From Excel to a decision system
Excel is a calculation tool. A decision system is a structure that records how a decision was made, which data supported it, who approved it, and what the result was. When the two are confused, a company entrusts its most critical decisions, pricing, discounts, inventory commitments, and budgets, to a file without an audit trail, dependent on one person’s knowledge, and containing errors that may remain invisible. For FMCG, retail, and mid-sized companies, this is a quiet but structural risk.
The problem is not Excel itself, but the weight placed on it. A spreadsheet is excellent for a quick calculation. Over time, however, the same file may become the company’s price list, budget, and single source of truth. No one made a deliberate architecture decision; the file simply grew. When a tool takes over the role of a decision system, it provides none of the safeguards that role requires.
In brief
- People make errors at a rate of approximately 3–5% in simple but non-trivial tasks such as cell entry, and those errors accumulate through formula chains. In Panko’s calculation, with 95% accuracy per formula, a spreadsheet containing 100 root formulas has a 99.4% probability of containing at least one error (Panko, EuSpRIG).
- Field audits find real errors in the great majority of spreadsheets. In the most recent audits using the strongest methodologies, the rate was at least 86% (Panko, “Spreadsheet Errors: What We Know”).
- The largest risk is not the error itself, but overconfidence. In one experiment, developers’ median estimate of the probability that they had made an error was 10%, while 86% had actually made one (same study).
- The solution is not to ban Excel, but to remove critical decisions from it and provide a single source of truth, validation, version history, role-based access, and an audit trail.
Why is Excel not a decision system?
Excel’s limitation is not its calculation capability, but its lack of decision infrastructure. A decision system requires four things: a single source of truth, input validation, an audit trail showing who changed what and when, and role-based access. Excel does not provide these structurally. As one enterprise analysis summarizes, spreadsheets lack built-in error detection, automatic audit trails, and safeguards preventing one person’s formula change from propagating through an entire reporting chain.
The deeper problem is duplication. Finance pulls the same data into one file, operations into another, and different filters and calculation logic produce two different versions of reality. Excel does not resolve this problem; it amplifies it. When a critical decision is made in that environment, no one can prove that the underlying number was correct.
Are spreadsheet error rates really that high?
Yes, and the cause is not simply carelessness, but human cognition. According to Panko’s research spanning several decades, people make errors at a rate of approximately 3–5% in simple but non-trivial tasks such as entering cells, writing text, or producing lines of code. The error rate appears small per cell, but spreadsheets contain thousands of cells and interconnected formula chains. The probability accumulates rapidly. With 95% accuracy per formula, a spreadsheet containing 100 root formulas has a 99.4% probability of containing at least one error.
Field audits support this conclusion. In audits using the newest and strongest methodologies, at least 86% of the spreadsheets examined contained real errors. This is not merely an academic curiosity. In 2010, two Harvard economists, Reinhart and Rogoff, published influential work arguing that high public debt reduces economic growth. A 2013 review by Herndon, Ash, and Pollin challenged the analysis partly because an Excel formula omitted five countries, Denmark, Canada, Belgium, Austria, and Australia. For completeness, the criticism was not limited to the spreadsheet mistake; it also addressed data selection and weighting, and subsequent literature continued to debate the debt–growth relationship. One point remains indisputable: the figures behind one of the world’s most debated macroeconomic studies had been produced in an unaudited spreadsheet.
Is the main risk the error itself, or overconfidence?
The greater risk is not only the error, but unjustified confidence that the spreadsheet is correct. In an experiment by Panko and Featherman, developers were asked to estimate the probability that their spreadsheets contained an error. The median estimate was 10%. In reality, 86% had made an error. When participants were asked to raise their hands if they believed they belonged to the error-free 14%, more than half did so.
This is the most dangerous combination: high error probability, weak error-detection ability, and high confidence. People are poor at finding their own mistakes, and the control formulas added to a spreadsheet have their own error rate. “Be more careful” or “add a check row” is therefore not a structural solution. The problem is not individual diligence, but the system.
What does a decision system do differently?
A decision system may perform the same calculation, but adds the safeguards that the decision requires. A number in Excel is “believed to be correct.” In a decision system, the number’s origin and validity can be traced.
| Dimension | Excel, the tool | Decision system, the structure |
|---|---|---|
| Source of truth | Duplicated files and names such as “v2_final” | One central source |
| Input | Free-form and unvalidated | Rule-based and validated |
| Error | Silent and difficult to detect | Detected through controls and alerts |
| Change | Who changed what, when, and why is unclear | Version history and audit trail |
| Access | Anyone who can open the file | Role-based permissions |
| Decision | Disappears inside the file | Recorded with inputs, approval, and outcome |
| Knowledge | Lives in one person’s head | Embedded in a transferable structure |
This transition does not require a huge enterprise platform. It means moving a critical decision out of Excel and into a structure with validation, version history, role-based access, and an audit trail. The same logic applies to price approval: when the decision becomes recorded and traceable on a screen, the organization can also learn from its outcomes.
Which spreadsheets are risky, and which are not?
Moving every spreadsheet into a system would be unnecessary and expensive. Simple calculations, one-off analyses, and short lists are low risk. Excel is excellent for these tasks and has no obvious replacement. The risky files share a common profile: they act as the single source of truth for something important, pricing, payroll, inventory, or customer lists; decisions are made from them; and they are usually maintained by one person.
Another point of honesty is necessary. The widely repeated statement that “88% of spreadsheets contain errors” is not one universal and fixed truth. In the field audits compiled by Panko, older studies using weaker methods found lower rates, while newer audits with stronger methodologies found at least 86%. The exact figure varies with methodology, and relying on one number too confidently would repeat the very error under discussion. Three conclusions remain consistent: errors are common, difficult to detect, and organizations are overconfident in spreadsheet accuracy. Decisions should be based on those three facts.
Conclusion
Excel is a calculation tool, not a decision system. Small and normal human errors at cell level accumulate through formula chains. Field audits find real errors in the great majority of spreadsheets, and most dangerously, organizations trust these files more than they should. When a critical decision, such as a price, discount, or inventory commitment, is made inside a file without a single source of truth, validation, or audit trail, no one can defend the number on which the decision depended.
The most practical conclusion from Panko’s research is that “be more careful” is not a solution because the problem is structural rather than attentional. People are poor at finding their own errors and do not know how many they have missed. The solution must also be structural: determine which spreadsheet is merely a calculation and which has effectively become a decision system. The latter should be moved into a structure with validation, version history, role-based access, and an audit trail.
How does GDP build it?
- Which spreadsheet we select: We identify a file that supports a decision, acts as a single source of truth, and depends heavily on one person.
- What we rebuild: We extract the business logic from the spreadsheet and rebuild it with validation, version history, and role-based access.
- What we record: We make the decision’s inputs, approver, and outcome traceable so the organization can learn from the result later.
Excel remains a calculation tool. Through our Decision Intelligence approach, we begin this transition lightly around one critical decision. We explain why the replacement screen should be decision-focused rather than merely “beautiful” in The beautiful-dashboard trap in management reporting.
Frequently asked questions
Why is Excel not a decision system?
Because it does not structurally provide four things a decision system needs: a single source of truth, input validation, an audit trail showing who changed what and when, and role-based access. Excel performs the calculation but does not preserve how the decision was made, what it relied on, or who approved it. As files multiply, multiple versions of “truth” appear.
Are Excel error rates really high?
According to Panko’s research, people make errors at a rate of approximately 3–5% in tasks such as cell entry. That small per-cell rate accumulates through interconnected formula chains. At 95% accuracy per formula, a spreadsheet containing 100 root formulas has a 99.4% probability of at least one error. In the strongest field audits, at least 86% of spreadsheets contained real errors.
Is the greatest risk the error itself?
No, the greatest risk is overconfidence. In one experiment, developers’ median estimate of their chance of having made an error was 10%, while 86% had actually made one. People are poor at detecting their own mistakes, and the control formulas they add can also contain errors. “Be more careful” is therefore not a solution; the problem is structural.
Which Excel files should be considered risky?
Simple calculations, one-off analyses, and short lists are low risk. Dangerous files share three characteristics: they are the single source of truth for something important, pricing, payroll, inventory, or customer data; decisions are made from them; and one person usually maintains them. When that person leaves, the only complete understanding of the file may leave as well.
Is moving from Excel to a decision system necessarily a large project?
No. Moving every spreadsheet would be unnecessary. The right starting point is one critical decision: move it out of Excel and provide a single source of truth, input validation, version history, role-based access, and an audit trail. The scope can expand after value is proven. Excel continues to serve as a calculation tool.
Sources
Academic: Raymond Panko, University of Hawaii, “Spreadsheet Errors: What We Know” and EuSpRIG 2015 paper. Cell error rates of 3–5%; accumulation through formula chains, including a 99.4% probability of at least one error in a spreadsheet with 100 root formulas at 95% formula accuracy; at least 86% error incidence in the strongest field audits; and the Panko & Featherman experiment in which the median estimated error probability was 10% while the actual rate was 86%.
Academic: Thomas Herndon, Michael Ash, and Robert Pollin, “Does High Public Debt Consistently Stifle Economic Growth? A Critique of Reinhart and Rogoff” (Cambridge Journal of Economics, 2013). The Excel formula error omitting five countries from the Reinhart–Rogoff 2010 analysis, alongside two additional methodological criticisms.
Industry: QuickLaunch Analytics on Excel’s lack of audit trails, error detection, and a single source of truth; Allen App Tools on the profile of high-risk spreadsheets, single source of truth, decision dependence, and reliance on one person.
Last reviewed: July 2026.