When to Automate and When Not To: A Framework for Leaders
- Automate tasks where "correct" has a stable, checkable definition. Keep judgment human where "good" depends on context that shifts case by case.
- Only 23% of Latin American organizations report any economic value from AI — a sign that automation decisions are often made without this filter.
- Latin America scored lowest in "talent" of six AI-competitiveness dimensions in a 2026 WEF/McKinsey survey — automating without the right people to own exceptions backfires.
- Three questions before automating anything: Is it a decision or a task? Does "good" have a stable definition? What breaks if it's wrong?
Automate the tasks where "correct" has a stable, checkable definition. Keep the judgment human where "good" depends on context that shifts from one case to the next. That's the whole rule. Most automation mistakes happen because a leader never separated the two.
Start with the real question: task or decision?
A task has one correct output that doesn't depend on who's asking — formatting a report, summarizing a document, routing a request. A decision has a correct output that depends on judgment, stakes, and context — who to hire, how to respond to an angry client, whether a number is good enough to act on. Automate the first category aggressively. Treat the second category as AI-assisted, not AI-owned.
Three questions before you automate anything
- Is this a task or a decision? If the correct output changes depending on who's judging it, it's a decision.
- Does "good enough" have a stable definition, or does it move with context? Unstable definitions need a human checking the output, every time.
- What actually breaks if this is wrong — and who finds out first, the customer or the team?
Why this matters more in Costa Rica and Mexico right now
Only 23% of Latin American organizations report any economic value from AI at all; just 6% report significant value (WEF/McKinsey, January 2026). A lot of that gap is companies automating decisions that needed to stay tasks, or leaving genuine tasks unautomated out of caution. The three-question filter above is what separates the 23% from everyone else.
The talent trap
Latin America scored lowest in "talent" of six dimensions measured for AI transformation readiness in the same WEF/McKinsey survey — employers cited unclear talent-needs vision, weak recruiting, and no clear AI career paths as major barriers. Automating a process doesn't remove the need for a skilled person to own its exceptions. If anything, it raises the bar for that person, because the easy cases stop reaching them.
This is the Realistic Design stage of CRAFT
Sequencing what to automate first — instead of automating everything at once — is the second stage of the CRAFT cycle: assess what's actually feasible, and sequence the work so it compounds instead of overwhelming the team that has to live with it.
FAQ
How do I know if something is a task or a decision?
Ask whether the correct output changes depending on who's judging it. If two competent people would agree on the one right answer regardless of context, it's a task — automate it. If the right answer depends on stakes, relationships, or judgment, it's a decision — keep it AI-assisted, not AI-owned.
Why do so few Latin American companies get real value from AI?
Only 23% report any economic value and just 6% report significant value (WEF/McKinsey, January 2026). Much of the gap comes from automating the wrong category of work — decisions treated like tasks, or tasks left manual out of caution — rather than from the technology itself.
Does automating a process reduce the need for skilled people?
No — it raises the bar for the people who own what's left. Latin America scored lowest in "talent" of six AI-competitiveness dimensions in a 2026 WEF/McKinsey survey; automation without a clear owner for exceptions is a common failure mode.
Should we automate everything we can, as fast as we can?
No. Sequence it — assess what's actually feasible and order the work so it compounds instead of overwhelming the team. That sequencing is the Realistic Design stage of the CRAFT methodology, not an afterthought.
- World Economic Forum — Latin America and the Intelligent Age — World Economic Forum
- Índice Latinoamericano de Inteligencia Artificial (ILIA) 2025 — CEPAL