You're Probably About to Automate the Wrong Thing
The CEO reads about automation, gets excited and picks the most complex cross-department process as the first target. Six months and $200K later the project has stalled, and the company decides automation doesn't work for them.
What went wrong was usually the choice of first process. Companies that get results start with something dull: a small, repetitive process they can measure. They get that one working properly and only then expand.
Step 1: Watch What Actually Happens
Process documentation is usually around 18 months out of date. People build workarounds, skip steps and add informal checks that never get written down.
Sit with the person who does the work and watch. Write down what they actually do, even where it differs from the SOP. Their workarounds show you where the process is really broken.
A data entry task that everyone calls "15 minutes" can take 45 when you watch it, because the person doing it is cross-referencing three spreadsheets nobody else knew about.
Step 2: Measure Before You Touch Anything
Get hard numbers:
- Volume: How many times per day/week/month?
- Duration: End-to-end cycle time?
- Labor hours: Person-hours per cycle?
- Error rate: How often do mistakes happen? What do they cost?
- Wait time: Active work vs. sitting in someone's inbox?
Wait time usually tells you the most. A "3-day process" often contains 2 hours of actual work and 2.5 days of waiting for approvals. Automation won't fix a queue problem, so fix the workflow first.
Calculate the annual cost of the manual process. That's your ROI baseline.
Process mining tools can speed this step up a lot. They read the event logs from your existing systems, map how the work really flows, and show where it gets stuck and for how long. Weeks of observation can become a KPI dashboard in a few days.
Step 3: Score and Pick
Rate each candidate on four axes:
Complexity (low = good): Simple, linear processes automate faster and fail less.
Volume (high = good): A process that runs 500 times a day pays back far faster than one that runs twice a week.
Standardization (high = good): If the steps and formats are the same every time, it's easy to automate. If every instance is a special case, expect the automation to be fragile.
Impact (high = good): Time saved, errors prevented, compliance improved, people freed up.
Your first target should score low on complexity and high on the other three.
Mistakes That Kill Projects
Automating broken processes. If the manual process has wasted steps, the automated version runs the same wasted steps faster. Fix the workflow, then automate the fixed version.
Skipping the pilot. Start with one team in one location, on a subset of the work, and watch it closely for two weeks. Fix what goes wrong, then expand.
Ignoring the humans. Automation changes jobs. If people feel threatened instead of supported, they'll work against it, whether they mean to or not. Bring them in early.
Fantasy timelines. Simple processes take 4-8 weeks. Complex cross-system flows take 3-6 months. Plan with those numbers.
Ignoring the integration layer. Most automation connects systems through webhooks, API integrations and data pipelines. No-code and low-code platforms like n8n, Make and Zapier handle straightforward ETL flows well. When you need custom data transformations or error handling across several APIs, budget engineering time on top of the drag-and-drop configuration.
When to Leave It Manual
Some things shouldn't be automated:
- Changes constantly: you'll spend more time updating the bot than it saves
- Needs empathy or negotiation: leave it with people
- Runs once a month for an hour: the ROI will never justify the investment
- One mistake = catastrophic damage: keep a person overseeing it
The Playbook
Pick one boring, high-volume process and map how it runs day to day. Measure it before you automate it, so that afterwards you can show the difference in numbers. Then use that result to make the case for the next one, and keep going one process at a time.