There's a quiet gap opening in finance tech conversations. On one side: adoption metrics — who's using something that looks like AI. On the other: decision advantage — whether a call got better, faster to defend, or cheaper to reverse when wrong.

Those are not the same.

What the sentiment numbers actually say

Deloitte Australia's CFO Sentiment Survey (Edition 21, 3 June 2026) reported roughly 90% of CFOs using AI in some form, but only 16% using it extensively. Use-case depth was concentrated in familiar places: invoice processing 32%, FP&A 30%, contract analysis 24%.

Read that as a peer, not a headline. Broad touch, thinner depth. Invoice and FP&A tools can save hours and still leave the quality of the cash call, the credit call, or the board narrative unchanged — if the inputs and ownership underneath are fuzzy.

I'm not anti-tooling. I'm pro-foundation. Soft on vendors: many of these products help. Sequencing and evidence decide whether help becomes advantage.

A desk test that cuts through the slideware

For any AI workflow in flight, ask:

  1. What decision does this change? Name it in plain language (e.g. "we release credit earlier with the same loss rate," not "we unlock insights").
  2. What data does it rely on — and who owns the definition? If the answer is "the data team" with no named finance owner, you're renting a demo.
  3. Propose vs approve: what can the tool draft, and what must a human sign?
  4. What's the metric in the pack? Hours saved is fine. Better is variance explained, exception cycle time, forecast error, or loss events — something a CFO would keep reporting after the pilot buzz fades.
  5. Can we reconstruct the call? Same audit-first test as any agentic path: inputs, logic, approval, log.

If you can't clear those five, you may still have useful task automation. You don't yet have decision advantage.

Where adoption tends to stop at "helpful"

Invoice coding and extraction often win on touch-time. FP&A assistants draft variance commentary. Contract tools surface clauses. All good — and all easy to mistake for strategy. Hours come back; the quality of the call stays flat.

Advantage shows up when AR timing assumptions in the 13-week actually improve because collections signal feeds the forecast; when recon exceptions fall because masters got cleaned, not because a model hid the break; when credit limits move with evidence you'd show a lender. That's the gap between "we use AI" and "we decide better" — and it's the gap CFOs get asked about once the pilot buzz fades.

What to do this quarter

Pick one workflow that's already "adopted." Run the five-question test. Keep what passes. For what fails, fix ownership and data definitions before you buy the next licence. Report one decision metric beside the adoption story so the board hears both.

That's how Brisbane and broader AU desks can stay practical: use the tools, keep judgement human, and don't confuse activity with edge.

CTA: Measuring the same gap between adoption and advantage? Join the conversation at financesignal.ai — peer thinking for finance people who have to live with the calls.