Data quality consulting for companies in Germany

What poor data quality costs your company

Poor data quality means records that no longer describe the customer, supplier or product they stand for. Staff notice it as an offer sent to a contact who left the company, or as a report that counts one customer twice.

In the DIHK digitalization survey of 4,686 companies, held in November 2025, 31 percent named poor data quality as a challenge in using their data, up from 28 percent a year earlier. Legal uncertainty came first, at 60 percent (DIHK, January 2026).

Records go stale because the things they describe change. A company changes its name, merges or closes, and a contact person moves to another employer. A table that was clean on the day of an import drifts from then on (data change management).

AI tools make the problem more visible. An assistant that answers from your CRM repeats an outdated address and gives no hint that it could be wrong.

How I improve data quality

I measure first and fix second, under rules that can be undone. Then the places where errors enter get checks. These rules are in production use on a client’s information platform; the concept is on Change management, de-duplication and workflows.

A duplicate scan reports more than its merges. It lists how many candidates each signal found, how many a contradiction ruled out, and which pairs stayed below the threshold for an automatic merge. Candidates come from the name, the address and shared numbers such as a VAT ID, because the English and the German name of one company can have no word in common.

A high match score does not override a contradiction. Two companies can share one address, and “Plant North” and “Plant South” are two sites of one maker. A dissolved company and a company founded later at the same address are two entities, and the scan checks such contradictions before it writes a merge.

A merge retires one record and keeps it as an alias of the other, with its identifiers. A correction adds the new value with its source and date and hides the old one. Both can be taken back, and the history shows who changed what.

Pairs that the rules cannot settle go to a person or to an AI agent that works under the rules the staff follow. An unsure answer leaves both records in place.

The last step is the entrance. New duplicates arrive through imports, web forms and staff who create a customer that already exists. An import account may add records and cannot overwrite existing ones, and a form can look for a matching record before it creates one.

Data quality checklist for your customer records

The questions I work through at the start of a clean-up. You can answer them for your own customer table.

1. How many records share a VAT ID, a commercial register number or an email domain with another record?
2. How many records name neither a source nor the date of their last check?
3. How many customers in the table were dissolved or renamed after they were entered?
4. How many contact emails bounced in the last twelve months?
5. Which required fields are empty, and in what share of the records?
6. Which values contradict each other inside one record, such as a German postcode with the country code of Austria?
7. Who may change a customer record, and is each change logged with its author?
8. Can a wrong change or a wrong merge be undone?
9. Through which doors do new records enter: imports, web forms, staff, other systems?
10. Which reports and AI tools read these records, and which of the problems above reach them?

Who I am

I am Michael Wutzke, based in Frankfurt, with more than twenty years in IT and media. I programmed the data stores of a global real estate database myself and later built the search of a real estate database with Elasticsearch. Today I help a client build an information platform with AI agents that follow the rules a human editor follows. I teach at Claude Hacker House and build software with Claude and Codex. I am also interested in open-source AI models that a company runs on its own servers. Details: Career stages.