The cost of not automating: how to think about it without the hype

Photo by Markus Spiske

Every article about automation ROI comes with numbers. "Businesses that automate save 30% of operational costs." "AI reduces processing time by 70%." These figures appear everywhere and they mean almost nothing — they're averaged across wildly different businesses, processes, and implementations.

I want to try a different approach. Not made-up percentages — just a way to think through whether automation is worth it for your specific situation, using your own numbers.

The basic framework

The cost of not automating a process is, approximately: the ongoing cost of doing it manually, minus the cost of automating it. If the first number is larger than the second, automation is worth considering. If not, it isn't. This is obvious in principle. The hard part is getting honest numbers.

Step 1: What does the process actually cost you today?

Pick one process. Estimate, as honestly as you can:

• How long does it take per instance? Not the fast version, the average version including the time to gather information, handle interruptions, and catch errors.
• How often does it happen? Per day, per week, per month.
• What does the person doing it cost per hour? Fully loaded — salary, benefits, office costs, whatever fraction of your overhead you want to attribute.

Multiply those three numbers together. That's your rough baseline manual cost per period. It won't be precise — but it will be in the right order of magnitude.

Now add: the cost of errors. How often does the process produce errors that require fixing? What does fixing them cost — in time, in customer impact, in downstream problems? This is often the number people undercount.

Step 2: What would automation actually save?

Automation doesn't eliminate all costs — it changes where the cost goes. A well-built AI agent can handle the bulk of a process autonomously, but you still need: someone to monitor it, someone to handle escalations, someone to update it when things change.

A realistic estimate for a well-built automation: it reduces the time-per-instance on the manual work significantly, probably to near zero for the typical case. The remaining human time is for exceptions and oversight. If your process currently takes two hours per day spread across your team, a good automation might bring that to 15–20 minutes of exception handling and monitoring. That's your savings estimate.

I'm deliberately not putting a percentage on this because it varies too much. Some processes are 90% automatable. Others have complex exceptions that require human judgement half the time. The only way to know is to scope it properly.

Step 3: What does automation cost?

Build cost plus ongoing maintenance. At SynthetixBiz, project costs start from €1,500 and scale with complexity. Monthly maintenance retainers start from €300. You can amortise the build cost over the period you expect to run the agent — typically 2–3 years is reasonable before major re-engineering.

Add: the opportunity cost of your team's time during the scoping and testing phase. This is usually a few hours of your time, not weeks.

What the calculation actually looks like

Let me walk through a real-ish example. A company has one person spending roughly 2 hours per day processing inbound supplier documents. That person costs about €40/hour fully loaded. So: €80/day, roughly €1,600/month, €19,200/year.

Build cost: let's say €3,500 for a properly scoped, properly tested document processing agent. Monthly maintenance: €400. Year one total: €3,500 + (€400 × 12) = €8,300.

Year one savings: €19,200 − €8,300 = €10,900. From year two, savings are €19,200 − €4,800 = €14,400 per year.

Those are illustrative numbers, not guarantees. The actual savings depend on how much manual time the agent genuinely replaces in your specific process. The point is: when you run the arithmetic with real numbers from your business, the case for automation either holds up or it doesn't. The vague percentages from industry reports don't help you make that decision. Your own numbers do.

When the numbers don't add up

Sometimes they don't, and that's fine. If the process only happens twice a week, or the people doing it are cheap to employ, or the build would be complex because your systems are hard to integrate — the ROI might not be there yet. Better to know that before investing, not after.

The discovery call is partly for this. I'd rather tell you the economics don't work than build something that doesn't pay back.

Automation is worth doing when it's genuinely cheaper than the alternative, for long enough to justify the build and change cost. Not because it's "the future." Not because your competitors are doing it. Because the numbers make sense.