Every business that comes to us about automation asks the same question in the first meeting: what should we automate? Usually they already have an answer in mind, and usually it's the wrong one.
The instinct is to automate whatever annoys you most. That's rarely the thing that pays back fastest.
Start With Frequency, Not Frustration
A task that takes two hours and happens once a quarter costs you eight hours a year. A task that takes four minutes and happens forty times a day costs you more than 600 hours a year. The four-minute task feels trivial. It isn't.
Before you automate anything, count. How often does this happen, and how long does it take? That number tells you where to start, not your irritation level.
We ask clients to track this for two weeks before we build anything. It almost always changes what we build.
The Three Filters
A task is worth automating when it passes all three:
- It's repetitive. The steps are the same every time. If the process changes depending on who's asking, automation will break on the exceptions.
- It's rule-based. You can write the decision logic down. "If the invoice is over $5,000, route to finance" works. "Use your judgment on whether this client is a good fit" doesn't.
- The input is structured. Data arriving in a consistent format is automatable. Data arriving as free text in a WhatsApp message is a much harder problem, though no longer an impossible one.
What Usually Pays Back First
In our experience the fastest wins are boring ones:
- Lead routing and first response. Someone fills in a form at 11pm and waits until morning. An automated acknowledgment plus routing to the right person cuts that to seconds, and response time is one of the strongest predictors of whether a lead converts at all.
- Appointment reminders and rescheduling. No-shows cost real money, and the fix is a well-timed message rather than anything clever.
- Moving data between systems. If someone is copying information from one tool into another, that's an integration waiting to happen.
- Report assembly. Pulling numbers from four dashboards into one summary every Monday is a job nobody enjoys and software does perfectly.
What to Leave Alone
Some things should stay human, at least for now.
Anything where being wrong is expensive and hard to reverse. Pricing calls, contract terms, hiring decisions. Automate the paperwork around them. Don't automate the judgment inside them.
Anything that is the relationship. A customer with a problem wants a person. Automating the apology is how a small complaint becomes a public one.
Anything you don't already understand. Automating a broken process just gives you a broken process that runs faster. Fix it on paper first. If you can't explain it clearly to a new hire, you can't explain it to software either.
Build Small, Then Widen
The most common mistake we see is scope. A company decides to automate its entire sales pipeline, spends four months on it, and ships nothing.
Pick one step. Ship it. Let it run for two weeks. See what breaks, because something will. Then take the next step.
That isn't caution for its own sake. Every automation makes assumptions about how work actually flows, and those assumptions are always slightly wrong the first time. Small pieces let you find that out cheaply.
Measure the Thing You Meant to Fix
Before you build, write down the number you expect to move. Hours saved per week. Response time. Error rate. Cost per ticket.
Check it 30 days later. If the number hasn't moved, the automation isn't working, however impressive the workflow diagram looks.
We've switched off automations we built ourselves because the measurement said they weren't earning their keep. That's a good outcome. The bad one is running something for two years that nobody ever checks.
Where to Start This Week
Pick the task your team complains about that also happens daily. Track how long it takes for five days. Write the decision rules on a single page.
If the rules fit on that page, it's automatable. If they don't, you've learned something more useful: the process needs work before any software can help it.
