Australian firms roll back AI agents over governance

Sinch found that 84% of Australian enterprises have rolled back or shut down an AI customer communications agent because of a governance failure, 10 percentage points above the global average in its survey.

The findings suggest the debate over corporate AI adoption in Australia has shifted. Rather than remaining stuck in pilot projects, many organisations have already put customer-facing AI into live use and are now dealing with what happens when those systems fail governance checks after launch.

The survey found that 63% of Australian enterprises have moved AI customer communications agents into production. The research covered 2,527 senior decision-makers across 10 countries, including 264 in Australia, and focused on large organisations in sectors including financial services, healthcare, telecommunications, technology, retail and professional services.

Privacy concerns were the most common trigger for a rollback in Australia. Among organisations that had withdrawn or shut down an AI agent, 45% cited concerns that personally identifiable information might be compromised, the highest rate recorded among the countries surveyed.

Other reasons included a lack of auditability, cited by 22% of respondents, and hallucinations, cited by 17%. The figures suggest that for many companies, the main barrier is no longer whether AI tools can be deployed, but whether they can be monitored and controlled once they are interacting with customers.

The data also highlighted the operational effects of failed deployments. Some 51% of Australian firms said support teams become overloaded when AI agents fail, while 26% expected reputational damage.

Investment focus

Despite the setbacks, spending plans remain strong. Sinch found that 67% of Australian enterprises plan to increase AI agent investment by 25% or more.

That money is not being directed first toward model development. In Australia, 61% of respondents said trust, security and compliance were an investment priority, compared with 59% who pointed to AI development itself.

The emphasis reflects a broader reassessment of what is needed to make customer-facing AI workable in production environments. The survey also found that 91% of Australian organisations viewed communications infrastructure as essential or very important to the success of AI customer agents.

Wendy Johnstone, Executive Vice-President, APAC, at Sinch, said: "Australian enterprises are ahead of the global average at getting AI communications agents into production but are rolling them back at a high rate. Having pushed past 'pilot purgatory', Australian firms are now struggling to make customer-facing AI reliable and safe."

She also outlined the commercial stakes for businesses trying to stabilise their systems. "Getting this right is critical. Those that succeed will be able to deliver tailored, timely experiences at scale before their competitors can," Johnstone said.

Vendor churn

The data also points to rapid movement in the supplier market. While 83% of Australian organisations said they were happy with their current AI communications vendor, 86% said they were also exploring alternatives.

The apparent contradiction suggests buyers are keeping their options open even when existing relationships remain broadly satisfactory. For suppliers, it signals a market where contract stability may be weaker than customer satisfaction scores imply.

Another pressure point is integration across multiple channels. Australian enterprises reported planning AI use across an average of 3.6 channels, the highest level of channel diversity measured in the study, according to Sinch.

Many companies are also building parts of that framework themselves. The survey found that 60% of Australian enterprises are custom-engineering infrastructure to preserve customer context across channels, a sign that off-the-shelf systems may not yet meet every operational requirement.

The mix of high deployment rates, frequent rollback decisions and continued spending suggests a market moving beyond experimentation but not yet settled on stable operating models. Large organisations appear willing to invest further, though with more attention on risk management and oversight than on AI development alone.

In Australia, the sample was drawn from enterprises with at least 1,000 employees. Some 68% of respondents came from organisations employing between 1,000 and 4,999 people, while 32% came from businesses with more than 5,000 employees. Respondents included C-suite executives, Vice-Presidents, Directors and Managers.

Johnstone said: "Australia's AI communications landscape is rapidly maturing. Organisations are shifting their investment focus beyond the AI model itself, prioritising the governance that ensures trustworthy service. At the same time, vendor loyalty is being tested by the need to quickly address new use cases as they emerge."

She added: "The underlying infrastructure is the key differential that gives organisations the confidence they need to deploy safe, scalable customer-facing agents."