How to get ROI from AI implementation.

Every board in Southeast Asia is asking the same question this year: what are we getting for our AI spend? Many CEOs cannot answer it. That is not because AI does not work. It is because most AI was bought as a tool and rolled out as a licence, not designed as a change to how the company makes money.

What does the evidence say?

PwC surveyed 4,454 CEOs in 95 countries and territories for its 2026 Global CEO Survey. Only 12% said AI had delivered both cost and revenue benefits. A further 33% saw a gain in one or the other, and 56% reported no significant financial benefit to date.

The same survey points to the cause. CEOs with strong AI foundations, meaning responsible AI frameworks and technology that integrates AI across the enterprise, were three times more likely to report meaningful financial returns. Leaders seeing both gains were two to three times more likely to have embedded AI across products, services, demand generation and decision-making.

Where does AI ROI actually come from?

In our work with enterprises across Southeast Asia, the return comes from four decisions, taken in this order.

1. One number. Start from a revenue or cost line the CFO already reports: speed to lead, win rate, cost per resolved ticket, days sales outstanding. If a use case cannot name its number, it is an experiment.
2. One redesigned workflow. Put the agent inside the work, with clear handoffs to people. Adding a chatbot to a broken process only makes the broken process faster.
3. One customer record. Agents are only as good as the data and context they read. Connect the CRM, messaging channels and core systems before you scale.
4. One owner. A business leader, not IT, is accountable for the number and reviews it every month.

Why do pilots stall?

Pilots are designed to prove the technology works. It usually does. What they rarely prove is that the business will change around it. Teams keep the old process running beside the new one, nobody retires the manual step, and the saving never reaches the P&L. We cover this pattern in From AI pilots to P&L.

How long should payback take?

Set the expectation by use case, not by vendor promise. Agree a baseline before launch, a target and a review date. If the number has not moved by the review, change the design or stop. Stopping a weak use case is a good outcome. It frees budget for the one that works.

What should the CEO personally own?

Three things. The choice of the first two or three use cases, ranked by value. The decision to change roles and processes, which only the top can make. And the scorecard: a short monthly view of the numbers each agent is meant to move, using the five metrics that prove AI is paying.

What does good look like after a year?

A short list of agents, each tied to a number the board already tracks, each with an owner, and each showing movement against its baseline. Fewer tools, not more. Manual steps retired, not running in parallel. That is what the 12% in PwC's survey have in common.

What to do on Monday.

1. List every AI tool and agent you pay for, with the number each is meant to move. Flag the ones with no number.
2. Pick one revenue workflow, such as lead response or service resolution, and record today's baseline.
3. Name one business owner for that workflow and give them the target.
4. Check what customer data the agent will read, and from how many systems. Fix that first.
5. Put the first monthly review in the diary now.