Ranking AI use cases by value.

Most leadership teams in Southeast Asia do not lack AI ideas. They have too many. Every function has a list. Vendors bring more. The result is a long backlog, a handful of pilots chosen by enthusiasm, and very little that reaches the P&L.

The evidence says this is common. McKinsey's 2026 State of AI survey finds nearly nine in ten respondents report regular AI use in at least one business function, yet only 37 percent attribute at least some EBIT impact to it, and only 6 percent qualify as high performers (McKinsey). Choosing well is a large part of the difference.

Why do most AI use case lists fail?

Three patterns repeat.

• Ideas start from tools. “We should use an agent for…” rather than “we need to move this number”.
• Readiness is ignored. A high-value idea built on fragmented data stalls in month two.
• Everything is a priority. Twenty use cases get a little attention each. None gets enough to change a process.

A ranking method fixes all three by forcing the same questions on every idea.

What goes on the one page?

List every candidate use case in rows. Score each from 1 to 5 on three columns.

1. Value

Which business number does this move, and by roughly how much? Name the number: qualified pipeline, conversion rate, cost per service contact, resolution time, revenue per customer, days sales outstanding. If nobody can name the number, the score is 1, whatever the enthusiasm.

2. Readiness

Is the data behind this use case in one place, complete and current? Is the process defined? Is there written context, such as policies, product data and approved answers? Is there an owner? A use case on clean, unified customer data with a defined process scores 5. One that needs three systems joined first scores 1 or 2.

3. Risk

What happens if the AI gets it wrong? Who sees the mistake: an employee or a customer? Does it touch personal data, money, health or legal commitments? Score low risk high, so that a 5 is always good news.

An illustration of how three typical candidates might score:

The scores are a starting point for debate, not a formula. The value of the exercise is that every idea is argued on the same three questions, in the same room.

How do you read the scores?

Plot value against readiness, and use risk to decide how carefully to proceed.

• High value, high readiness: fund now. These are your first two or three.
• High value, low readiness: fix the foundation. Fund the data and process work, not the agent yet.
• Low value, high readiness: quick wins only if they are nearly free. Do not let them crowd the list.
• Low value, low readiness: drop them. Say so explicitly.

High-risk use cases in the top-right box still go ahead, but with people reviewing outputs, a smaller scope and a stricter launch gate.

Which use cases usually rank highest?

In customer-facing operations, the first winners tend to be bounded and close to data the business already holds. Info-Tech's SoftwareReviews suggests starting with bounded use cases such as “meeting capture, CRM record updates, lead qualification, or first-draft follow-up” before broader deployment (SoftwareReviews). We would add first-wave service questions on messaging channels and account research for sales.

These are not the most exciting ideas on the list. They are the ones most likely to reach production quickly and prove that the method works. See Customer Agent in practice and Prospecting Agent for enterprise sales for two examples.

How do you keep value claims honest?

Every value score should come with a one-line calculation the finance lead accepts: the number, today's baseline, and a conservative estimate of the change. “Cut first response time on WhatsApp enquiries in one market” is testable. “Transform customer experience” is not. Where the business has no baseline, the first piece of work is to measure one; that alone often reorders the list.

Treat vendor-reported averages as a sense check, never as the estimate. Your starting point, your data and your process decide your result.

Who should be in the room?

The ranking is a leadership exercise, not an IT one. Bring the CEO or MD, the heads of sales, marketing and service, finance, and whoever owns data and technology. Each function brings its candidates. Finance tests the value claims. Technology tests readiness. Legal or the data protection officer tests risk.

McKinsey reports that high performers are twice as likely as others to say their senior leaders demonstrate commitment to AI initiatives and that they have defined processes to measure the impact of those initiatives (McKinsey). The ranking session is where both start.

What happens after the ranking?

1. Name an owner for each funded use case, from the business, not the technology team.
2. Record a baseline for the business number before anything is built.
3. Redesign the process, not just the tool. McKinsey finds nearly three-quarters of high performers have fundamentally redesigned workflows (McKinsey).
4. Set a review date. Re-score the whole list each quarter as readiness improves and new ideas arrive.

Where to start

Gather every AI idea in the business on one list this month. Score it together in one session. Our AI readiness assessment runs this exercise with your leadership team and tests readiness against your actual data.