To spot errors in company data before they reach your reports, decide what a correct record looks like in each area, check systematically for the most common problems (duplicate records, missing fields, impossible values, dates outside the period and totals that do not match between tables) and fix them at the source rather than in a spreadsheet. A short, regular check before the month-end close stops a wrong figure from reaching the board meeting.
Most small and mid-sized companies have been there. Two different sales figures for the same month turn up in a meeting, or a customer shows an average collection period that cannot be right, and the discussion moves from the decision to which number is correct. The report is rarely the problem. The data behind it is. This article covers which errors to look for, how to organise the check and what an analytics tool can do for the team.
Why do data errors end up in reports?
Company data is written by many people, in different systems and often in a hurry. Errors come in through familiar routes.
- Manual entry. A customer created twice with the tax number typed differently, an item with the wrong unit, a date with the wrong year.
- Imports and integrations. A file imported twice, or an integration that failed halfway and left incomplete records behind.
- System changes. After an ERP migration, old and new codes live side by side and do not always match.
- Rules that changed. A new product family or a new cost centre that older reports know nothing about.
A report has no way of knowing that a value is wrong. It adds up whatever it finds. That is why the check has to happen earlier, in the data itself.
What are the most common data errors in a small business?
Data quality is measured by how many records fail simple tests. These are the types of error that most often distort a report.
| Type of error | Example | Effect on the report |
|---|---|---|
| Duplicate record | The same customer with two accounts, or the same invoice posted twice | Inflated sales or costs, customers counted twice |
| Missing field | Invoices with no cost centre, items with no product family | Amounts that fall outside every grouping |
| Impossible value | Negative quantities on a sale, zero prices, margins above 100% | Distorted averages and totals |
| Date outside the period | A document dated next year, or a due date earlier than the issue date | Transactions in the wrong month and meaningless average periods |
| Inconsistency between tables | Invoice lines that do not add up to the header total, orders for a customer that does not exist | Different totals depending on which report you open |
Duplicates deserve particular attention. A customer with two accounts splits their sales in two, and neither half looks important. An item with two codes shows stock under one and zero under the other.
How do you check the data step by step?
You do not need a data quality project to get started. A simple routine, repeated every month, deals with most cases.
- Pick the tables that feed decisions. Customers, items, invoices, orders and financial transactions are usually enough. Do not try to check everything at once.
- Write down what a correct record is. For instance, a customer has a valid and unique tax number, an invoice is dated within the period and its total equals the sum of its lines, and an item has a family and a unit.
- Count the failures for each rule. How many customers have no tax number, how many invoices have a total that differs from their lines, how many items have no family. The count shows the size of the problem and, month by month, whether it is improving.
- Fix it at the source. A correction made only in a spreadsheet is gone the following month. The wrong record has to be corrected in the ERP or in the system it came from.
- Close the door the error came through. If duplicates always come from the same import, the fix belongs in the import, not in a monthly clean-up.
- Repeat before every close. A short check before month-end saves hours of explanations afterwards.
A practical example before the month-end close
Imagine a manufacturing company with around 150 people, an ERP holding five years of history and a monthly sales and margin report for the board. At the last close, the margin on one product family rose unexpectedly, and the finance team spent two days working out why.
A check before the close would have shown three things. A batch of sales invoices with a zero item cost, because the items had been created again without their standard cost. Two customers with duplicate accounts, each holding part of the sales. And a batch of returns dated the following month, which fell outside the period.
None of these is unusual, and none of them shows in the final report. They show in the data, when someone looks with the right questions. In this example, the odd margin disappeared once the cost of the new items was corrected, and the board meeting could discuss the business instead of the numbers.
Which tools help find inconsistencies?
There are several ways to run this check, and the right one depends on the volume of data and on who does the work.
- Spreadsheets. Fine for a start, with filters and formulas that count duplicates and empty fields. The limit appears when the data grows or when the check depends on a single person.
- Rules in the ERP itself. Mandatory fields and validation at entry prevent many errors from getting in. They do not catch the ones already there.
- Database queries. Someone who knows SQL can write a set of checking queries and run them every month. It works, but it stays in the hands of whoever can write and maintain them.
- Data analytics tools. Some scan the database for inconsistencies and point out what needs attention, without anyone having to know where to look.
Whatever the tool, it should show where each number comes from. An answer that cannot be checked is one more place for an error to hide. We cover this in how to check what an AI tells you about your company data.
How does EngiAnalytics help find errors in the data?
EngiAnalytics is the Engibots conversational data analytics platform. As well as answering questions in natural language, it carries out exploratory discovery, scanning the database for inconsistencies and showing what should be corrected before it reaches a report.
Every answer comes with a plain-language sentence describing what the query returns and, if that sentence quotes a figure that is not in the returned data, the reader is warned. The SQL query behind each answer can be opened and copied, next to the table of returned data, so that any figure can be checked at the source. At the start, a getting-started screen shows what the product has already understood about the data, from the business areas to the period available.
EngiAnalytics connects to MySQL and PostgreSQL databases, including ERP and CRM databases running on those engines, and to CSV and Excel files, with no migration and no replacement of existing systems. The team learns where to look, and corrections are still made in the source system.
A good starting point is to choose three tables, write two or three rules for each and count the failures before the next close. That first number already tells you how far you can trust the reports that reach the board today.
Frequently asked questions
What is data quality?
It is the extent to which a company's data is complete, correct, free of duplicates and consistent across systems. It is measured by counting the records that fail simple rules, such as a customer with no tax number or an invoice whose total does not match the sum of its lines.
What are the most common data errors in a small business?
Duplicate records, missing fields, impossible values, dates outside the period and inconsistencies between tables, such as invoice lines that do not add up to the header total.
How often should the data be checked?
At least before every month-end close, and always after a large import or a system change. A short, regular check works better than a yearly clean-up.
Should errors be fixed in the ERP or in the report?
In the ERP, or in whichever system the data came from. A correction made only in the report or in a spreadsheet is lost the following month and leaves the error waiting for the next report.
What does EngiAnalytics exploratory discovery do?
It scans the database for inconsistencies and shows what should be corrected before it reaches a report.