Operate
A system is what remains when the person who knows leaves the room.
Founders are told to delegate. Delegation without a system is just distributed guessing — the work leaves your calendar and returns as escalation. What transfers is not the task; it is the judgment encoded around the task.
Operating systems are how a company holds quality without holding the founder's attention. Done well, they make most decisions unnecessary and the remainder fast.
How this goes wrong
Documenting steps instead of decisions
Most SOPs describe what to do when everything goes normally, which is precisely the case that never needed documenting. The value is in the exception path — what to do when reality diverges, who decides, and at what threshold it escalates.
Answer these honestly
- Which three tasks consume the most of your week, and which of them has a written decision rule?
- When your team escalates, are they missing information or missing authority?
- If you were unreachable for two weeks, what would break — and would you find out?
- Which recurring work is being done by a person because nobody has questioned it?
- Where does the same decision get re-litigated more than once a month?
The method
Map by frequency and judgment
High-frequency, low-judgment work gets automated. High-frequency, high-judgment work gets a decision rule. Low-frequency, high-judgment work stays with a human and gets a threshold.
Write the exception path first
Document what happens when it goes wrong, who decides, and the number above which it comes to you. The happy path mostly documents itself.
Apply AI where volume meets tolerance
AI earns its place on high-volume work where a small error rate is survivable and checkable — drafting, classification, summarisation, first-pass review. Not on irreversible decisions.
Assign an owner, not a committee
Every system needs one name against it who is accountable for it working and permitted to change it.
What installed looks like
- ✓Escalations arrive with a recommendation, not a question
- ✓Every recurring process has one named owner
- ✓Thresholds are numeric — a rupee value or a percentage, not "if it seems big"
- ✓A two-week absence is invisible in output
The three most common mistakes
Systematising before stabilising
Encoding a process that is still changing weekly produces documentation nobody trusts and everybody ignores.
Automating a broken process
Automation makes a bad process faster and harder to see. Fix, then automate.
Owning the system yourself
If the founder owns the SOP, the SOP is the founder. Nothing was delegated.
Pillars that move with this one
Knowing the pillar is not the same as knowing it is yours.
Ninety minutes to run the diagnostic on your own numbers and leave with the binding constraint named plus the pricing architecture and the 90-day cadence that follow from it.
Save my seat — free →