A single folder glowing faintly inside a half-open steel filing cabinet drawer, representing controlled access to sensitive customer data
AI-generated image

Can You Put Customer Data Into an AI Tool? A Five-Question Checklist

PM
Peter Mangin
Founder, AI Innovisory
6 min read

You have 500 customer comments, a deadline, and an AI tool that could summarise the lot in seconds. The spreadsheet also contains names, email addresses, purchase histories, and whatever customers happened to mention when they wrote in.

Can you upload it?

Sometimes. It depends on why you are using the data, whether that use is permitted, and what the tool does with it once it arrives. Our full guide to NZ data privacy and AI covers the legal detail: IPP3A, the six-level classification framework, and the workflow to run before any AI use. This is the shorter version: five questions worth asking at the moment you are about to hit upload.

Can I put customer data into an AI tool in New Zealand?

Sometimes. It depends on what the data contains, whether your intended use fits the purpose it was collected for, what the tool does with what you upload, and whether your organisation has approved that tool for that task. Start by asking whether you need identifiable information at all.

1. Does the AI Need to Know Who This Is?

Take a retailer trying to understand why customers complain about delivery. The useful information is the complaint theme, the delivery timeframe, and whether the customer received updates. Names, phone numbers, and unrelated purchase histories add little to that task.

Before uploading, check what the file actually contains. Read the comments themselves, not just the column headers. A customer may have explained that they missed a delivery because they were in hospital. Deleting the name column does not remove that detail, and removing names does not automatically make the remaining data anonymous. A location, an unusual purchase, and a distinctive complaint can still identify someone.

Our NZ Data Classification Checklist walks through the six levels we use to answer this: from public information suitable for any tool through to sensitive personal data that needs an approved enterprise tool with a data processing agreement, or no AI at all.

2. Are You Allowed to Use It This Way?

Does the NZ Privacy Act limit how I can use customer data with AI?

Yes. Information Privacy Principle 10 generally limits use of personal information to the purpose it was collected for. Using complaints to improve delivery communications and using them to predict which customers are financially vulnerable are different proposals, and each needs its own check.

Write the task down in one sentence: "We want to understand recurring delivery complaints so we can improve customer communications." That gives your team something concrete to assess against the purpose the data was originally collected for.

Check where the information came from, too. Data collected indirectly, a purchased list, an enriched contact record, a client-supplied database, generally requires notification under IPP3A, in force since 1 May 2026. A marketing opt-in does not automatically cover every later use, and it does not answer whether a supplier has met their own notification obligations.

3. What Happens After You Upload It?

A familiar brand or a paid subscription does not settle this on its own. Confirm, in writing, whether the tool trains on what you input, what it retains and for how long, who can access it, where it is processed, and how a customer's access or correction request would be handled.

"Not used for training" answers one question. It does not tell you everything about storage, access, or onward sharing. Overseas processing is not automatically prohibited either: a provider processing information solely on your behalf is generally treated differently from one using it for its own purposes, but your organisation still has to check the arrangement and take responsibility for it.

4. Does Your Approval Actually Cover This Task?

A tool approved for drafting public-facing copy may not be approved for analysing customer complaints. Look for instructions that specify which information can go into which tool, for which tasks, under what conditions. "We use this AI tool" is not an instruction. It is the absence of one.

Where the work touches Māori communities, knowledge, or interests, bring appropriate Māori data governance expertise into the assessment before you proceed. Removing names does not resolve questions about collective rights in data, and individual consent does not extinguish them either, a distinction we cover in more depth in the guide to AI and Māori data sovereignty.

5. Who Checks the Answer Before Anyone Acts on It?

Should I trust an AI summary of customer feedback without checking it?

No. Before a finding changes a campaign, a policy, or how a customer is treated, someone needs to read the underlying comments, check the themes against the records, and look for complaints the summary overlooked. A fluent answer is not the same as a verified one.

Suppose the AI concludes that customers want faster delivery. Before anyone builds a campaign around that, someone reads the underlying comments and finds the real, recurring problem is missed delivery windows and poor updates, not speed. Promising speed would not have solved it.

Name the person responsible for that check before the output reaches a decision. If the finding changes who gets an offer, or how a customer is treated, check it more carefully, not less.

Your next upload, as a five-question table:

1
What does the AI need?Which information is necessary, and what can be removed first.
2
Is this use permitted?Where the information came from, and whether the proposed use fits.
3
What happens after upload?How the provider handles, retains, and shares it.
4
Does approval cover the task?Which instructions apply to this data, this account, this use.
5
Who checks the answer?Who verifies it before anyone acts on it.

Write down the decision so the next person does not have to guess: "We can use this tool for this task, with this information, under these conditions." Then, for the detail behind each question, the classification levels, the notification rules, the tool selection guide, work through the full data privacy and AI guide.

Based on the Privacy Act 2020 (NZ), IPP3A (in force 1 May 2026), and Te Mana Raraunga Māori Data Sovereignty Principles. Not legal advice. Contact the Office of the Privacy Commissioner at privacy.org.nz for guidance on your specific situation.

Want your team working from the same answer, not five different guesses?

AI Innovisory helps New Zealand organisations turn this checklist into an instruction people can actually follow, through governance advisory and hands-on workshops that build the judgement, not just the policy.

Book a strategy call

Enjoyed this article?