Frank Fillmann, EVP and GM, Australia and New Zealand
Over the past three years, I’ve spoken with hundreds of Aussie and Kiwi business leaders about what AI means for their employees, customers and business growth. Today’s conversations focus on moving fast to capture the opportunity while managing trust, cost, and ROI.
Together we’ve been able to take this new incredible intelligence capability and harness it with the data guardrails and business logic they already have. Customers like Xero, ANZ Bank, and Fisher & Paykel are trailblazers, unlocking trapped value in their businesses and delivering better employee and customer experiences.
What’s become crystal clear: AI models alone cannot run a company. It’s the pairing of probabilistic AI models and deterministic systems which deliver the innovation to unlock AI’s true potential.
The next wave of AI innovation
Probabilistic AI can interpret context, generate responses and reason through complex problems. Deterministic systems provide the trusted data, business rules, permissions and workflows that organisations rely on every day.
Combined, they create something far more powerful than either can deliver independently: AI that can not only understand and recommend, but also act within clearly defined business guardrails.
Together, with dynamic, context-aware user interfaces like Claudeforce and Slackbot you have this next phase of AI with unprecedented potential.
The Australian and New Zealand context
Leaders are under pressure to improve productivity, manage rising costs and meet ever-higher customer expectations in a challenging economic environment.
While there are no silver bullets, we agree with the Treasury that AI represents one of the most significant opportunities in almost two decades to unlock new capacity, accelerate growth and improve customer outcomes across the economy.
I’m seeing some amazing trailblazers innovate and move at pace, yet so many businesses are yet to deliver true agentic transformation. They may have some failed pilots, perhaps resources spent on building their own model or vibe coding a project that works on day one then falls over.
The rest of the world isn’t waiting. Companies with growth ambitions are finding ways to transform their businesses into agentic enterprises, fast.
Agentic AI momentum
On the Salesforce platform, we’ve seen Agentic Work Units (AWU), which represents one discrete unit of work completed by an AI agent, increase by 97 per cent quarter on quarter. That’s seven billion AWUs delivered to date.
Customers are seeing real ROI today. EasyPark has reported 40 per cent year-on-year growth, supported by Agentforce. The Grout Guy has also grown 40 per cent year-on-year. Meanwhile, Sutton Tools saw online orders increase fourfold in the year they first started implementing Agentforce.
For more businesses to achieve these results, a new approach to governing the risks and opportunities of AI transformation is needed.
Trust governance: Compliant and safe AI
Employees turn to consumer AI tools when they believe enterprise alternatives cannot meet their needs. While this reflects enthusiasm for the technology, it also creates governance risks around data, security and compliance.
There may be hundreds of agents lurking in businesses today. A unified control plane, like Agent Fabric ensures security, data access, logging, and costs are fully visible and rogue AI less likely to be whirring away.
Trailhead, our free online learning platform, also supports workforce education and adoption so teams can get the most from approved and trusted AI. The answer is not simply tighter controls. It is ensuring approved enterprise AI is secure, reliable and genuinely useful in the flow of work.
Cost governance: Beyond Tokenomics
An explosion of usage has for many companies resulted in an explosion in cost. For example, Uber used its entire AI budget in just four months.
The reality is, not every task requires the most sophisticated or expensive model. There are both technical and commercial approaches which can help customers deliver both amazing innovation and prudent cost management.
For example, Agentforce originally ran on a single rented model which meant the bill for tokens grew linearly with traffic. We broke out several different tasks and tuned them to specific open-source models to accomplish them.
Today, these models run a growing share of Agentforce, cutting costs and running more effectively. Businesses need an approach that matches the right level of intelligence to the right task, balancing performance, risk and cost.
ROI governance: The five per cent problem
MIT research highlights the gap in reaching implementation and scale, with only five per cent of AI agents making it from pilot to production.
We’ve seen a 70 per cent quarter on quarter increase in agents going into production, and we’re aware how important it is for businesses to have a clear path to ROI.
Our advice is simple: start with focus and move with urgency. Initial projects should focus on quantifiable ROI rather than attempting to solve the most difficult, long-standing problems, allowing technology and function leaders to earn the right to move to larger, more complex implementations.
This is where Salesforce and our partner ecosystem’s Forward Deployed Engineers (FDEs) can also help. They are embedded in our customer’s teams to help them to remove blockers and accelerate AI adoption.
With the right approach and the right support, moving AI agents from pilot to production can happen faster, with less time spent in the sandbox.
Dreamforce 2026 is almost here
Dreamforce 2026 must be our most anticipated yet. We have a chance to accelerate Australia and New Zealand’s agentic enterprise transformation and we can’t wait to show you how.
