When Excel stops being the right tool

The real trigger is the workbook’s role, not its file size

A large workbook can be appropriate, and a small workbook can be dangerous. File size, row count, and calculation time matter because they can reveal strain, but they do not measure business risk. The better question is what happens if the workbook is wrong, unavailable, changed without approval, or understood by only one person.

I would treat a workbook as business-critical when it materially affects money, customers, employees, operations, reporting, or compliance. Once it crosses that line, it needs an owner, a backup owner, a recovery plan, documented inputs and outputs, and a scheduled review—even if it opens quickly today.

• Does the business treat the workbook as the official record?
• Can several people or automations change it at the same time?
• Does it assign work, collect approvals, or track process status?
• Does it move data between systems or hold the only copy of important logic?
• Would an error affect payroll, billing, inventory, customers, compliance, or management reporting?
• Could the process continue if the workbook owner were unavailable tomorrow?

Employee count changes the controls—not the basic fit

A 12-person company can have a high-risk payroll workbook. A 400-person company can have a perfectly reasonable one-person analysis. Headcount is useful context because growth usually adds editors, handoffs, departments, reporting expectations, and security requirements. It should change how formal the review becomes, not decide the answer by itself.

• 1–25 employees: name the owner and backup, use controlled storage, document the purpose, and set a review date.
• 26–100 employees: add access groups, change notes, source documentation, recovery tests, and a quarterly review for critical workbooks.
• 101–250 employees: keep an inventory of critical workbooks, assign risk tiers, separate data from presentation, and require independent checks for material models.
• 251–500 employees: define architecture standards, service expectations, audit evidence, support ownership, and funded modernization paths for high-risk workbooks.

Good uses for Excel

• Exploratory analysis where the questions and calculations are still changing.
• Forecasts, scenarios, and financial models with documented assumptions and independent review.
• Reconciliations and exception analysis using exports from trusted systems.
• Ad hoc and last-mile reporting where Excel presents data but is not the master record.
• Small, bounded lists with one clear owner, limited editing, and no complex approval process.
• Prototypes used to learn the requirements before committing to a longer-term solution.

Bad uses for Excel

• The only system of record for customers, orders, inventory, cases, payroll, or compliance evidence.
• A shared operational queue where several people assign work, change status, and approve outcomes.
• An integration hub that imports, transforms, and redistributes data between several systems.
• A security boundary that depends on hiding tabs, locking cells, or emailing different copies.
• The only home for critical macros, pricing rules, allocation logic, or institutional knowledge.
• A high-concurrency file that people, scripts, and scheduled automations may write to simultaneously.

What public spreadsheet failures actually teach

In 2020, Public Health England’s use of an older Excel file format contributed to 15,841 positive COVID-19 cases being omitted from reporting. The incident is often reduced to an Excel row-limit story. The more useful lesson is that a consequential data pipeline reached a file-format boundary without a control that caught the missing records before publication.

In the SEC’s 2013 JPMorgan settlement, spreadsheet miscalculations appeared within a broader failure of valuation and internal controls. Excel did not single-handedly cause the losses. Material models can exist in spreadsheets, but they need proportionate assurance: clear ownership, input controls, version management, independent checks, documented changes, and a way to reproduce the result.

Review immediately when one of these conditions is true

• The workbook cannot open or complete its normal work on standard managed equipment.
• It has produced a materially wrong result or silently omitted data.
• A core business process cannot continue when the workbook or its owner is unavailable.
• People and automations may write to the same workbook at the same time.
• There is no trained backup owner, tested recovery copy, or documented manual fallback.
• Sensitive data cannot be restricted, retained, or audited at the level the business requires.

Schedule a review when the operating baseline starts to slip

Numbers help a team act before frustration becomes failure. I recommend using the following as starting thresholds, then tightening them for faster or more consequential processes. They are screening criteria, not guarantees and not Microsoft’s technical limits.

• A routine user action makes someone wait more than 30 seconds.
• An attended refresh, recalculation, open, save, or close takes more than five minutes.
• There are two crashes, sync conflicts, failed refreshes, repair prompts, or support incidents within 90 days.
• The team spends more than two hours a month repairing, reconciling, rerunning, or working around the file.
• The workbook has more than one external data connection, more than one automation writer, or dependencies nobody can fully explain.
• A new department, location, customer group, or regulatory obligation will depend on it.

Where Excel belongs in the automation and data landscape

A healthy pattern is source systems → governed data and business rules → workflows and integrations → reporting and analysis → Excel for exploration, controlled modeling, review, or export. Excel can be an excellent interface at the edge of that landscape. It is a fragile center when it must simultaneously store the master data, enforce the workflow, integrate systems, calculate results, and prove who changed what.

Keep authoritative records in a durable system of record. Put repeatable transformations in a governed data pipeline. Run assignments, approvals, reminders, and exception routes in a workflow tool. Use reporting tools for shared metrics. Let Excel consume trusted data and return bounded, validated inputs when that is the best experience for the people doing the work.

Choose the next category by the job the workbook is doing

• Shared records with relationships, validation, and controlled concurrent edits: consider a database-backed application or managed work-management platform.
• Assignments, approvals, reminders, handoffs, and escalation: consider workflow or case-management software.
• Scheduled imports, transformations, and movement between systems: consider an integration or data-pipeline platform.
• Standard dashboards and shared metrics: consider a reporting or business-intelligence platform connected to governed data.
• Complex planning and modeling: keep Excel as the modeling surface, but move stable inputs, outputs, and history into governed services where practical.

Do not replace a critical workbook in one leap

The safest modernization is usually staged. Stabilize the current process first, then separate the workbook’s different jobs. A workbook may be a database, calculator, workflow, report, and user interface all at once. Those parts do not have to move on the same day—or to the same product.

• Name the business owner, technical owner, purpose, users, inputs, outputs, and downstream decisions.
• Make a tested backup and record the current performance and failure baseline.
• Inventory macros, queries, links, connections, scheduled tasks, formulas, manual steps, and hidden dependencies.
• Decide which record is authoritative and remove duplicate or conflicting sources.
• Separate data storage, business rules, workflow, reporting, and user-interface responsibilities.
• Run the old and new paths in parallel for a defined period and reconcile material outputs.
• Train the primary and backup owners, document exception handling, and test recovery.
• Retire the old path deliberately, preserving the records and evidence the business must retain.