The gap
84% of companies plan to increase their AI investments, yet only 40% maintain the data quality required to use AI effectively. The McKinsey Global AI Survey 2025 highlights what many Swiss SMEs experience in practice: the tools are readily available, but the data foundation is missing.
The result: companies purchase licenses for AI tools, connect them to existing repositories, and end up disappointed when outputs prove unusable. The issue is rarely the AI model itself, but rather that the underlying data is incomplete, outdated, or contradictory.
Why AI fails without data
AI models do not create knowledge out of thin air: they identify patterns in existing data. If that data contains errors, the AI simply scales those mistakes. A CRM with 30% outdated contact records produces AI recommendations that target the wrong people. A product database with inconsistent categories generates search results that confuse buyers rather than assist them.
The paradox: the more advanced the AI, the faster it amplifies poor data. While a simple rule-based system fails predictably, a sophisticated AI model generates plausible-sounding nonsense that is far more difficult to detect.
Three common mistakes
1. Deploying AI tools before auditing data. The first step is never purchasing software licenses. The first step is an honest audit: where does your data reside, how current is it, and how complete are the records? Without this baseline inventory, choosing any AI tool is pure guesswork.
2. Connecting too many systems at once. Some SMEs attempt to link CRM, ERP, and marketing automation all at once. In practice, 68% of SMEs digitalize without a clear strategy at precisely this stage. Standardize and clean one core system first, then move to the next.
3. Unclear data ownership. Data degrades over time: contacts change jobs, firms merge, and addresses change. Without a designated owner who regularly monitors and maintains data hygiene, any AI investment loses its value within 12 months.
What works
Start with a single high-priority data stream, typically your CRM. Cleanse the records: eliminate duplicates, update outdated contacts, and populate missing fields. Depending on database size, this process usually takes one to two weeks.
Only then should you introduce process automation: automatic enrichment of new leads, structured categorization by industry and region, and proactive alerts for dormant contacts.
Then, and only then, does AI add genuine value: predictive scoring for closing probability, recommended outreach timing, and automated segmentation by account potential. This succeeds because the underlying data architecture is reliable.
First step
Evaluate your most critical data source today: What percentage of records is actively verified? How many duplicate entries exist? Which fields are systematically missing?
If you can answer these three questions, you have a clear starting point for your data strategy. If you cannot, you know exactly where your initial work must begin.
