Automation
What to automate first, and what to leave alone
Most automation projects fail because they start with the most visible process rather than the most expensive one. A practical way to choose.
By TezCommerce · · 6 min read
The first automation conversation almost always starts in the wrong place. Someone has seen a demo, they are excited about a specific tool, and they want to know whether it can be applied to the process that annoys them most. Annoyance and cost are not the same thing, and the gap between them is where automation budgets go to die.
Score before you build
Before scoping anything, we ask a client to list every recurring manual process and score each one on four axes. It takes an afternoon and it consistently changes the priority order.
- Frequency: how many times a week does this happen?
- Duration: how long does one instance take, honestly, including the context switch?
- Error cost: what happens when someone gets it wrong? A typo in a caption is not a typo in a remittance reconciliation.
- Stability: has this process changed in the last six months? Automating a process that is still being invented is how you build something you throw away.
Frequency times duration gives you hours. Error cost tells you whether those hours are the whole story. Stability tells you whether to build now or wait. High on the first three and stable on the fourth is where you start, regardless of which process is most irritating.
What automates well
- Moving structured data between systems that do not talk to each other. Orders into accounting, leads into CRM, courier status back into the store.
- Reconciliation. Comparing two lists and flagging what disagrees is exactly what computers are for and exactly what humans are worst at.
- Scheduled reporting. If someone builds the same summary every Monday, that is a script.
- Triggered notifications. Stock below threshold, order stuck in transit, payment failed, lead gone cold.
- First-draft generation. Product descriptions, reply drafts, report commentary, where a human reviews before it goes out.
What to leave alone
Anything where judgement carries real consequence and the process is not yet stable. Anything a person does five times a month: the build will never pay back. Anything that requires a human relationship, which includes most of your key account communication. And anything where the current process is broken: automating a bad process gives you a bad process running faster, which is worse, because now it is harder to see.
Fix the process, then automate it. In that order, always.
Build for failure from day one
The difference between an automation that helps and one that hurts is what happens when it breaks, and it will break, because an API will change or a credential will expire. A workflow that fails silently is far more dangerous than no workflow, because your team has already stopped checking.
- Every workflow alerts a named person on failure. Not a shared inbox.
- Retries with backoff for transient errors, hard stop and alert for real ones.
- A log of what ran, when, and what it touched.
- A documented manual fallback, so a broken automation is an inconvenience rather than an outage.
Start with one
Pick the highest-scoring process, build it properly, run it for a month, and measure whether the hours actually came back. One well-instrumented automation that your team trusts is worth more than six that nobody is sure are still running.
