Banks waste AI spending without workflow redesign as only 10% see gains

Embedding AI into operations and simplifying processes is key to cost-income gains in banking.

Financial institutions won't see substantial returns from their artificial intelligence (AI) investments if they don't redesign workflows, as rising transaction volumes and declining headcount do not translate into proportional cost reductions.

AI optimisation is primarily an operating model issue, not a technology deployment challenge, said Christopher Saunders, Partner and head of Advisory, Financial Services at KPMG Thailand, at the Asian Banking & Finance and Insurance Asia Summit in Thailand in 28 April.

Saunders said a small group of financial institutions, which he estimated at around 10%, are beginning to realise stronger outcomes by simplifying operating models and integrating AI into redesigned workflows.

The majority, however, continue to focus on deployment rather than structural change, limiting the impact of AI investments.

Saunders said this reflects a structural issue where operational complexity offsets efficiency gains.

He noted to industry peers at the Amari Bangkok, that whilst digital adoption has reached maturity in markets such as Thailand, with transaction growth moderating to around 5% to 8% annually, cost pressures continue to build from regulatory requirements, cyber security, real-time monitoring and biometric controls, alongside recurring cloud and engineering costs.

Saunders said many institutions initially realise efficiency gains through targeted digital and AI use cases, but later add new platforms without removing legacy systems. This creates duplicated processes and fragmented workflows, limiting the net value generated.

Without workflow redesign, AI increases speed but does not improve outcomes because bottlenecks such as approvals, handoffs and manual reconciliation remain unchanged, he said.

“AI will not transform your business, but people will,” he said, adding that value is only created when AI is embedded into end-to-end workflows and decision-making is integrated into operational processes.

Saunders outlined three requirements for effective AI optimisation, namely redesigning workflows, enabling employees to use AI in daily operations, and reinforcing organisational change through accountability and adoption.

He said institutions that succeed move decisions into the flow of work, reduce manual handoffs and embed controls directly into processes, resulting in faster execution, lower costs and improved cost-to-income performance.

He added that changing outcomes requires changing the flow of work itself, not simply accelerating existing processes with new tools.

By contrast, he said organisations that add AI on top of unchanged systems only speed up existing inefficiencies, resulting in higher activity but limited performance gains.