A woman monitors an AI-powered document management system in a campus office, with cloud storage and servers shown above.
The University of Waikato wants a vendor to find, classify and risk-assess its unstructured data, and it has gone to market to get one. The university is seeking a technology partner to deliver an AI-powered unstructured data security and governance platform, as part of an information management effort that spans Microsoft 365 and on-premises storage. The tender is live and no supplier has been chosen. For Microsoft 365 administrators, it is a useful case study in what "Copilot readiness" looks like when it is written into a procurement document.

A woman monitors an AI-powered document management system in a campus office, with cloud storage and servers shown above. What the tender is​

The procurement is RFP 332893, "Unstructured Data Security & Governance Solution – Supply, Implementation and Support". It is listed on the Government Electronic Tenders Service (GETS).

  • Issuer: University of Waikato's Information Technology Services department.
  • Type: Request for Proposals, with the university as the sole agency.
  • Opened: 18 November 2026, 3:00 p.m. New Zealand time.
  • Closes: 7 December 2026, 3:00 p.m. New Zealand time.
  • Regions listed: Waikato and Bay of Plenty.
  • Categories: business-function software, content management, data management and query, and security and protection software.
  • Contact: the university's procurement team. GETS says registration is needed to download the full documents.

Reseller News reports that the contract is expected to start in February 2027. I could not confirm that date in the public GETS overview, so treat it as the outlet's reporting.

How much data is involved​

According to Reseller News, the RFP puts the estate at roughly 740TB of unstructured data. That is about 364TB in Microsoft 365 and a further 375TB on network file shares. The two figures add up to 739TB, so "around 740TB" is a rounding rather than an exact sum. The breakdown comes from the reporting. The public GETS overview does not show it.

The scope matters. The GETS overview describes a toolset that uses AI and machine learning to discover, analyse, classify and assess risk exposure across Microsoft 365 and on-premises network file shares. It also says the toolset should help keep that data well governed after the initial clean-up. This is therefore not a Microsoft 365-only project. It also isn't a plain chatbot purchase. It is a discovery, classification and governance job across two very different storage environments.

What the university says it wants​

The GETS notice gives the goals:

  • Build a better structure around file storage to limit storage costs.
  • Manage governance, security and compliance risk.
  • Reduce the risk that sensitive data is accessed by AI tools.
  • Speed up the university's M365 Information Management project and cut project staff effort.
  • Provide functionality to maintain data hygiene once the project is complete.

Reseller News reports more detail from the full RFP:

  • A catalogue of holdings: the university wants an accurate picture of what exists, who owns it and where it is stored.
  • Metadata and sensitivity labels: it wants these applied, and records management improved in line with the Disposal Authority for New Zealand Universities.
  • A consolidated view: one view across Microsoft 365 and the file shares, flagging content to retain, migrate, exclude or dispose of.
  • Controls: the platform should apply sensitivity labels, data loss prevention (DLP) controls and retention policies. It should also find and fix excessive permissions and over-exposed content.
  • A multi-year relationship: the university describes a multi-year contract supporting the M365 project and the ongoing hygiene of its unstructured data.

The AI conditions​

Three conditions stand out:

  1. Human review is capped. Respondents must show how AI and machine learning can classify content automatically. The university says people should review only the up to around 20 per cent of content that the AI tools cannot classify. That is a stated expectation. It is not a measured result, and it does not show that 80 per cent of the university's content is easily classifiable.
  2. Accuracy may be tested. Shortlisted suppliers may have to demonstrate classification accuracy on a sample of university data.
  3. No training without consent. The university says its data must not be used to train AI models without consent.

The university also acknowledged the short timeframes. It asked for concise proposals with the best pricing up front, and the RFP says it will have no time to negotiate.

Why Microsoft 365 admins should care​

Reseller News frames the project around governance, compliance and Copilot readiness. The sources I reviewed do not say the university has deployed Copilot, and they do not describe any incident. The stated concern is simply that sensitive data could be reached by AI tools.

That concern matches Microsoft's own guidance. Independent write-ups of that guidance note that Microsoft 365 Copilot only returns content a user already has permission to open. In practice, oversharing in SharePoint and OneDrive becomes oversharing in Copilot answers. Microsoft's documentation points admins to Microsoft Purview and SharePoint Advanced Management to assess oversharing risks. It also describes DLP policies that can stop Microsoft 365 Copilot and agents from summarizing data carrying selected sensitivity labels.

The Waikato RFP list reads like the same checklist, applied to a bigger and messier estate: find the data, label it, fix permissions, apply DLP and retention, and keep it clean. It also adds the on-premises file shares that Microsoft's cloud tooling does not cover on its own.

Analysis​

These are my own reading of the evidence, not claims from the RFP.

  • The hard part is the combined scope. Classifying content well is only one test. The supplier must also cover cloud and local stores, work out ownership and disposition, deal with access exposure, and support the work after clean-up.
  • Discovery does not shrink storage by itself. Savings depend on retention and disposal decisions made afterwards. No cost target or contract value appears in the public sources I reviewed, so avoid assuming either.
  • Labels are only as good as their coverage. One analysis of Copilot security warns of sensitivity labels that were switched on but never applied to real content. The accuracy demonstration and the 20 per cent review figure address exactly that gap.
  • The sample test is the right idea. Vendor claims about classification accuracy are hard to compare on paper, and running them on real data is a sensible check.

Takeaways for IT teams​

If you are preparing a similar clean-up, the Waikato requirements make a handy checklist:

  1. Inventory unstructured data across both cloud and on-premises stores, including owners and locations.
  2. Apply labels and retention rules according to your records obligations.
  3. Find and fix excessive permissions before enabling AI features.
  4. Plan for human review of what automation cannot classify.
  5. Set contractual limits on vendor use of your data for AI training.
  6. Plan for the work to continue after the initial clean-up.

What we don't know​

The detailed RFP attachments were not accessible. That means no product names, evaluation weightings, storage system details or architecture specifics. The public listing shows no supplier or award. The next milestone is the 7 December 2026 close date.

 

References

  1. University of Waikato seeks AI data governance partner to tame 740TB of unstructured data Reseller News 2026-11-19T10:36:39+00:00
  2. GETS | University of Waikato - 332893 - Unstructured Data Security & Governance Solution - Supply, Implementation and Support gets.govt.nz
  3. GETS | University of Waikato - Current tenders gets.govt.nz