You Can't Automate a Mess

Everyone wants to talk about automation. AI agents, workflow builders, no-code tools that promise to 10x your operations. And some of them actually deliver.

But here’s what nobody wants to hear: if your process is a mess, automating it just gives you a faster mess.

The prerequisite nobody mentions

Automation works when the work is standard. When the steps are the same every time, the inputs are predictable, and the outputs are defined. When that’s true, automation is cheap and fast and the ROI is obvious.

The problem is that most businesses trying to automate aren’t there yet. Their processes have exceptions baked in. Somebody “just knows” which orders go to which supplier. There’s a spreadsheet with rules that only one person understands. Half the workflow runs through email threads.

That’s not a process. That’s institutional memory held together with good intentions. You can’t hand that to a machine.

Standard vs. non-standard work

This is the distinction that saves businesses a lot of money and frustration:

Standard work follows the same path every time. Order comes in, system checks inventory, routes to the right supplier, generates a shipping label. Same steps, same logic, same outcome. This is automation’s sweet spot.

Non-standard work requires judgment. A customer wants to change their shipping address after the order was placed. A supplier is out of stock on one variant but not another. A new product doesn’t fit the existing pricing model. These require a human to evaluate the situation and make a call.

The mistake is trying to automate the non-standard work before you’ve standardized it. The automation doesn’t know what to do with the exceptions — because you haven’t decided what to do with the exceptions. You’ve just been handling them ad hoc.

Process first, then automation

The work that actually matters happens before anyone writes a line of code or configures a workflow:

Name the process. If you can’t describe the steps, you can’t automate them. Write it down. Not in a fancy tool — a document, a whiteboard, a napkin. “When X happens, we do Y, then Z.”

Find the exceptions. Where does the process break? Where does someone have to make a judgment call? Those are the edges of your automation boundary. Everything inside the boundary can be automated. Everything outside needs a human — for now.

Standardize first. Can you make the exceptions less exceptional? Can you create a rule for the most common edge cases? Every exception you turn into a rule is an exception you remove from someone’s plate.

Then automate. Now you have something a machine can run. The steps are defined. The edge cases are handled. The exceptions route to a human with the context they need to make the call.

This sequence — name it, find the edges, standardize, then automate — is the difference between automation that works and automation that creates new problems.

Options are valuable, but expensive to keep

There’s a related idea that comes up constantly with businesses evaluating automation: the desire to keep every option open.

“We might want to add a third supplier.” “We might switch to a different fulfillment model.” “We might expand into a new product category that works completely differently.”

All of those are real possibilities. And planning for them isn’t wrong. But every option you engineer into a system adds complexity. Complexity adds cost — in build time, in maintenance, in cognitive overhead for the people operating it.

The right question isn’t “could we ever need this?” It’s “do we need this now, and what does it cost us to keep this option open?”

Sometimes the answer is yes — the flexibility is worth the cost. Often the answer is no — you’re paying today for a future that may never arrive. Build for what’s true now. Adapt when the reality changes.

The bottom line

If someone tells you they can automate your operations and they haven’t spent serious time understanding your processes first, be skeptical. Automation is a multiplier. It makes good processes great and bad processes worse.

The unsexy work — mapping the process, finding the exceptions, standardizing the rules — is where the real value lives. The automation is just the last step.

Start with the mess. Clean it up. Then let the machines run.

Thinking about automation but not sure where to start? Sometimes a systems review is the best first step. Let’s talk.