Migration · spreadsheets

The workbook is not the problem

Excel is excellent and the workbook is usually correct. What makes it a risk is that a critical, complex, monthly calculation ended up somewhere with no version control, no tests, no audit trail, no access control, and exactly one person who can maintain it.

Which workbook?

Tell us what it produces and how long it takes each month. That is usually enough to scope it.

1 / 3
Back-tested against 24 monthsThe expert stays centralParallel until it agrees twice

Stay if

Three cases where the spreadsheet should stay.

Not everything in Excel is a liability. These are the cases where replacing a workbook costs more than it returns, and we would say so before quoting.

It is genuine ad-hoc analysis

Exploring a question, modelling a scenario, testing a hypothesis. Excel is the right tool for that and always will be. What gets replaced is the monthly production run, not the ability to explore.

It runs twice a year

A calculation performed occasionally, by one person, with low stakes, rarely justifies a build and its ongoing maintenance. Better documentation is the cheaper answer.

The rules are still contested

If four people describe the calculation differently, building it encodes one of those descriptions and makes the disagreement permanent. Settle the rules first; that is a cheaper engagement.

Real reasons

Three that hold up under examination.

One person understands it

The single most common reason and the most legitimate. A board-relied-upon number with one maintainer who has never taken two consecutive weeks off during a close is a real business risk.

An auditor has started asking

Revenue schedules, consolidation, and allocation workbooks are the three auditors probe hardest, and the questions get harder each year as the amounts grow.

It has stopped scaling

The consolidation that took two hours at four entities takes two days at eleven. Manual workbooks scale linearly with complexity while the close window does not.

What the back-test usually finds

On a typical engagement, running the replacement against twenty-four months of historical inputs produces differences in several months — and more of them turn out to be spreadsheet errors than replacement errors. That is uncomfortable, it is why the back-test exists, and it is worth deciding in advance how you want prior periods handled if it happens.

Extractread source, no writesMapaccounts, customers, vendorsLoadinto a staged tenantReconciletrial balance, per periodgate · must tieReviewyour controller signsCut oversource goes read-onlygate · must tievariance → back to mapping, never waivednothing advances past a gate until the trial balance agrees to the penny

Extraction is archaeology

The logic in a mature workbook is rarely documented and frequently not what anybody believes it to be. There are hard-coded adjustments from a prior year, exceptions for a customer who left, formulas that differ in one column of one tab, and at least one number nobody can explain that everyone works around.

Finding those requires reading the workbook cell by cell with its owner, and it takes longer than the build. It is also frequently the most valuable part of the engagement, because several of the discoveries are errors that have been running for years.

Reading the workbook takes longer than rebuilding it, and it is where the value is. Several of the things you find will be wrong rather than merely undocumented.

The workbooks we replace most often

Multi-entity consolidation, with intercompany eliminated from memory and currency applied at a rate somebody looked up. Revenue schedules interpreting contract terms into deferrals. Commission calculations with tiers, splits, and clawbacks that nobody outside finance can verify. Overhead allocation formulas agreed years ago that four people now describe differently.

Each of these has the same profile: high stakes, high complexity, monthly, one owner. That profile is the qualifier rather than the subject matter.

Tested against its own history

The replacement is validated by running it against twenty-four months of historical inputs and comparing to what the spreadsheet produced. Every difference is investigated before launch and resolved into one of three categories: the replacement is wrong, the spreadsheet was wrong, or a rule changed and neither is wrong.

All three occur. That test is not optional in our methodology, and a replacement that has not been back-tested is a replacement nobody should trust with a number the board reads.

Keep the expert

The person who owns the workbook is the domain expert, and replacement projects that treat them as an obstacle fail at the edge cases. They know why the odd adjustment is there and they are the only reliable validator of the result.

What changes is what they spend time on. Rebuilding the mechanics monthly becomes reviewing exceptions and explaining variances, which is the work they were hired for and usually the work they would rather be doing.

Parallel until it agrees twice

Both run side by side for two full cycles. The spreadsheet is retired when it has agreed twice, not when a date in a plan arrives. We have extended parallel periods on several engagements and have never regretted it.

Questions

What spreadsheets customers ask.

Do we lose Excel entirely?
No. Ad-hoc analysis, scenario modelling, and exploration stay in Excel, which is the right tool for them. What gets replaced is the monthly production run.
What if the spreadsheet turns out to be wrong?
The back-test finds it, and it happens more often than the reverse. We tell you what we found and you decide how to handle prior periods.
How long does it take?
Three to six weeks for most workbooks, of which reading the existing one is roughly a third. Then two full parallel cycles before retirement.
What happens to the person who owns it?
They stay central — they are the domain expert and the only reliable validator. Their time moves from rebuilding mechanics to reviewing exceptions.
What does it cost?
$10,000 to $28,000 depending on complexity. The reading phase is the variable part, because a fifteen-year-old workbook contains more surprises than a three-year-old one.

Tell us about the workbook.

The one everyone depends on and one person understands. That is the one worth replacing first.