
The 12-month break-even rule for automation
I'm done with the obsession to automate everything. Most of the time it's just a sophisticated way to set your budget on fire - and the maths is easy enough to do before you start.
You know the situation... Someone spots a repetitive task and immediately reaches for Claude, Python or a workflow tool like Zapier to 'solve' the burden of manually doing the task.
We tell ourselves we're being efficient, but often we're just masochists who'd rather play with the shiny new tech than have an awkward and/or frank conversation with a stakeholder. I include myself in that - I love building. Building stuff is the fun bit of the job, and it's the bit nobody checks the numbers on.
The rule I live by is: If an automation doesn't pay for itself in time saved within a year, don't bother automating it.
It's not overly poetic or sexy - it doesn't make for much of a LinkedIn post either (trust me) - but it keeps the focus on the only question that matters, which is whether the work is ROI positive.
Why twelve months you may ask? It's short enough that you can't hide behind "it'll pay off eventually", and about as long as anything survives before the platform changes, the client changes or the person who understands it leaves.
Note: I'm well aware that sometimes you need to automate for reasons that have nothing to do with time i.e. accuracy where a human error would be expensive. Just be sure it's actually your reason, and not a story you're telling yourself after you've decided to build the thing anyway.
The time-saving matrix
For example, a five-minute task you do every single day costs you about 21.7 hours a year - three working days! So if that automation takes three days to build, you haven't made a saving. You've broken even, twelve months later, having also carried the risk that it broke in month four.
The chart below is the whole rule in one grid. Find how often you do the task, find how much time it takes you each time, and the cell is your entire (time) budget for building it.
What gets me every time is how brutal the low-frequency rows are. A monthly task that saves five minutes is worth one hour of build time. Not one hour a month - one hour total. Not many things can be built for that! Be honest with yourself, and accept that you may have to do it manually for now, and stop thinking about automating it.
Bear in mind that the build is the cheap part. The chart only counts build time, so be honest that you're also signing up for:
- Maintenance: Things need fixing when an API version changes or a column gets renamed.
- Silent breakage: An automated process fails silently and can quietly poison a dashboard for many weeks.
- Knowledge tax: Someone has to understand it. If that's only you, you've built a liability, not an asset.
My rule of thumb is whatever you think the build will cost, add a third for the year of looking after it. If the maths only just works, it probably doesn't.
Real-life scenarios in practice
A team spends half a day every week, stitching platform exports into a single dashboard and report.
Automation: worth it!
The data and analytics engineering pipeline took well under five weeks to implement, and cleared break-even in a few months. Not only was it mathematically worth it, the analyst I worked with was way happier, and that's priceless!
--
A monthly board report (Excel) where someone reformatted a few tables and pivots by hand. Twenty minutes a month, so four hours a year.
Automation: not worth it
The work alone would have exceeded that in an afternoon, let alone the fact that the layout changed every other quarter anyway from execs manually entering in the data! Still manual, and nobody has ever complained.
--
Alerting to catch data quality issues.
Automation: not worth it (it's a trap)
It definitely sounds essential, but the underlying task was a ten-minute weekly check - about 8.6 hours a year. Even a crude solution like a scheduled query and an email alert would take that long. And it's fixed and not able to automatically check new or other data points easily. It's annoying, but it works.
The AIlephent in the room
It's an obvious statement to make, but Gen AI has collapsed build times. Things that took me three days in the 'before times', genuinely take an hour now, which drags a load of tasks from the "not worth it" side to the "worth it" side. Nice.
What's changed is the failure mode. Building the wrong automation used to cost you say a week, so you (and your stakeholders, etc.) noticed. Now it costs an hour, so nobody tends to notice. Just be careful you don't end up with forty half-maintained automation scripts nobody else can explain. It could cost you or the business far more time in the future to unpick and understand what was a 'quick fix' at the time!
WTF is a 'billable hour'??
If you're an agency or freelancer, this may be uncomfortable to hear because it's an argument for doing less work.
The legacy model is built on selling hours. A three-day automation build is three days of revenue, even if it only saves one day per year of human time. Or the other way round where a task is kept manual to keep 'logging the time', and thus bill the client off the back of it.
Let's not kid ourselves, the clients have worked this out. They can see the difference between an agency that's busy and an agency that's useful, and increasingly they're buying outcomes rather than timesheets.
Agencies that charge for the hour rather than the value will find themselves very busy, very broke and very irrelevant. Sorry not sorry.
The honest version of this job is sometimes telling a client not to buy the thing they asked for. That's a much easier conversation when your commercial model doesn't punish you for it.
Automate for ROI, not for your ego, and definitely not for your timesheets.
