Australian Payments Plus, known as AP+, was formed in 2022 when three of Australia's core domestic payment organisations merged: BPAY Group, eftpos, and NPP Australia. It now operates the infrastructure millions of Australians rely on every time they tap a card, pay a bill, or transfer money in real time.
This context matters when reading AI adoption numbers. AP+ does not have the luxury of moving fast and fixing things later. It works under financial regulation, cybersecurity obligations, and scheme rules that govern how banks, merchants, and payment processors all connect. If something breaks or a decision is wrong, the consequences flow across the entire financial system.
That is what makes AP+'s ChatGPT and Codex results more credible than the typical AI case study. Organisations under this kind of pressure do not adopt tools lightly.
The productivity results come from an internal survey. Across the staff using ChatGPT Enterprise, 77% report saving more than 2 hours each week. Eighty per cent say their creativity or work quality improved. AP+ employees have also built over 300 custom AI assistants and more than 1,000 tailored project spaces inside the platform, which points to wide adoption rather than a handful of power users skewing the numbers.
The sharper results come from Codex. OpenAI describes Codex as an AI that can write and run software code, investigate systems, and complete technical tasks that would otherwise require a skilled developer to do manually. At AP+, a team used it to trace a subtle timing error buried across payment logs and reconciliation records. That investigation dropped from 4 hours to 30 minutes. Separately, product teams that used to spend days or weeks building early-stage simulations of payment journeys now build them in a single day.
These are not theoretical efficiency gains. In financial services specifically, a 4-hour investigation taking 30 minutes means faster resolution for the banks and businesses that depend on AP+'s infrastructure.
There is a broader trend here. Recent data on Codex adoption across enterprises shows non-developer workers are now adopting the tool at a pace far exceeding the rate at which software engineers first picked it up. Legal teams, finance departments, and operations staff are using it for tasks like data analysis, structured reporting, and process automation. AP+'s experience fits that pattern: the tool began with technical teams and is spreading outward.
The governance approach AP+ describes is worth noting for any organisation in a regulated sector. They focused on giving employees approved, secure tools rather than leaving people to experiment with consumer-grade AI on their own. They brought privacy, security, and compliance teams in at the start rather than after the fact. In financial services, where employees might otherwise paste sensitive data into public AI tools without thinking through the risks, this structure matters.
The practical lesson is not that ChatGPT and Codex are magic. It is that a regulated organisation can deploy them at scale if the rollout is treated like any other compliance-sensitive process: governed, monitored, and communicated clearly to the teams accountable for risk. AP+ got broad adoption by letting teams learn from examples within their own work, rather than through generic training sessions.
For business operators outside finance, the AP+ case makes the most concrete argument for starting: the time savings are real, the quality improvements are measurable, and the risk of not starting is falling behind organisations that already have a year of internal learning behind them.