Where AI Actually Pays: Innovature BPO COO Shares Three Questions to Ask Before Automating Anything

At AusCham Vietnam’s executive workshop on AI and customer experience, Quang Doan argued that the largest returns from AI sit in the back office, and that most failed pilots fail before any technology is involved. 

Most organizations now accept that AI can improve how they operate. The harder question, and the one that decides whether the investment pays, is which process to improve first. 

That question was the focus of a session by Quang Doan, Chief Operating Officer of Innovature BPO, at the AusCham Vietnam executive workshop “The Future of Customer Experience: AI, Automation and Smarter Decision Making”, held on 1 October 2026 in Ho Chi Minh City, which brought together business leaders, technology specialists and operations professionals. 

The back office is where the return is measurable 

Mr. Quang’s starting point was a gap many attendees recognized: attention and budget go to customer-facing AI, while the process-driven work behind it, in finance, data processing and service operations, receives far less scrutiny. Yet that is where results are easiest to measure, in accuracy, turnaround time and capacity. 

Hotels, factories, clinics and schools run very differently at the front, he noted, but their back offices repeat the same steps every day: invoices arrive, data is re-keyed, approvals are chased, month-end waits. Because that work is countable, it can be measured, priced and improved. 

Three questions before any automation decision 

Rather than begin with technology, Mr. Quang offered a test any organization can apply to any process before buying a tool or engaging a provider. 

1. Does it have a unit? Invoices, transactions, claims and enrolments can all be counted. If a process cannot be counted, it cannot be measured, automated or priced, and most failed pilots fail at this question first. 

2. Is the rule written down, or does it live in someone’s head? If the rule has never been documented, AI will fail until it is extracted. “That extraction is the actual project,” he told attendees. “The software is the easy part.” 

3. Who eats the error? A wrong email costs an apology. A wrong payment costs cash. The answer determines how much control, auditing, and human review need to sit around the machine. 

A process that fails the first question should not be automated yet; one that fails the second needs its rules written down before any software is bought. 

How Innovature BPO applied the test to a 1,000-location AP operation 

Accounts payable shows the test in practice: every business has it, the invoice is a countable unit, and the error is cash leaving the business. Where it usually fails is question two, because approval thresholds, GL coding and exception rules are rarely written down. 

He pointed to a US veterinary healthcare network of more than 1,000 locations running four separate finance systems. Innovature BPO took over accounts payable and expense processing but started with the rules, not the volume: its team documented how each site actually coded, approved and paid, then consolidated that into one policy signed off by the client, with a named owner for every exception type and weekly KPI reporting. 

The operation now handles around 29,000 transactions a month at a 1.5% error rate against a 4% industry benchmark. When the client later replaced its finance platforms, the same rules carried over without being rebuilt, and approval and payment release stayed with the client throughout.