The trust paradox stalling Australia's AI ambitions

The assumption in most data strategy conversations is straightforward: build structured programmes, standardise your workflows, invest in governance, and trust follows. New research conducted by Quest Software with Corinium Global Intelligence across 154 senior data leaders in Australia and Singapore suggests otherwise.

Even organisations that have done everything right are still rebuilding data assets from scratch rather than reusing the ones they have: Not because they lack the tools, but because they don't trust what's already on the shelf.

The paradox at the heart of structured delivery

Only 25% of surveyed organisations operate a structured data products programme with standardised, repeatable workflows. The remaining 74% work ad hoc or through hybrid approaches, and for them, duplication is largely expected; different teams, different requirements, no shared infrastructure.

But for organisations that apply a structured delivery approach, the leading cause of rebuilding isn't siloed teams or unclear requirements: it is trust and data quality concerns. Teams are actively choosing to rebuild data assets rather than reuse work that already exists, was built to be shared and is technically findable.

Dig a layer further and the conditions for distrust become clear. The research points to a cluster of systemic issues: data silos that prevent cross-organisational visibility, poor communication between departments and a basic lack of awareness that relevant assets exist elsewhere. It's not that people distrust what they can see. It's that the conditions for trust such as data classification, shared context, managed data pipelines and cross-functional visibility aren't in place to begin with.

The evidence gap everyone is avoiding

The trust deficit isn't only visible in duplication. It shows up in compliance too in a way that should concern any organisation operating under Australia's current regulatory environment.

What the evidence trail requires is no mystery. A comprehensive trust model has the following non-negotiables: data quality metrics, data curation, source origination authority, usage and social proof, and governance and sensitivity status. Five components: each one visible, each one verifiable. Survey respondents are significantly more confident in their ability to meet compliance obligations than they are in their ability to produce clear, auditable evidence that those obligations have been met. Between 11% and 30% of IT spend is already absorbed by compliance activity, yet the evidence gap persists. Organisations are spending heavily on compliance and still can't demonstrate it when it counts.

Confidence and demonstrability are not the same thing. Meeting an obligation and proving you've met it requires different infrastructure. Most organisations have built for the former. Under frameworks like the Privacy Act, the Consumer Data Right, and Australian Prudential Regulatory Authority (APRA) standards, the latter is increasingly what regulators expect.

A trust model that looks complete from the inside but can't be evidenced from the outside is a liability.

Where AI production stalls

Nearly 90% of surveyed organisations are either running active AI or generative AI work, or still in the experimentation phase. Most are trying to move from proof of concept to production; and most are stalled close to the starting gate.

The bottleneck isn't compute or capability. It's whether someone can open a data catalogue, find an asset, read its lineage and trust score, understand who owns it, and quickly make a confident reuse decision. Right now, most can't. Can you?

Three moves that change the equation

The fixes aren't insurmountable. But they require treating trust as an engineering exercise, not aspiration.

1. Make trust visible inside the data product, not adjacent to it

Quality scores, classification tags, lineage transparency, certification status and usage history need to be part of the asset itself, not a separate governance layer most consumers never look at. If a data consumer can't quickly evaluate an asset's trustworthiness, it will get rebuilt.

2. Close the evidence gap, not just the compliance gap

Build the audit trail as you build the data product. Governance retrofitted after the fact is both more expensive and less credible. Organisations treating compliance documentation as a retrospective exercise are setting themselves up for exactly the evidence gap this research has identified.

3. Design for reuse confidence, not just reuse availability

A catalogue full of undocumented assets is not a viable ‘marketplace’. The organisations that move AI into production fastest will be those whose people can browse, evaluate and commit to a data asset the way they'd evaluate anything else worth relying on. Ratings, usage history and lineage transparency: this is the infrastructure of trust.

The data your organisation needs to move AI from pilot to production already exists somewhere inside it. The problem is that nobody trusts it enough to use it. The organisations that solve this challenge will compound their AI advantage. Those that don't will keep paying the hidden cost of distrust, one redundant build at a time.

*Susan Laine is the Global Data Chief Technology Officer and Data and AI thought leader at Quest Software. She has three decades of experience leading data intelligence initiatives at major enterprises globally.

Image credit: iStock.com/champpixs