Trust in government AI will increasingly depend on clear examples of where, when and how we choose to deploy artificial intelligence and where we choose not to, according to one of the public sector’s largest technology suppliers.
SAP’s vice president for global public sector Ryan van Leent told the Commercial Disco podcast that building trust will matter as much as technical innovation, as government agencies inch closer to deploying AI at scale.
“Just because we can do something with AI doesn’t mean we should,” he said.
But he stressed that good governance isn’t about being risk-averse. Instead, it’s about choosing the right tools, at the right time, in the right context to deliver real public value.
Mr van Leent pointed to SAP shelving a highly accurate emotion-detection tool, because the risks outweighed its potential benefits, describing the decision as an example of responsible AI shaping long-term trust rather than holding innovation back.
“We built emotional AI capable of detecting human emotion from facial expression and tone of voice that had more than 70 per cent accuracy,” he said.
“But we decided not to productise it. The risk was too high.”
It’s an unusually candid window into the judgement calls shaping the next era of AI deployment and a reminder that trust grows not just from what technology can do, but from how it is deliberately applied.
SAP now detects human sentiment based on “what you say not how you say it”, Mr van Leent said, precisely because public trust depends on companies showing both where they can innovate and how they set boundaries.
That distinction matters in a political environment still marked by the trauma of previous automated decision-making failures.
Mr Van Leent said public servants remain understandably cautious, and strong AI ethics policies mean little without clear transparency and human oversight.
To support that shift, SAP provides an “AI Analysis” view inside its enterprise products, showing the steps an agent takes, the data sources it draws from, and the reasoning behind its recommendations.
“Human oversight depends on AI transparency,” he said. “Users need to understand how the agent got there.”
Trust isn’t the only hurdle. Mr van Leent highlighted the sheer number of AI prototypes that never reach production.
Despite AI now supporting around a quarter of tasks within Australian organisations, forecast to exceed 40 per cent within two years, most government efforts remain stuck at proof-of-concept stage.
“We need to get AI out of prototypes and into production,” he said. “There are three things required: shifting to embedded AI, uplifting skills and confidence, and building public trust.”
Embedded AI is where he expects real scale to emerge.
“Embedded AI will scale in a way custom AI solutions never can, because it’s ready to go out-of-the-box,” Mr van Leent said. “This is what gets AI beyond experimentation.”
Global examples already show what production-level AI can do in government. in Germany, Hamburg’s Ministry of Finance uses machine learning to support processing of social benefit applications, saving 33,000 hours of manual review.
The city of Antibes in France uses AI for budget optimisation, far exceeding human capacity by making 138,000 AI-supported decisions.
“Those are the kinds of scenarios where AI is already delivering real value,” Mr van Leent said.
The next shift will be more radical, with Mr van Leent predicting AI will become “the new UI”. Instead of clicking through CRM or ERP screens, users will interact with AI copilots that navigate systems on their behalf.
“The user interface becomes largely irrelevant,” he said. “You’ll interact with the copilot, and it will represent you to the application and execute transactions for you.”
SAP is an industry partner in a new Australian Research Alliance for Enterprise AI, involving the University of Queensland, QUT, UNSW, the University of Sydney and the University of Melbourne.
The alliance is exploring what the transition to agentic AI will mean for large-scale government operations and where the guardrails need to sit.
Despite concerns about talent shortages and soaring salaries, Mr van Leent said the public service is already ahead of the general population on AI literacy.
“Eighty-four per cent of Australian public servants say they’re ready to use AI,” he said. “The next step is building confidence through applied scenarios.”
His advice: don’t start with ambitious, public-facing services. Instead, overhaul back-office processes and everyday tasks that build familiarity without risking public confidence.
“There’s enormous business value in technically simple AI capabilities,” he said. “That’s where experience and trust grow.”
Do you know more? Contact James Riley via Email.
Thank you Ryan and Trish for this thoughtful piece. It is not always that we see such excellent alignment between industry and research. The challenge of getting AI from PoC to Production is at the forefront of our work at the University of Queensland’s Centre for Enterprise AI, working together with our Alliance members QUT, UNSW, UMelb and USyd (enterpriseai.org.au). I totally agree with your emphasis on choice of the use-case – just because we can, doesn’t mean we should – well said! We look forward to working with SAP on what the transition to agentic AI means for large-scale government operations.