Why the grid fits how we think

Why are spreadsheets so popular? Ask anyone in finance and you’ll hear one word: flexible. Researchers in human-computer interaction (HCI), the field that studies how people work with software, have spent more than forty years taking that word apart. What they found reads like a description of how an FP&A team builds a budget.

This page is part of the research behind In defence of Excel, the first post in our series Spreadsheets: love & limits. It gathers the evidence: the classic studies, how people learn and share spreadsheet skills, why professionals model in a grid, and what research since 2010 has added. Each finding names something we do every day without thinking, and a strength is easier to protect once it has a name.

Why are spreadsheets so popular?

The most influential answer comes from Nardi and Miller’s fieldwork around 1990: long interviews with eleven spreadsheet users, several of them in finance (a CFO, a controller, a finance-department manager and accountants). “Spreadsheets succeed”, they concluded, “because they combine an expressive high level programming language with a powerful visual format to organize and display data.” Their users were not failed programmers but “business professionals or scientists or other kinds of domain specialists whose jobs involve computational tasks”. An FP&A analyst, in other words: a finance expert who computes.

They also watched people with no particular interest in computers build spreadsheets of their own accord, drawn by the “twinkling lights” of cells that update themselves. Read side by side, the studies since then agree on a short list of design properties.

Why does a spreadsheet feel so direct?

In 1983 Ben Shneiderman coined the term “direct manipulation” for software in which you work on the thing itself, through “Rapid, incremental, reversible operations whose impact on the object of interest is immediately visible”. His showcase was VisiCalc, and his example could be any sales model today: costs, sales, commissions and profits by district and month, “so that the impact of changes on profits is immediately apparent”. He also described what this does to the people using it:

“Users experience less anxiety because the system is comprehensible and because actions are so easily reversible. Users gain confidence and mastery because they initiate an action, feel in control, and can predict system responses.”

So the sense of control an analyst feels over “my model” isn’t vanity. It’s what the design produces. Hutchins, Hollan and Norman added in 1985 that spreadsheet programs “have been valuable, in part because their output format continually shows the state of the system as values are changed”. You never have to ask where the forecast stands. It’s on the screen.

Alan Kay called the spreadsheet “a simulation kit” in 1984, with “few mystifying surprises because the only way a cell can get a value is by having the cell’s own value rule put it there”. Every number has one visible source. And you generalise by example: write January’s formula against January’s numbers, check it, and drag it to December. Nobody writes “for each month”.

Weak on abstractions, cheap to start

Thomas Green and colleagues built a framework for judging how usable a notation is, the cognitive dimensions of notations, set out with Petre in 1996. It asks, among other things, whether you must make decisions before you have the information you need (premature commitment) and whether you can check unfinished work as you go (progressive evaluation). Spreadsheets “recompute at every opportunity, so that the user can develop a solution bit by bit”, says Green and Blackwell’s 1998 tutorial, which calls the spreadsheet “an abstraction-hating system” and means it as praise:

“One reason why spreadsheets are so popular is that they are weak on abstractions. In fact, not surprisingly, many potential end-users are repelled by abstraction-hungry systems.”

Those systems, it explains, “suffer from a sort of delayed gratification, because the appropriate abstractions must be defined before the user’s immediate goals can be attacked”. Anyone asked to set up dimensions, hierarchies and versions before typing a single number knows the feeling.

Alan Blackwell’s attention-investment model explains why this matters. Most decisions to start something programming-like, he argued, rest on “an implicit cost-benefit analysis”: attention spent now against attention saved later, weighed by the risk that it never pays off. A formula in a cell costs seconds and shows its answer at once. In 2003 Simon Peyton Jones of Microsoft Research and two academic colleagues concluded that “the commercial success of spreadsheets is largely due to the fact that many people find them more usable than programming languages for programming-like tasks”. Debugging in Excel, they noted, “means looking at values that are continuously displayed in intermediate cells”. Eyeballing the subtotals is the spreadsheet’s debugger.

How do people learn spreadsheets?

Nardi and Miller’s best-known line is about motivation. The spreadsheet’s “single biggest advantage”, they wrote, “is not cognitive but motivational: after only a few hours of work, spreadsheet users are rewarded by simple but functioning programs that model their problems of interest”. The design makes this possible. Its “fundamental layer”, the formulas and the table, is “completely self-contained and independent from the advanced layer of more sophisticated features”. You can run a department budget without touching a macro, and pick up the rest when a task calls for it.

One of their informants held “an accounting position of considerable responsibility” and had not learned to iterate over a range. “Despite what would be a fatal gap in her knowledge in a traditional programming language,” the authors wrote, “Jennifer is a successful spreadsheet user.” Learning pivot tables one year and dynamic arrays the next isn’t a failing. It’s how the tool is learned.

Learned from colleagues, not courses

Formal training is rare. In a survey of 1,597 business spreadsheet users published in 2006, the commonest ways of learning were books and manuals (53.6%) and colleagues’ demonstrations (52.3%), and only 37.7% had ever had classroom instruction. In 2018 Sarkar and Gordon interviewed seven people aged 35 to 65, among them accountants and a budget manager, and found that “feature adoption in spreadsheets is informal, opportunistic, and social”. People learn a function when a problem needs it and ask colleagues, who “see themselves as ‘helping’ and not ‘teaching’”. One participant, describing a finance firm, called it “almost an apprenticeship, somewhat medieval in its nature”.

They also saw attention investment at work. People adopt a feature only when the reward seems worth the attention, and a professional accountant explained why many stop at formulas:

“automation to me is a very fine double-edged sword. You can achieve efficiencies but it is significant cost initially and if the world changes an inch, you can negate any investment. So a lot of my colleagues never automate beyond formulae … they’ll pivot table and they’ll refresh but they won’t do a lot of macros”

Anyone who has rebuilt a report after a reorganisation will recognise the reasoning, and the grid never forces anyone past formulas. That is a large part of why the barrier to entry is so low.

Expertise spreads through the files

Nardi and Miller found that “Spreadsheet co-development is the rule, not the exception”. They named three roles, “non-programmers, local developers and programmers”, where local developers are the colleagues with deeper skills who “typically serve as consultants for non-programmers”. Every finance team has its Excel person. Knowledge of the business travelled in templates: one CFO built the budget templates his controller and her staff filled in, with “a basic structure for data that he works out because of his greater knowledge of the business”. It works because the grid is legible:

“Users are able to understand and interpret each other’s models with relative ease because the tabular format of the spreadsheet presents such a clear depiction of the parameters and data values.”

Sarkar and Gordon called this percolation: “visible features percolate better”. One modeller had picked up SUMIFS because “I must have just seen that in a spreadsheet somewhere”. A 2024 survey of 100 users in administrative and finance roles found that people are more willing to share when they feel confident and expect credit for it, and less when writing things down feels like work. They also felt less sure of their general spreadsheet skills than of their skills in their own job. Most of us know someone who runs the whole forecast and still insists they aren’t really an Excel expert.

Why does finance use Excel for modelling?

Most models start as a question, not a specification. Many of Nardi and Miller’s users began with “only a general goal in mind”, such as “maximize profits over the next three quarters”, and found out what mattered “in the process of actually trying to solve it”. One, who built business-plan models for joint ventures, explained why he didn’t hand the job to someone else:

“I found it easier to develop that myself than to go to somebody and say here’s what I want, here’s what I want, here’s what I want … I felt that I was learning as I went … I was learning about all the variables that I needed to think about.”

A specification assumes you already know what the model should be. In modelling, building it is how you find out. The studies show the grid helping in four ways.

You see the whole problem. “Virtually every user in our study reported that an advantage of spreadsheets is the ability to view large quantities of data on one screen”, Nardi and Miller wrote. Twelve months by forty P&L lines on one screen is a thinking aid, not an accident of layout.

Restructuring is cheap. A manager who ran a finance department at a large corporation prized what he called “reconstructive surgery”: “I can really move things around, change the whole look and feel and objective of a spreadsheet very quickly.” Participants often noted that “there is no penalty for adding new things as they come to light”. A new cost line or scenario column is one insert away.

What-if is the core loop. Kay noted that most VisiCalc buyers “exploited it to forecast the future rather than to account for the past”. In a 2019 interview study of 11 users, published as a preprint, “A common comparison was a best-case and worst-case scenario”, and people saved scenarios next to each other because “the grid layout allows to put scenarios side by side”. Base, upside and downside in adjacent columns is the standard FP&A pattern for a reason.

The model is also the conversation. Nardi and Zarmer describe a CFO who “does not organize his thoughts until he has the spreadsheet artifact to look at, react to, and critique”. A controller and her CFO refined the annual budget estimates by passing the sheet back and forth, until “the model emerged in successive approximations”. And the finished model is the presentation, because in a spreadsheet “the development environment and the presentation environment are the same”.

Hendry and Green, who interviewed ten people who used spreadsheets by choice, summed up the bargain in 1994. The strengths “allow quick gratification of immediate needs”, while the weaknesses “make subsequent debugging and interpretation difficult”. The benefits arrive now and the costs later, when someone else inherits the file. For work due at Friday’s meeting, we’d call that a rational trade.

What research since 2010 confirms

The classic studies were done with VisiCalc, Lotus 1-2-3 and early Excel. Since 2010, researchers, many of them at Microsoft Research in Cambridge, have gone back to the same questions with new methods, and we found no recent study that contradicts the classic account. A study of more than 15,000 Enron spreadsheets found that “76% of spreadsheets in the presented corpus use the same 15 functions”, echoing Nardi and Miller’s “fewer than ten functions”.

The clearest confirmation came at CHI 2022, the main conference on human-computer interaction. Chalhoub and Sarkar interviewed 21 users and asked why “the flexible grid remains overwhelmingly successful” despite efforts to augment or replace it with formal data structures. Their answer: the grid starts “its abstraction gradient at zero” and “requires very little premature commitment”. One participant put it better: “It’s freedom to put things where my mind wants to put them.” Another drafts first and formalises later: “In the early drafting of [my spreadsheet], I liked just to add the column headers, add a bit of data in, and then later in the process, make it into a table.”

A study the same year followed 12 “data workers”, people who work with data without being professional analysts, and asked a question finance teams will recognise:

“Spreadsheets, the canonical table tools, pervade the data ecosystems of organizations, even when purpose-built business intelligence (BI) tools are widely available. Why? Are these organizations, and their people, simply unsophisticated in their data strategy? Our investigations … says emphatically: no!”

Tables gave their participants “a trusted window” onto their data, with “both precise control and a sense of confidence and ownership”. The authors’ most salient finding was “the sheer irreplaceability of the table form”.

The extensions mostly concern what happens after a model is built. In a 2021 study of 15 people reading others’ spreadsheets for work, participants spent around 40% of their time looking for information, and 12 of the 15 went back to the author, much as Nardi and Miller had observed in 1991. When Excel gained LAMBDA, which lets users write their own functions, an analysis of nearly 2,700 online comments found such abstractions “viewed both as helpful and harmful”, the trade-off Green and Blackwell described. And a 2026 study of AI agents for spreadsheets starts from the premise that “spreadsheet programmers tend to work iteratively”, though its 24 participants preferred the agent’s plan mode. How AI changes modelling is an open question.

What the evidence doesn’t show

Most of this evidence is qualitative, with 7 to 21 people per interview study. The big numbers come from a survey about training and from collections of files and comments. The “few hours” to a working model is an observation, not a measurement. The account’s weight comes from consistency: the same drivers across four decades and several methods, and a 2022 study, with 18 of its 21 interviewees aged 18 to 34, hearing the same words as the interviews of 1990.

We found no controlled experiment comparing spreadsheets with dedicated planning tools on FP&A what-if work, and no large study of how FP&A professionals learned Excel. Outside Nardi and Miller’s finance informants, few of the people studied work in FP&A at all. And these studies explain why the grid wins for exploring a problem and explaining it to others. They say little about planning across many entities, versions and contributors. That question belongs to the next post in the series, which looks at where Excel breaks.

The usual explanation for finance’s attachment to Excel is habit. The research describes something more precise: a tool that supplies the structure, shows every consequence at once, asks for nothing before the first number, and leaves the model with the person who understands the business. That is the case we make in In defence of Excel. For how many people program this way, see Is Excel a programming language?, and for whether the love is generational, Excel across generations.