Two years ago, "analysing data" meant opening Excel or waiting for a report from an analyst. Today, any manager can type a question in plain English and get a chart back in seconds. That shift is real and useful — but it brought a new question most SMEs haven't stopped to answer yet: what's safe to paste into an AI chat, and what should never leave the company?
This guide is practical, not alarmist. Generative AI is a genuinely powerful tool for data analysis — as long as you know exactly where it helps and where it exposes the business to unnecessary risk.
What AI Does Well in Data Analysis
Used correctly, AI speeds up parts of the work that used to take hours:
- Summarising reports that already exist — pasting a table of monthly results and asking for a three-sentence executive summary
- Suggesting formulas — asking for the right DAX or Excel formula for a specific metric, instead of searching forums
- Exploring hypotheses — asking "what other metrics should I look at to understand this sales drop?" before opening any tool
- Drafting insights from already-aggregated numbers — turning a table of category totals into a clear paragraph for management
Notice the pattern: in every one of these cases, the data entering the conversation is already summarised, aggregated, or generic. That's exactly where the security line begins.
What Should Never Be Pasted Into a Public Chat
Before pasting any file or table into an AI tool, ask yourself: does this identify a real person?
- Customer lists with name, email, or phone number
- Financial data with IBAN, card numbers, or tax details
- Entire sales or CRM files with no anonymisation
- Individualised HR data — actual salary, performance reviews, health information
The risk isn't just whether a tool "uses your data for training" — that varies by product and plan. The underlying risk is simpler: pasting personal data into a tool with no data protection agreement with your company is, by itself, a data transfer outside the perimeter GDPR requires you to control.
Not All AI Tools Are Equal
The most important — and least discussed — difference is where the data actually lives:
- Copilot inside Microsoft 365 or Power BI, with a business licence: data stays within the organisation's tenant, under the same data protection agreement already signed with Microsoft.
- ChatGPT or similar tools on a free or personal plan: outside that contractual perimeter, unless it's explicitly the Enterprise or Team tier with a no-retention policy.
A simple rule of thumb: if the tool you're using doesn't have a signed data processing agreement (DPA) with your company, treat it as public — and never paste something there you wouldn't post on social media.
A Safe Workflow, Step by Step
- Aggregate before asking. Totals by category, month, or region — never row-by-row with identifiers.
- Strip direct identifiers. Name, email, tax ID, and phone number aren't needed for AI to spot a sales pattern or a trend.
- Test the logic on a fictional sample before applying the same reasoning to real data inside a tool with proper AI integration and contractual security.
- Treat the response as a draft, not a fact. Manually confirm any number that feeds into an important decision.
Where AI Fails Silently
The biggest risk isn't AI refusing to answer — it's answering with complete confidence based on poorly organised data. If your data model has no clear definition of "active customer" or "margin," AI won't ask. It will simply compute something plausible, written fluently and convincingly — which makes the mistake more dangerous than an obvious human error, because nobody questions it.
It's the same logic that has always applied to traditional dashboards: the tool is only as good as the data behind it. That hasn't changed with AI — it's just gotten easier to forget.
Quick Checklist
- ✅ Data aggregated, no direct identifiers
- ✅ Tool with a data protection agreement with the company
- ✅ Critical numbers manually confirmed before deciding
- ❌ Customer or HR files pasted without treatment
- ❌ A decision made solely from an AI answer, with no validation
In the end, the most important question isn't "which AI tool to use" — it's whether your data is clean, modelled, and ready for any tool, dashboard or AI, to work on with confidence. That's exactly the starting point of the data analysis services at PC Data Insights: diagnostics and cleanup before any automation enters the picture. Get in touch via the form or WhatsApp to discuss the current state of your company's data.