Skip to content

No account required

Start a project

Field guide

How spreadsheets go wrong.

And how to tell, before the number reaches anybody who will act on it. Every one of these is useful whether or not you ever use this product.
Guides
27
Each answers
one question
Tool required
none

Guide

Why two people get different totals from the same spreadsheet.

It is almost never arithmetic. It is nearly always one of four structural problems, and each leaves a signature you can look for.

6 min read

Guide

How to tell a real finding from a coincidence.

Most findings that fall apart do so for one of four reasons, and all four can be checked in minutes.

7 min read

Guide

Five things to check before you trust a dashboard.

A dashboard is a set of decisions someone made and then hid. These are the five worth surfacing before you act on one.

5 min read

Guide

Finding duplicates that Remove Duplicates will not find.

The built-in tool compares cells. Real duplicates disagree in every field, which is exactly why they survived.

6 min read

Guide

Why Excel changes your numbers without telling you.

Four conversions happen on open, none of them warns you, and all four are irreversible once you save.

7 min read

Guide

Comparing two versions of a spreadsheet without drowning in false changes.

Three things break a file comparison, and all three make it report far more changes than happened, or far fewer.

6 min read

Guide

Dates in more than one format, and why it is worse than it looks.

A date column in two formats silently misplaces rows in every period comparison, and about a third of them cannot be recovered at all.

6 min read

Guide

Why your CSV opens as one column, and what that means.

The most common CSV complaint has nothing to do with the file, and the second most common one is invisible until somebody searches for a customer.

5 min read

Guide

Numbers stored as text, and the total that is quietly short.

The most dangerous data problem is the one whose symptom is a number that looks about right.

5 min read

Guide

Joining two spreadsheets, and the match rate nobody reports.

The join usually works. What goes wrong is that nobody checks how much of it worked.

6 min read

Guide

When the average describes nobody.

The mean is the right answer to a narrower question than most people are asking when they compute one.

5 min read

Guide

Six ways a chart misleads without containing a false number.

Almost every misleading chart contains only true numbers. The misleading part is the encoding.

6 min read

Guide

The data dictionary, and why inheriting a file without one costs a week.

The cheapest document in analytics to produce, and the one whose absence causes the most repeated work.

5 min read

Guide

Analysing survey results without measuring who answered.

Sample size does not fix a systematic absence, and it is the absence that breaks most survey conclusions.

7 min read

Guide

Why two dashboards give different numbers for the same thing.

Two people are usually both right about a number, and disagreeing about which number it is.

5 min read

Guide

Analysing survey results in Excel when half the answers are words.

A survey export mixes ratings, tick-boxes and sentences in one grid, and each of the three needs different handling before any of them can be averaged.

6 min read

Guide

Cleaning a messy spreadsheet in the order that saves the most work.

Order matters more than effort. Each repair done out of sequence has to be done again after the one it depended on.

6 min read

Guide

Turning a Power BI report into a presentation without screenshotting it.

A dashboard is built for monitoring and a meeting needs an argument. The move between them starts from the data model, not from the screen.

6 min read

Guide

The messy-file test: a reproducible way to check whether a data tool gets the numbers right

Most AI data tool reviews compare speed, price and charts. This one tests whether the numbers are correct.

9 min read

Guide

We compute every figure twice. Here is what happens when the two answers differ.

A language model may decide what to compute. It never decides what the answer is.

8 min read

Guide

Eight things that are actually wrong with your messy CSV

A CSV is not messy in general. It is usually broken in one of eight recognizable ways.

9 min read

Guide

A dashboard is not a set of charts

A dashboard is not a set of charts. It is a small number of decisions, each backed by a number someone can verify.

7 min read

Guide

A deck built from a spreadsheet has to survive being questioned

A data deck succeeds in the meeting after the one where it was presented, when somebody asks where a number came from.

7 min read

Guide

The famous spreadsheet errors were not arithmetic errors

The famous failures were not bad maths. They were correct calculations performed on the wrong rows, ranges or types.

9 min read

Guide

Julius AI alternatives, and the four questions that actually decide between them

Nearly every comparison is written by one of the products on it. This one says so, and includes the cases where the competitor is the better choice.

9 min read

Guide

A close that balances is not the same as a close that is right

A close that balances can still be wrong. These checks catch the structural failures a trial balance cannot see.

8 min read

Guide

The question is not whether the model is clever. It is whether it is allowed to decide the answer.

The important question is not whether the model is clever. It is whether the model is allowed to decide the answer.

9 min read