Data Readiness Playbook for Small Businesses

Data Readiness Playbook for Small Businesses

"We don't have enough data for AI" this is the most common objection we hear from SME owners. And in most cases, it is not true. You have more data than you think. The question is: is it in the right form?

Myth: "AI Needs Big Data"

Forget everything you have read about Big Data. Large language models and pre-trained AI systems have drastically reduced the amount of data you need for an SME project. For most use cases, 6-12 months of historical business data is sufficient. That is typically a few thousand data points, not millions.

An example: to build a demand forecasting system for a restaurant, we don't need millions of data points. We need the daily revenue figures from the last 12 months (365 data points), weather data (publicly available), the regional event calendar (publicly available), and the reservation history. That is everything. No data warehouse, no data lake, no Hadoop clusters.

Where Your Data Already Exists

Most SMEs are already collecting valuable data without knowing it: your POS system stores every transaction with timestamp, items, and amounts. Your email inbox contains thousands of customer inquiries with patterns. Your accounting software has years of financial data. Your online shop logs every click, every order, every return. Your calendar documents appointments, utilization, and availability. Even your Excel spreadsheets and Word documents are data sources.

The Four Dimensions of Data Readiness

We assess an SME's data readiness along four dimensions: availability, accessibility, quality, and relevance.

Availability asks: does the data even exist? In most cases, yes. The question is just where. Sometimes data is hidden in systems you didn't know stored data (e.g., your phone system logging call times).

Accessibility asks: can we access the data? This concerns technical aspects (does the system have an export function or API?) and legal aspects (are we allowed to use the data for AI?).

Quality asks: is the data consistent and complete? The most common quality problems: missing fields (30% of customer profiles without email address), inconsistent formats (phone numbers sometimes with area code, sometimes without), duplicates (the same customer under three different spellings), and outdated data (addresses from five years ago).

Relevance asks: does the data contain the information the AI needs? If you want to forecast demand, you need historical sales data, not just revenue figures.

The Data Readiness Check: 10 Questions

Answer these ten questions honestly to assess your data readiness: Do you have digital records of your core transactions (orders, invoices, appointments)? Do these records cover at least 6 months? Is the data in a system that allows export or API access? Is your customer data consistently captured? Can you assign timestamps to your transactions? Do you have access to the relevant data sources? Is the data current? Are there obvious gaps or inconsistencies? Is the data stored in compliance with data protection laws? Is there someone on your team who understands and can explain the data?

If you answer 7 or more questions with yes, your data is probably AI-ready. At 4-6, preparation is needed but typically takes 1-2 weeks. Below 4, we recommend first improving data collection, which is often possible with simple measures like consistent data entry and regular backups.

Preparing Data: Practical Steps

If your data needs preparation, these are the most important steps: consolidate data from different sources into a uniform format. Clean obvious errors and duplicates. Fill in missing fields where possible. Define standards for future data entry to ensure long-term quality. Document what individual data fields mean.

This preparation doesn't have to be perfect. The rule is: 80% clean is enough to start. The AI will recognize patterns even with imperfect data. Results simply improve as the data gets cleaner.

Data Protection: What You Need to Consider

In Switzerland, the revised Data Protection Act (revDSG) applies. For AI projects, three points are particularly relevant: personal data may only be processed with a legitimate interest or with consent. Data must be processed in Switzerland or in a country with adequate data protection. Affected persons have the right to information and deletion.

For most SME AI projects, these requirements can be met with simple measures: processing on Swiss servers, anonymization where possible, and transparent customer information.

Conclusion: Data Is Not a Barrier

Data readiness is rarely a knockout criterion for AI in SMEs. In our experience, 90% of SMEs have enough data for a first AI project. They just don't know it. The biggest hurdle is not the lack of data, but the uncertainty about whether existing data is "good enough." Our advice: have your data assessed by an AI specialist before making assumptions.