Introduction
Artificial intelligence is only as good as the data that drives it. This principle sounds trivial but is systematically underestimated by the majority of companies. McKinsey confirmed in a comprehensive study in June 2026 what experienced AI consultants have been preaching for years: data quality is the decisive factor for scaling AI initiatives.
The reality in most SMEs is sobering. Data is scattered across silos, formats are inconsistent, responsibilities unclear. CRM systems contain outdated contacts, ERP data has gaps and Excel spreadsheets with critical business information exist only on individual laptops.
This article shows how SMEs can build a solid data foundation that serves as the basis for successful AI transformation.
1. The Data Reality in SMEs
Current studies paint a clear picture. According to NTT Data, more than 82% of companies exceed their original budget for digital transformation projects - often due to unexpected data problems.
Typical data problems fall into four categories:
Fragmentation: Data exists in isolated systems without connection. The CRM knows the customer differently than the ERP, marketing automation has its own contact lists and sales maintains additional Excel spreadsheets.
Inconsistency: The same information is captured differently in different systems. Address formats vary, product names are not standardized, date formats change between systems.
Incompleteness: Critical data fields are not maintained. Customer profiles have gaps, transaction histories are incomplete, interaction data is missing.
Obsolescence: Data is not regularly updated. Contact information is outdated, market data comes from past periods, competitive information is no longer current.
2. The Five Dimensions of Data Quality
Completeness
Target: at least 95% for AI-critical fields.
Accuracy
Target: at least 98% accuracy for business-critical data.
Consistency
Target: at least 99% agreement for key fields across systems.
Timeliness
Target: depends on data type (contact data: max. 6 months, transaction data: real-time).
Relevance
Target: every collected data field must be assigned to at least one AI use case.