Which processes to automate: 5 signs yes, 2 no

The question isn’t whether to automate. It’s what to automate first, without burning budget on processes that don’t deserve it.

This is the operational checklist we use in the first calls with a COO. No FOMO: two of the seven signs tell you to stop.

In short:

• A process is ready for automation when it’s high-volume, repetitive, rule-based, with structured output and a measurable baseline.
• As a rule of thumb (from our experience): above ~20-30 repetitive tasks per week per person automation starts to pay off.
• Two signs say do NOT automate: the process will change within 3 months, or every case is an exception requiring human judgement.
• Without a timed baseline, any savings estimate is an opinion, not a number.
• At Navily, automating moderation and enrichment cut −70% of operational time: all 5 “yes” signs were present.

The 5 signs a process is ready

1. High volume, high repetition

If one person repeats the same task dozens of times a week, it’s the first candidate. The value of automation grows with volume.

Rule of thumb (from our experience, not a market stat): under ~10 tasks/week it’s rarely worth the sprint; above ~20-30, yes.

2. Clear rules (even if complex)

The process doesn’t need to be simple. It needs to be describable. If you can tell a new hire “when X happens do Y, except in cases Z”, an agent can learn it.

If the answer is “it depends, you pick it up over time”, the process isn’t ready yet: it needs to be codified first.

3. Structured output that flows into your systems

The best processes to automate produce something that lands in a CRM, ATS, ERP or management software (e.g. TeamSystem, Zucchetti, Odoo). Ticket triage, invoice parsing, CV screening, lead enrichment.

If the output is manual copy-paste between tools, you’re paying a person to shuttle data. That’s exactly the work an AI agent removes.

4. High-volume unstructured text as input

CVs, tickets, emails, PDFs, reviews, property listings. When the bottleneck is reading and interpreting large volumes of text, AI delivers the most value.

This is the case for Customer & Compliance Automation: triage and moderation of volumes a human team can’t handle without headcount growth.

5. A measurable baseline exists (or you can create one)

This is the sign most people skip. If you don’t know how much the process costs today, you won’t be able to say whether automation worked.

A real baseline = timed measurement on 10-20 real tasks, not “it feels like it takes us half a day”. One week of work that unlocks everything else.

The 2 signs that say NO

Here I do the opposite of what an AI salesperson does.

NO 1. The process will change within 3 months

If you’re replacing the management system, redesigning the workflow, or the market is shifting the rules under your feet, don’t automate now. Building an agent on an unstable process means rebuilding it next quarter. Wait until it stabilises.

NO 2. Every case is an exception requiring judgement

Senior executive search, complex negotiations, strategic decisions. Where recurring rules don’t exist, automation doesn’t just fail to help: it introduces hard-to-audit errors. Here the human stays central, full stop.

The real case: Navily

The boating community Navily had all 5 “yes” signs: high volume of user content to moderate, clear moderation rules, output flowing into their systems, massive text input, measurable baseline.

Result: −70% operational time on moderation and enrichment. No magic, just a match between the signs and the intervention.

How to apply it in practice

Run this exercise on 3-4 processes: score each against the 5 “yes” signs and check the 2 “no” ones. The process with the most “yes”, zero “no” and the easiest baseline to measure is your first sprint.

If you want help with the scoring, the 3-minute check-up gives you a first read.

CTA

Want to figure out which of your processes goes first? Take the check-up (3 minutes) or see how we worked on high-volume processes with our AI agents.