The data you need for AI depends entirely on what kind of AI you are using. For everyday generative AI assistants such as ChatGPT, Copilot, Gemini or Claude, you often need no special dataset at all: clear instructions and the relevant documents for the task are enough. For AI connected to your business systems you need organised, accessible information. Only custom-built AI models need large, clean, labelled datasets. Many organisations delay AI because they assume they need the third when they only need the first.
Level 1: Everyday AI assistants
What they need: your instructions (prompts), plus whatever you give them for the task, such as a document, an email thread, a spreadsheet or meeting notes.
What you need to prepare: very little. The quality of results depends mainly on how well people brief the AI and check its output.
Data considerations: use approved business tools with appropriate privacy terms, and follow clear rules about confidential and personal information.
Typical uses: drafting, summarising, research, analysis of a single document or spreadsheet, meeting notes.
Level 2: AI connected to your information
What it needs: access to collections of your documents or systems, such as a shared drive, knowledge base, CRM or helpdesk.
What you need to prepare: organised, reasonably current information; clear permissions so people only see what they should; removal of obviously outdated or duplicate material.
Data considerations: access controls matter. If your shared drive is a mess of old versions, AI will happily quote the wrong one.
Typical uses: answering staff questions from internal policies, searching past proposals, summarising customer history.
Level 3: Custom AI models
What they need: large volumes of relevant, consistent, often labelled historical data, plus technical expertise.
What you need to prepare: significant data engineering, quality checking and governance.
Typical uses: demand forecasting, fraud detection, specialised prediction or classification.
Where most organisations should start
Start at Level 1. It delivers real value quickly and needs no data project. Move to Level 2 for specific workflows where connecting AI to your information clearly helps. Consider Level 3 only for a specific, high-value problem that off-the-shelf tools cannot solve.
A quick data readiness check for Level 2
- Do people know where the authoritative version of key documents lives?
- Are outdated documents archived or clearly marked?
- Are permissions set so sensitive information is restricted?
- Is there someone responsible for keeping key information current?
If the answers are mostly yes, you are probably ready for Level 2 on that information.
The data rules that matter at every level
Whatever level you are at, staff need simple, clear rules on what information must never go into AI tools, which tools are approved, and how to handle personal data in line with UK GDPR. Our guide on protecting customer data when using AI covers this in more depth.
For agent-based AI, which has its own data needs, see what data you need for agentic AI. Our Leading AI Innovation & Organisational Transformation course focuses on Level 1 and the safe use of business information, including how to pull insights from reports and spreadsheets without formulas.
Frequently asked questions
Do we need a data strategy before using AI?
Not to start with everyday AI assistants. A data strategy becomes important when you connect AI to your systems or build custom models.
Can AI work with messy spreadsheets?
Often yes, for analysis of a single file. AI assistants can help clean, summarise and spot patterns, though results should always be checked.
Is our data used to train AI models?
It depends on the tool and plan. Business versions of major AI tools typically offer terms that limit use of your data, but check each vendor's terms carefully.