How do you find and eliminate toil systematically?

WHAT THIS TESTS This checks whether you can manage toil as a measurable engineering program, applying prioritization and proving impact, rather than reactively scripting whatever annoys you most.

A GOOD ANSWER COVERS Start by making toil visible and quantified. Capture it through lightweight time tracking, on-call and ticket analysis, and periodic toil surveys, recording how often each recurring task happens and how long it takes. Aggregate to find total toil load, ideally as a fraction of team time. Then prioritize: rank candidates by return on investment, roughly frequency times time-per-occurrence relative to the effort to automate, while also weighting tasks that carry outage or error risk. Tackle the highest-ROI items first, and prefer incremental automation, semi-automating before fully automating. Finally, define success metrics up front: reduction in toil hours per engineer, fewer on-call pages or manual interventions, faster resolution, and engineering time reclaimed for project work. Compare against the baseline you measured before starting.

COMMON WRONG ANSWERS Automating the most annoying thing regardless of payoff. Skipping measurement so you cannot prove improvement. Trying to automate everything at once. Counting scripts written rather than toil removed. Ignoring the risk that bad automation creates new toil or new failure modes.

LIKELY FOLLOW-UPS How do you account for automation that itself needs maintenance? What if the highest-toil task is hardest to automate? How do you sustain low toil over time? When is manual work acceptable?

ONE CONCRETE EXAMPLE A team surveys a month of on-call and finds certificate rotation consumes eight hours monthly across engineers and risks outages when missed. It ranks high on both cost and risk. They automate rotation with monitoring and alerts, measure the result, and the eight hours drop to near zero with no expiry-related incidents the following quarter. The before-and-after toil-hours number proves the win objectively rather than relying on a feeling that things improved.