power Automate as an orchestration engine, or Copilot Studio as a place to design persistent agents. Instead, Excel concentrates intelligence where it has always been strongest: at the point where humans inspect data, test assumptions, and make decisions.
That’s why the pyramid matters. As Excel climbs the hierarchy from arithmetic to assistance it becomes a more capable decision environment, not a background automation service.
In other words, the AI stack gives Excel reach, while the intelligence hierarchy explains Excel’s judgment.
Why this distinction matters in real workflows
Most Excel problems that show up in organizations are not caused by a lack of AI. They’re caused by confusion about what Excel is being asked to do.
Teams try to:
- automate before they’ve structured data
- analyze before they’ve stabilized inputs
- add Copilot before they’ve clarified logic
- treat Excel like an orchestrator instead of a decision surface
When that happens, Excel feels fragile. AI feels unreliable. Automation feels risky. This hierarchy gives you a diagnostic tool.
If something breaks, the question isn’t “Do we need more AI?” It’s “Which layer is missing or overloaded?”
If Copilot’s answers don’t make sense, it’s often because the workbook never made it past structural or analytical intelligence. If a report takes hours to refresh, the problem is usually procedural, not computational. If an automated process keeps failing silently, it’s probably doing work that belongs in Power Automate or Copilot Studio, not in Excel at all.
Understanding where Excel should stop is just as important as knowing how far it can go.
Agent Mode, revisited through the pyramid
This is also why Agent Mode fits naturally at the top of the hierarchy, rather than sitting off to the side as something fundamentally new.
Agent Mode can build a strong first-pass model. It can reason through data, draft formulas, validate results, and iterate under human supervision. That part of the work feels impressive, but any experienced analyst knows it’s often the easy part.
What’s harder is everything that comes after: keeping models up to date, bringing in new data, handling changes in structure, pushing results to other systems, and preserving enough context for the work to survive beyond the current session.
Agent Mode doesn’t do this well, and Excel on its own isn’t designed to either. It doesn’t persist intent over time, run independently in the background, or manage long-lived workflows. That’s by design.
This is where the broader Excel AI stack comes in. Power Query, Dataflows, Dataverse, Office Scripts, Power Automate, and Copilot Studio exist to handle persistence, orchestration, and execution across systems, allowing Excel to stay focused on visible logic and decision-making.
Seen through the pyramid, this isn’t a limitation. It’s coherence. Excel remains the place where assumptions are testable and accountability is clear, even as it plugs into a much larger AI and automation ecosystem.
The takeaway: stop asking “How advanced is the user?”
The most important shift this hierarchy encourages is a conceptual one. Instead of asking “How advanced is the person using Excel,” start asking: “what level of intelligence is Excel operating at in this workflow, and is that the right level for the job?”
That question leads to better workbook design, more reliable automation, and far more productive use of AI. It also makes it much easier to explain Excel’s value to stakeholders who assume that “AI” means replacing spreadsheets rather than strengthening them.
If you want to apply this hierarchy, start with one workbook you already rely on. Ask which layer of intelligence it’s really operating at today, and whether that’s the right level for the decisions it’s meant to support. Most Excel pain shows up when those two don’t match.
And if you’re responsible for analytics workflows or Excel standards on a team and want a second set of eyes, I work with organizations on exactly this kind of diagnostic. You can book time with me here: