How Do You Get Your Data Ready for AI? Quality, Access, Permissions and Sensitive Data

The short answer

Getting data ready for AI means four things: quality (accurate, current, consistent), access (reachable by approved tools), permissions (only the right people can see it) and sensitivity (you know where personal and confidential information sits). You do not need perfect data across the business. You need reliable, well-governed data for the specific use cases you are starting with.

Why data decides AI outcomes

AI models are capable, but they work with what they are given. Point an AI assistant at a shared drive full of outdated policies and three versions of the price list, and it will confidently quote the wrong one. Connect a forecasting tool to a ledger with inconsistent coding, and the forecast will be built on noise.

AI also changes the risk profile of existing data. Information that sat unnoticed in a forgotten folder can now be found and summarised in seconds. That makes data readiness both a value question and a risk question.

The four pillars of AI data readiness

Quality: fix what the use case depends on

Data quality work can expand endlessly, so anchor it to use cases. For each priority use case, ask:

• Which data does it rely on? For an AI knowledge assistant, that is policies and procedures. For AI invoice processing, it is supplier master data and purchase orders.
• Is there a single source of truth? If three versions of a document exist, decide which is authoritative and archive the rest.
• Is it current? Set review dates and owners for content AI will draw on.
• Is it consistent? Standard naming, codes and formats make AI output far more reliable.

A practical example: before launching an internal AI assistant to answer staff questions about HR policies, a business reviews its policy library, removes superseded versions, confirms each remaining policy has an owner and review date, and stores them in one location. The AI then answers from current, approved content rather than whatever it happens to find.

Access: bring data within reach, safely

Much business knowledge lives in places AI tools cannot or should not reach: personal drives, inboxes, local spreadsheets and old systems. Improving access means:

1. Moving important shared content into managed, backed-up repositories.
2. Connecting core systems through approved integrations rather than exports emailed around.
3. Retiring or archiving legacy stores that nobody maintains.
4. Avoiding workarounds such as staff uploading files into unapproved AI tools to get answers.

A data and technology strategy helps decide where data should live long term, so AI readiness work supports the broader architecture rather than adding more silos.

Permissions: the readiness issue most businesses miss

AI assistants in productivity suites search across content each user can access. That is the right design, but it exposes years of loose sharing. Common findings include:

• Sites or folders shared with all staff that contain salary, performance or board material.
• Anonymous links to sensitive files that never expired.
• Former project teams and guests who still have access.
• Sites with no owner, so nobody reviews who can see what.

Fixing permissions before switching on AI is far easier than responding after something sensitive has been surfaced. A Microsoft 365 security review is an efficient way to find oversharing in Microsoft environments, and our Microsoft Copilot rollout guide explains how this fits into a rollout.

Sensitive and personal information

Under the Privacy Act and the Australian Privacy Principles, organisations covered by the Act must handle personal information in line with the purpose it was collected for, keep it secure and be careful about disclosure, including to AI vendors. The OAIC's guidance on commercially available AI products recommends caution about entering personal information, particularly sensitive information, into publicly available AI tools.

Practical steps include:

• Discover – identify where personal, health, financial and other sensitive data is stored.
• Classify and label – apply sensitivity labels so protection follows the data. Tools such as Microsoft Purview support classification, labelling and data loss prevention.
• Minimise – only give AI access to the data the use case needs; remove or de-identify what it does not.
• Check vendor terms – understand where data is processed and stored and whether it is used to train models.
• Plan for incidents – AI-related exposure of personal information may trigger obligations under the Notifiable Data Breaches scheme. See our guide to notifiable data breaches in Australia.

A practical data readiness checklist

1. List your priority AI use cases.
2. For each, map the data sources it will use.
3. Assign an owner to each data source.
4. Assess quality: duplicates, currency, consistency, completeness.
5. Review permissions and remove oversharing.
6. Locate and label sensitive and personal information.
7. Decide what the AI may and may not access.
8. Fix the critical gaps, then pilot.
9. Review data quality and permissions regularly, not once.

Data readiness is one part of broader readiness. Our AI readiness assessment guide covers strategy, people, process and governance as well.

How The Technology Office prepares data for AI

The Technology Office helps businesses get their data ready for AI in a focused, use-case-led way, so effort goes where it unlocks value and reduces risk. We are independent and sell no data platforms or AI products, which means our advice is about fixing the right things, not buying more tools.

Through our AI consulting and data strategy services, we map data to priority use cases, work with your IT provider to tighten permissions and protect sensitive information, and then run measured pilots against a baseline. We work with businesses of every size and industry across Australia.