The "secret sauce" problem every AI project eventually runs into
At a major AI summit in Sydney this week, Telstra's CEO took the stage alongside some of the biggest names in Australian business.
The conversation was about what actually separates companies that are getting results from AI from those still waiting. Her answer wasn't about models or data infrastructure. It was about something much harder to pin down.
Vicki Brady has been leading one of Australia's largest companies through a multi-year AI transformation. She's not someone hedging on the technology. Which made it notable when she said that the thing setting companies apart usually isn't their data. It's the knowledge that never gets written down.
"As organisations we need to be clear on what differentiates us," she said, "and sometimes that secret sauce is in that layer that's not codified."
It's also exactly what Sugarwork was built to solve.
Automation without context won’t work
Most AI deployments follow a similar arc.
There's a business case, a vendor, a roadmap. The technical build goes well enough. Then somewhere between proof of concept and production, things get sticky. The outputs need constant correction. The automation keeps missing the nuances. The tool works, technically, but it doesn't quite work.
This isn’t an unusual outcome. A March 2026 survey of 650 enterprise technology leaders found that 78% have AI pilots running, but only 14% have successfully scaled one to organisation-wide deployment.
The most commonly cited reason isn't model performance or integration complexity. It's insufficient domain training data: the structured, context-rich understanding of how work actually gets done in a specific organisation.
What usually happens is that the system was trained on documented workflows. How work is supposed to happen, not how it actually does.
The exception handling, the workarounds, the judgment calls your best operators make without thinking, none of that made it into the spec. It was never written down. It lived in people, and it stayed there.
Peter Tonagh, executive chairman of data analytics firm Quantium, made a prediction at the same summit. Within 18 to 24 months, he said, far more organisations will be recording every major conversation in an attempt to capture what's actually happening inside their businesses.
He's right about the direction. But a recording tells you what was said. A transcript tells you what was said. Neither tells you what it means: how a conversation connects to adjacent workflows, which parts represent genuine exceptions versus standard practice, what a new hire or an AI system would need to know to replicate the outcome reliably.
The value isn't in the capture. It's in what you do with it. Ambient capture that feeds into a structured understanding of how work actually gets done is a fundamentally different thing from a searchable archive of meetings.
The knowledge that drives your business is probably not where you think it is.
It's not in your documentation, your process maps, your knowledge base, or the SOPs someone wrote during an ISO audit in 2019. It's in the heads of your most experienced people, passed on through conversation and shadowing and watching someone do it once and figuring it out from there.
IDC estimates that Fortune 500 companies lose $31.5 billion annually from failing to share and preserve this kind of knowledge — and that was before organisations began accelerating AI
investment on top of an already leaking foundation.
Two things become possible once you recognise this.
The first is continuity. When the knowledge your best people carry is captured and structured, it stops being fragile. Processes stay reliable through transitions, onboarding gets faster, and the expertise that took years to build doesn't have to be rebuilt from scratch every time someone moves on.
Research across more than 300 studies consistently shows that organisations with strong knowledge transfer practices outperform those without, particularly in knowledge-intensive roles where accumulated expertise is the product.
The second is genuine AI readiness. AI systems perform well when they have structured, context-rich inputs — a clear picture of how work actually gets done, not just how it's documented.
Only 12% of organisations currently have data of sufficient quality to support AI applications. The ones that close that gap first are the ones that will see real returns on their AI investment rather than watching pilots stall at the production threshold.
Sugarwork was built for exactly this.
It captures the knowledge your people carry through structured conversations and turns it into something usable: process maps, role clarity, decision logic, dependency structures. Not a transcript. Not a meeting summary. Something a team can actually work from, that an onboarding programme can be built around, that an AI system can operate against.
The point isn't documentation. It's closing the gap between how your organisation thinks it works and how it actually works, before a transformation initiative runs into it the hard way.
Brady was right. The secret sauce is in the uncoded layer. The organisations that find it first, and structure it into something usable, are the ones that will get the most out of what comes next.
