AI Data Requirements: What You Need to Get Started

AI Data Requirements: What You Need to Get Started

Data is the fuel for AI. But what data do you need, and how much? This guide gives you clear, practical answers.

The Good News First

You need less data than you think. Thanks to pre-trained AI models, 6 to 12 months of historical business data is sufficient for most SME projects. For an AI agent you need your most frequent customer inquiries from recent months, ideally 200 to 500 example inquiries, your product information, price lists, and FAQ, and example responses showing your quality standard. For workflow automation you need 3 to 6 months of transaction data such as orders, invoices, or appointments, examples of correctly processed transactions, and examples of exceptions and special cases. For business insights you need at least 12 months of sales data with timestamps, customer data with purchase history, and optionally external data like weather, calendar, or market data.

Where to Find Your Data

Most SMEs already collect all necessary data, just not consciously. Check these sources: POS system and online shop (transaction data), accounting software (financial data), email inbox (customer inquiries and correspondence), CRM (customer data and interactions), calendar and booking system (appointments and availability), Excel spreadsheets and documents (informal data collections), and even WhatsApp and social media (customer inquiries).

Checking Data Quality

Four quality dimensions are relevant for AI. Completeness asks whether important fields are filled. If 30% of your customer data lacks email addresses, that is a problem for email-based automation. Consistency checks whether same things are captured the same way. If addresses are sometimes entered with country codes and sometimes without, this must be cleaned before AI use. Currency checks whether data reflects the current state. Customer data from 5 years ago is less valuable for behavior analysis than current data. Accuracy asks whether data is correct. Typos in phone numbers, wrong categorizations, and swapped fields must be corrected.

Preparing Data: Practical Steps

If you have identified data problems, follow these steps: Consolidate data from different sources. Clean duplicates and obvious errors. Standardize formats like date formats, phone numbers, and addresses. Fill critical missing fields. Document what the data means.

Most SMEs underestimate this step. Plan 1 to 2 weeks for data preparation. It is well-invested time that improves the quality of the entire AI project.

Don't Forget Data Protection

If your data contains personal information, you must observe the Swiss Data Protection Act. For AI projects this means: only use data necessary for the purpose. Inform customers when their data is used for AI. Anonymize data where possible. Ensure processing on Swiss servers.

Checklist: Are You Ready?

Check these points: Do you have digital business data for at least 6 months? Is the data exportable from your systems? Are there obvious quality problems? Are the relevant data sources identified? Is data protection considered? Do you have someone on the team who understands the data?

If you can check most points, you are ready for an AI project.