Weekly letters · Issue 1

Does AI pay here? Five questions to start 2027

Paul K.H. Kim ·

For years the question about AI was “Can it?” Can it write, reason, code and plan? By 2026 the answer was clearly yes. The question boards, ministers and business owners ask now is different: does it pay, and who answers for it?

This letter is about that question. Each week I look at what is changing in AI across the United States, China and Korea, and what it means for your organisation and your career. It draws on my book, The AI Triangle 2027, which ends not with forecasts but with three playbooks: one for executives in large organisations, one for owners and founders of smaller firms, and one for professionals.

You may wonder why a book about 2027 makes no forecasts. Chapter 1 grades the loudest AI predictions made about 2026, and the lesson was humbling. The technology moved faster than expected, while adoption, robots and returns came more slowly than the boldest voices promised. A plan that works whichever way 2027 turns out is worth more than a guess about which way it will.

The gap

Companies spent about five times as much on AI in a year, yet the share seeing any profit from it stayed at about 37 percent. Only about four in 33 AI trials ever reach real use. None of the best-known AI failures of 2025 and 2026 came from a weak model. Each was a management failure that a simple check would have caught.

That is good news. Management failures can be fixed, in a ministry, a multinational or a firm of fifty people.

Five questions to find your gaps

Each playbook in the book opens with five questions. These are the ones for a large organisation, public or private. Count a yes only if you could show the evidence to your board tomorrow.

  1. Is one named executive accountable for AI across the organisation, with a budget and the authority to stop projects?
  2. Do you have a complete list of your AI pilots and tools, with what each one costs?
  3. Has at least one AI system been running in daily work for more than six months, with its result measured in a number your finance team already tracks?
  4. Does every live AI system have a test set, a fixed list of questions with known right answers, that is run before every change?
  5. Does every AI tool have a monthly spending ceiling, per person and per team?

Zero or one yes: not started. Two: experimenting. Three or four: proving. Five: scaling. Then look at your nos. The first one is where to start.

A smaller firm has its own five, and they are simpler. Does one person own the job of driving AI? Is everyone who uses AI on a business account with a spending limit? Is one painful process written down, with what it costs today?

What question 3 looks like in practice

When my team built an AI gate that reads container numbers at a port terminal, we agreed the number before the pilot began: 95 percent recognition. We ran it on one lane, beside the old system, so nothing in operations depended on it. The first version read only about two containers in three. We studied the failures, labelled new images and retrained, round after round, until it reached the target. Only then did we ask for the roll-out budget, and the numbers won it.

It needed no better model than anyone else could buy. It needed one owner, a number agreed in advance and a pilot that could fail cheaply.

How this letter works

Each week I will take one signal from the news and say what to do when you see it, drawn from the “If this happens, do this” tables in the playbooks. For example: if your main AI vendor raises its prices, run your test set on the next-cheapest model that meets your data rules, and switch if it passes.

Some dates are already known. On 1 January 2027 new US state rules on AI start. Around 22 January the grace period on fines under Korea’s AI Basic Act can end. On 2 December the EU’s rules for high-risk AI begin. Each will get its own issue when it lands.

Your turn

Which of the five questions did you answer no to first? Tell me on LinkedIn. I will use the most common answers to choose what this letter covers next.