Most ops advice starts from a simple premise: if a task repeats, automate it. The logic feels airtight. Repetition is waste. Waste is cost. Cost should be eliminated.
We believed this for a while. Then we noticed something. The processes we automated fastest were not always the ones that produced the best outcomes. Some of our highest-value work happened in the seams — the moments where a human paused, read the room, and did something a script never would.
So we built a framework for deciding what not to automate. It changed how we operate.
Think about the last time a company surprised you with good service. It probably was not a perfectly timed drip email or a frictionless checkout. It was a person who noticed something, made a judgment call, and acted on it.
Those moments live in the seams between automated steps. Onboarding a new customer. Triaging an escalation. Coordinating with a partner on a tricky integration. These are repeatable tasks that follow patterns. But the value is not the pattern — it is the deviation from it.
When a new customer mentions during onboarding that they are migrating off a competitor mid-quarter, that changes the conversation. When an escalation comes in from a team that just closed a funding round and is about to triple their usage, the priority math shifts. A script handles neither of these. A person handles both, often in the same sentence.
We ask three questions before automating any process.
Does this task benefit from context that changes between instances? Sending a deployment notification does not. Walking a customer through their first week does. If yes, we keep a human in the loop.
Does the outcome affect a relationship? Generating an invoice does not build trust. Responding to an invoice dispute does. Wherever the outcome shapes how someone feels about working with us, we want a person making the call.
Will automating this make us blind to something we need to see? This is the most important question. Every manual process is also a sensor. When someone triages escalations by hand, they develop intuition about which customers are struggling, which integrations are fragile, which documentation is failing. Automate that away and you lose the signal along with the toil.
If a process passes all three filters — low context variance, no relationship impact, no signal value — automate it immediately. If it fails even one, pause.
The answer is not "keep everything manual." That does not scale either. The answer is to automate everything around the human touchpoint so the person doing the work can focus on judgment.
For onboarding, we automated account provisioning, environment setup, and welcome messages. The onboarding conversation itself stays human. The person running it arrives with full context already assembled, zero time on data entry, and nothing to do except pay attention.
For escalation triage, we automated collection, deduplication, and initial severity classification. A person still reads the escalation, decides who handles it, and writes the first reply. They spend their time understanding the problem instead of routing tickets.
For partner communication, we automated scheduling, status updates, and document sharing. The actual conversation — where priorities are negotiated, misunderstandings are caught early, and commitments are made — stays between people.
The pattern: automate the scaffold, protect the seam.
Three things happened when we adopted this framework.
First, the automated parts got faster. Removing ambiguity about what should be scripted made it easier to invest in those scripts properly. No one debated whether to automate provisioning. It clearly passed all three filters. So we built it well and moved on.
Second, the human touchpoints got better. When onboarding is not also data entry, the person running it has more attention for the customer. Quality went up because we stopped diluting the work that mattered with the work that did not.
Third, we kept learning. Manual processes continued to surface patterns we would have missed. A cluster of escalations from customers using a particular workflow told us something no dashboard would have. A recurring onboarding question led us to rewrite documentation before it became a support burden.
Automation ROI is not just time saved. It is time saved in the right places and time preserved in the right places. The highest return comes when you protect the seams that build relationships and compound knowledge — then automate everything else so those seams get full attention.
The default narrative says automate everything repeatable. The better rule: automate everything that does not get smarter when a person does it. The rest is where your advantage lives.
Be the first to comment.
0 comments
Loading comments...