Finding the Friction: What Actually Deserves Automation

Most automation projects fail before a single line of code ships, because the operator picked the wrong target. Signal over noise starts with target selection, not tooling.

The friction audit

Every business runs on a stack of small, repeated decisions and hand-offs. Somewhere in that stack is the friction that's actually costing you money and morale. Find it before you build anything.

Walk your own operation for one week and log every task that meets two conditions:

• It repeats. Daily, weekly, or per-transaction, not a one-off.
• It has a visible cost. Time, errors, delay, or a person you can't hire because this task eats their week.

Do not log tasks just because they feel tedious. Tedium is not the bar. Cost is the bar.

Sorting the list: three buckets

Once you have a list, sort it into three buckets:

1. High-frequency, low-judgment. Data entry, status lookups, formatting, routing. These are the best automation candidates: high repeat count, low ambiguity, easy to verify correctness.
2. High-frequency, high-judgment. Customer triage, pricing exceptions, quality review. These are automatable, but only with a human-in-the-loop pattern and a governance layer (see Episode 3), not a blind hand-off.
3. Low-frequency or high-novelty. Strategic planning, one-off negotiations, anything that changes shape every time. Leave these alone. Automating a task that happens twice a quarter is a hobby project, not an operations improvement.

The "would a new hire ask about this" test

A fast heuristic: if you handed this task to a new hire on day one with a one-page instruction sheet, could they do it correctly? If yes, it's a strong automation candidate. The task is well-specified enough that a system (rules-based or AI-assisted) can do it too. If the honest answer is "it depends, they'd have to ask me," the task still has undocumented judgment in it. Document that judgment first. Automate second.

Why this matters more than the tooling choice

Operators new to AI adoption tend to start with "what tool should I buy," when the real first question is "what task, specifically, costs me the most and repeats the most." Tools are not the mission. Capability is. Pick the friction first, and the tooling choice gets a lot easier (often a lot smaller) than the vendor pitch decks suggest.

What NOT to automate yet

Hold off on:

• Anything you haven't done manually, correctly, at least a few dozen times yourself. If you can't yet describe the steps, an agent can't either.
• Anything where an error is expensive and hard to reverse (financial postings, customer-facing commitments, legal language) until you've built a review gate around it.
• Anything that's about to change. Don't automate a process you're planning to redesign next quarter. You'll be maintaining two things instead of shipping one.

Field exercise

Spend the next five business days keeping a friction log. For each entry, record: task name, frequency per week, minutes per occurrence, and a 1-10 judgment score (1 = pure rule-following, 10 = pure expert judgment). At the end of the week, sort the log into the three buckets above and pick exactly one High-frequency, low-judgment task to carry into Episode 2 for baseline measurement. Resist the urge to pick your biggest, scariest problem first. Pick the one you can measure cleanly.