When a leader doubts a report, the first thing worth asking is where its records came from. That answer begins with what can go wrong in them. Data quality issues are the ways those records can be wrong, missing, repeated, old or written in different formats. Each one also starts at a place in the firm where a named person can look. Five kinds follow, each with something you can notice as well as someone to ask. This article names the kinds but leaves the fixing to other work. This page is part of the use case on reliable reports and AI answers.
Last reviewed 2026-11-02 by Manish Sharma.
Data Quality Issues: A Working Definition and Five Kinds
A few terms come before the five kinds.
- Data quality is how far the records a firm keeps can be trusted for the job they are used for.
- A record is one row about one customer, order or item.
- Analytics means turning records into the totals and charts that reports show.
- In this article, artificial intelligence means software that carries out work a person would otherwise do, such as reading or sorting.
The data quality issues in this article fall into five kinds:
- Values that are wrong.
- Values that are missing.
- Records that are repeated, also called duplicates.
- Records that are out of date.
- Values written in different formats or units.
Numbers that disagree between two systems are a separate problem, closer to unifying business data, and this article leaves them out. The first two kinds can start when someone fills in a field. The other three can build up as records age, move or merge. To trace figures that differ from one system to another, see four reasons business numbers disagree between systems.
Wrong and Missing Values: Where They Start
Both of these kinds can start at the moment a person fills in a field.
A wrong value is one that could not be true. For example, a delivery date falls before the order date, or a phone number sits in the quantity column. It enters when someone types into the wrong box, or when the person who designed the form left a field open to any text. Ask: who enters this field, and what does the form let them type?
A missing value is a blank where the report owner expects something. For example, a customer record has no region. It starts when a field is optional on the form while the report owner counts on it. A field the form lets staff skip may be left blank. Ask: which empty fields do your reports count on, and who can make them required?
Poor data quality at the point of entry comes down to two places to look. The first is the form itself, and the second is whoever fills it in. Neither is a question about software alone. After all, a person chose what the form would accept.
Duplicate, Old and Mixed-Format Records: Where They Start
These three kinds can build up after entry, as people copy records, leave them alone or share them between teams.
A duplicate is the same customer entered twice, perhaps as “Al Noor Traders” and “Al-Noor Trading”. It appears when someone adds a new customer without searching first, or when two lists are merged into one. Ask: who creates a new customer, and what do they check first?
An out-of-date record was right when someone entered it. Then the customer moved or the product changed, and nobody asked anyone to update the record. Ask: who is told when a customer or supplier changes?
A mixed format means one team writes dates day-first and another writes them month-first. It also covers a column where some weights are in kilograms and others in pounds. Ask: who decides the format for each field?
In each of the three, the gap sits with a person or a habit. The questions name people rather than systems. Software can take on some of the spotting, and the options are laid out in four places where a check on records can run.
Key Takeaway Records can go wrong long after someone typed them in correctly.
Data Quality Issues: Signs, Sources and Who to Ask
Within Signisys’ working description, a report reader can spot a mark for each kind, and someone can be asked about each.
| Kind of issue | What a report reader notices | Where it starts and who to ask |
|---|---|---|
| 1. Wrong values | A total that could not be right | At entry: the owner of the form and the team that enters the data |
| 2. Missing values | Blank cells in a report | At entry: the owner of the form and the report owner |
| 3. Duplicates | The same customer counted twice | At entry or after a merge: the person who creates customer records and the system administrator |
| 4. Out-of-date records | For example, returned mail or an old address on an order | As records age: the owner of the customer or supplier list |
| 5. Mixed formats | A date column that sorts oddly | Between teams: the person who sets the standard for each field |
Take one strange figure from a recent report and find its row above. Then ask the person named in the last column.
This page puts no order of size on these data quality issues, and one report can show more than one at once. Instead, the table is a reasoning aid.
What These Issues Do to Reports and AI Answers
A leader who reads a report built on repeated customers counts some customers twice. An average on the page can also shift. In the same way, a team that asks an AI assistant about out-of-date records gets an answer that was true once.
Analytics turns rows into totals, so one wrong row moves a total without any warning on the page. When a team builds an AI assistant on its records, an old record can come back as a confident sentence. In both cases, the reader sees the result and not the row behind it.
This is reasoning about what can happen, not a measure of how far it reaches in any firm.
Researchers have also tried to sort these problems into groups. A 2024 paper by Qiaolin Qin, Heng Li and Ettore Merlo, Wrangling Data Issues to be Wrangled, reports a literature review and proposes a taxonomy of data issues that sorts each one along an attribute dimension, meaning what the problem is in itself, and an outcome dimension, meaning how it shows up. The authors used it to label issues from an industry partner’s data warehouse. It is one proposal among several in the research literature.
Clean formatting on a page says nothing about the rows behind it. Ask where the rows came from before you act on a total.
Data Quality Issues: Tracing an Odd Deliveries Report
We made up the furniture maker below to show all five kinds in one place, and it comes from no client’s records. For example, a furniture maker ships to retailers, and its sales head reads a report of deliveries by city.
To begin with, one city shows a total far above the rest. Someone at the order desk typed a postcode into the quantity column. The order-desk lead is the first person to ask. On a long list of orders, nobody filled in the city, and the form leaves it optional. Then the owner of that form can look.
One retailer appears three times under slightly different spellings. That points to whoever creates customer records. On recent orders, staff still pick an old warehouse address, which sends the question to the owner of the customer list. Finally, the date column mixes two styles, a matter for the person who sets the format for each field.
Read this way, the data quality issues in a single report point to five different people. Your own reports, however, will show their own mix.
Key Takeaway Trace each odd figure to a row, and each row to a person.
Where to Start Looking in Your Records
Begin with a report you already doubt, then set each oddity beside the matching row of signs above.
Three questions stay with you after that. Which kinds sit in your own records? How common is each one? And what would putting it right mean for your forms, lists and teams? In the end, those answers come from your own records and from the people who keep them.
An outside review is worth asking for in one case. That is when two reports built from the same records show different totals, and their owners are unable to trace which record caused the gap.
Signisys provides data and AI services through its data integration and business intelligence solution and its data and AI service page. The box below describes a Signisys service, not neutral advice, and it makes no promise. Bring one report you doubt and the names of the people who keep its records.
Data Quality Issues FAQ: Entry, Signs and Owners
Which data quality issues start at the point of entry?
Do data quality issues start with people or with systems?
Can data quality issues be seen in a report?
How can a leader tell that records are repeated?
Which of these issues can our own team look at first?
Security Insights, Delivered.
Practical, vendor-neutral analysis of threats, tools, and architecture decisions.