Everyone wants the shiny part. The dashboard with the clean charts, the AI assistant that answers questions in plain English, the monthly report that lands on time and tells the real story. Nobody puts their hand up for the part that makes all of that possible, which is sitting with a messy spreadsheet and fixing the same customer name that has been spelled three different ways since 2019.
There is an old line in our world that never stops being true: garbage in, garbage out. The smartest tool on the market will still hand you a confident, beautifully formatted wrong answer if the data feeding it is a mess.
We see this constantly. A business invests in reporting or automation, switches it on, and the numbers do not match reality. The instinct is to blame the tool. Almost always, the tool is fine. The data underneath it is the problem, and that is where the unglamorous work lives.
What dirty data actually looks like
Dirty data is rarely one big dramatic error. It is a thousand small ones that add up quietly until your reports drift away from the truth. The usual suspects:
- Duplicates. The same customer captured three times because someone could not find the first record, so they made a new one.
- Typos. A misplaced zero turns R5,000 into R50,000, and suddenly one invoice skews your whole month.
- Inconsistent formats. Dates as 01/07, 1 July and 2025-07-01 all in the same column. Phone numbers with and without the country code. Cape Town, CPT and Capetown all meaning the same city.
- Missing values. Blank fields where a value should be, so totals quietly undercount and whole segments go missing.
- One thing spelled many ways. A single customer living in your system as ABC Trading, A.B.C. Trading and ABC Trading (Pty) Ltd, counted as three separate accounts.
How it quietly ruins your reports and your AI
The damage is sneaky because nothing breaks. The report still runs. The chart still renders. It just tells you something slightly untrue, and you make decisions on it without ever knowing.
You think your top customer is worth a certain amount, but their spend is split across three duplicate records, so they look like three medium customers and never get the attention they deserve. Your sales total looks healthy until that misplaced zero is found weeks later. Your AI tool, which learns from whatever you feed it, learns your bad habits and repeats them back to you with total confidence.
A report is a mirror of your data. If the mirror is dirty, you do not see a clearer business. You see a distorted one, and you trust it anyway.
How to clean what you already have
If your data is in a state, you do not fix it by feel. You work through it in a sensible order so you are not chasing your tail.
The order that works
- Agree your standards first. Decide how a date, a phone number, a province and a customer name should look. Write it down. Without a standard, cleaning is just personal preference.
- Deduplicate. Find and merge the records that are really the same thing. This is often the single biggest win.
- Validate against the standard. Flag anything that does not match the format you agreed, then fix it.
- Settle on one source of truth. Pick the system that owns each piece of data, so there is no more arguing about which copy is correct.
The "one customer, three names" test
Pick your biggest customer and search for them in your system. If they show up more than once, under slightly different spellings, you have found your starting point. Almost every business we work with fails this test on the first try, and fixing it usually improves the numbers more than any new tool would.
Prevention beats cure
Cleaning data is satisfying, but the real win is stopping the mess from forming again. Dirty data is mostly created at the point of entry, by people in a hurry, typing into a box that lets them type anything. Tighten that box and you fix the problem at the source.
- Use dropdowns instead of free text wherever there is a fixed list of options, so nobody invents a fourth spelling of a province.
- Make the fields that matter required, so records cannot be saved half empty.
- Design forms that check formats as people type, so a phone number or an email has to look right before it is accepted.
- Warn on likely duplicates before a new record is ever created.
It is ongoing, not a one-time project
Here is the part nobody likes to hear. Clean data is not a job you finish. New records arrive every day, people make mistakes, suppliers change names, and slowly the entropy creeps back in. The businesses that stay on top of it treat data quality as a small, regular habit, a quick check each month rather than a giant cleanup every few years.
It is genuinely the least glamorous work we do. There is no demo that makes a deduplicated customer list look exciting. But every good dashboard, every reliable report and every AI tool that earns its keep is standing on top of it. Get the boring part right, and everything built on it suddenly starts telling the truth.
