Not "adopt AI" — remove expensive routine

Bringing AI into your business · Lesson 1 / 24

The right question

Most failed AI projects start with "where could we apply AI". That's a question about technology, and it produces pilots that demo beautifully and don't survive to the second quarter.

The working question is different: what expensive routine do we have, and would AI remove it. The difference isn't cosmetic — it decides who owns the project. In the first case it's IT or an innovation team; in the second it's a manager with a real problem.

Signs of a good candidate

Repeats often — dozens or hundreds of times a week. A one-off task won't repay any implementation, however elegant.
Consumes manual hours. People spend hours on the same action.
Works on text or data — emails, documents, tickets, reviews, transcripts.
An error isn't instantly critical. There's room for a human check before the result leaves the building.
The result is verifiable. You can tell a good output from a bad one — otherwise you can neither tune it nor measure it.

Typical quick wins

First-line triage and draft replies in support.
Drafts of emails, proposals, product descriptions.
Summaries of meetings and long documents.
Turning reviews and feedback into structure.
Extracting fields from incoming documents.

Where AI usually doesn't repay

Tasks where an error is expensive and verification is costly. Tasks that happen monthly. Tasks where the real bottleneck isn't the text but the approvals between people — there AI speeds up a step that was never the constraint.

Insight. Look not at where AI is applicable but at where you're spending a lot, often. It's applicable almost everywhere; it repays in a narrow set of places.

Common mistake. Starting with the most visible task instead of the most repayable one. The visible one gets you an industry write-up; the repayable one gets you budget for the next project.

Pro tip. Ask department heads one question: "what do your people spend time on that feels pointless to them". That's a candidate list assembled in an hour.

Cheat sheet

Start with the pain, not the technology.
Candidate: frequent, manual, text-based, with room to check.
The result must be verifiable.
Rare and unverifiable tasks aren't first.

1. Which question starts a workable implementation?

2. Why does verifiability matter?

3. Which task won't repay?

Task — checked by AI

Survey yourself and at least one manager in your company and build a list of 7 routine candidate tasks. For each state: how often it repeats (per week), how many manual hours it takes, whether it works on text or data, and whether the quality of the result can be checked. Mark which candidates you're rejecting and why.

🔒 Answer the question correctly to move on to the next lesson.