A reader asked the sharpest question I have seen this month.
What kind of problems fall into the 20 percent that still needs a senior.
Most automation pitches answer the opposite. They sell the 80 percent that was always going to work, and they stay quiet about the part that breaks careers.
So here is the honest answer, from the enterprise stacks I audit across DACH automation deployments.
The 80 percent is the easy promise. The 20 percent is the whole engagement.
What the 80 percent actually is
The 80 percent is the work that has a shape.
A clear input, a known rule, a predictable output. Invoice in, line items out. Lead in, enrichment out. Ticket in, routed ticket out.
This work should be automated without apology. It is repetitive, it is well understood, and a human doing it by hand is waste.
If a vendor is still selling you the automation of the 80 percent as the hard part, they are selling you the easy thing at the price of the hard one.
What the 20 percent contains
The 20 percent is not a smaller version of the 80. It is a different kind of problem. In every audit it breaks into four classes.
- Judgment under ambiguity. There is no rule, because the situation is new. Someone has to weigh trade-offs that the system was never told about.
- Exceptions with real consequences. The edge case that touches money, a regulator, or a customer relationship that took years to build.
- Context that lives in a head, not a database. The reason this client gets the exception, the history nobody wrote down, the thing everyone on the team knows and no field captures.
- Decisions that need a name on them. When it goes wrong, accountability has to land on a person, not a workflow.
Every one of these reports as a normal task to an automation. None of them is a normal task.
Why teams automate the 20 anyway
The pressure is real and I understand it.
The 80 is automated, the savings are visible, and the natural next ask is to push for the rest. The demo of the 20 always looks fine, because a demo runs the happy path, and the happy path is the part of the 20 that behaves like the 80.
So the team ships it. Then a real exception arrives, the one with money on it, and the automation handles it with the same confidence it handled an invoice.
That is the moment a senior should have been holding the decision, and was not.
What that failure costs
When the 80 percent fails, you lose developer hours.
When the 20 percent fails, you lose the thing that does not come back on a retry.
A regulated decision made without review. A key account handled like a routine ticket. A judgment call the system was never qualified to make, made anyway, at scale, before anyone noticed.
The cost is not in the error count. It is in which errors they are.
The honest posture
Automate the 80 percent ruthlessly. Free your seniors from it completely.
Then do the harder work. Find the boundary where the 20 percent begins, and build the stack so it calls a human at exactly that boundary, with the context already gathered and the decision framed.
The goal is not to automate the senior away. It is to make sure the senior is spent only on the 20 percent that is actually theirs, and is never bypassed on it.
A team that gets this right looks slower on the org chart and is far safer in production. A team that automates the 20 to look fully autonomous is one exception away from the incident that ends the project.
What this writeup does not hand you
I run the boundary version of this in client engagements. Where exactly the 20 percent starts in a given stack, how a workflow is forced to recognise it has crossed into judgment territory, and the escalation that puts the right senior on the right decision are what I bring to an audit.
I am not pasting them here, and the reason is honest. The boundary is specific to each company's risk, and a copied recipe gives a team false confidence that it has drawn the line correctly when it has not.
Drawing that line wrong is the failure. The discipline is in drawing it right.
One question for you
Look at your own automation.
Find the place where it stops asking a human and starts deciding alone, and ask whether that line was drawn on purpose or by accident.
If it was drawn by accident, tell me in the comments what kind of decision sits on the wrong side of it. I will reply with the question I would ask first to find out whether it belongs in your 80 or your 20.
Top comments (0)