Before you automate it, ask whether it should exist
Name one process you suspect should not be there. Answer ten evidence questions about it. Six gates run in order and the first one that catches it decides, which is why a tidy report nobody reads still loses and a messy process with a named consumer still survives.
Free, no signup, no AI, and nothing you type leaves your browser.
Think of one process you suspect should not exist.
The report nobody reads. The approval nobody changes. The form everyone works around. Name it, answer ten questions about it, and watch which rung of the ladder catches it. Nothing you type leaves this browser.
Does anything, or anyone, actually use what this produces?
Opens, replies, a decision it triggered, an API call, a downstream job. Ninety days is the window. A distribution list is not a reader.
indirect or partial consumption
Keyboard: space re-runs the ladder, A runs the worked example, B switches to the portfolio board, R clears the floor.
Pain nominates. Pain cannot defend.
An automation request list is a map of where the organisation hurts, not a map of where automation belongs. Ask a team how AI could improve a process and nobody in the room is allowed to give the honest answer, which is sometimes that it should not exist. So the backlog fills with rubber-stamp approvals to accelerate, reports nobody reads to summarise, and workarounds to automate around.
Evidence decides existence, never a score.
The tool keeps a health score and shows it to you. No rung of the ladder reads it. That is the point: a healthy process can still be merged, and a dying one cannot be rescued by points. The score describes the shape of the work. The gates decide whether the work should be there at all.
The six gates, in the order they run
The ladder is deterministic. Given the same ten answers it returns the same verdict every time, for every process, in every department. Nothing below reads the health score and nothing below is weighted.
- 01
Ownership
UNOWNEDQ10 = unowned
A process nobody can evidence in ten minutes is not scored, it is put on a 14-day notice. An owner who surfaces inside the window re-enters scoring like any other candidate. The register is a holding pen, never a graveyard.
- 02
Existence test
ELIMINATEQ1 = no consumption AND Q2 = no anchor AND Q3 = no decision changes
All three limbs absent is the only route to ELIMINATE, and no health score can rescue it. Any one limb surviving sends it on down the ladder. No regulatory anchor is required to qualify as useful.
- 03
Duplication
MERGEQ6 = duplicate exists
MERGE applies whatever the health score. Two healthy processes producing one output are still one process wearing two costumes.
- 04
Path reality
REDESIGNQ5 = exceptions above 40%
Automating fiction builds a bot for the process nobody runs. Redesign on the path that works, then re-score.
- 05
Automation applicability
AUTOMATION CANDIDATEQ1 = evidenced AND Q3 ≠ none AND Q5 = exceptions ≤ 10% AND Q9 ≠ under 2 hrs AND Q10 = evidenced
This is the only door to the Automation Applicability Gate, and the verdict is CANDIDATE, never AUTOMATE. The gate runs after triage, on evidence, and can still route the process back to manual by design.
- 06
Default
SIMPLIFYeverything that reaches here
SIMPLIFY is the floor of the ladder, not a consolation prize: strip the unanchored fields, cut the unread covering narrative, keep what a consumer actually consumes.
The ten questions, and the evidence each one wants
Every question has an answer you can point at. That is what separates triage from a workshop where the loudest person wins. Take these into the room and ask them about one process before you write a single line of automation.
Does anything, or anyone, actually use what this produces?
Opens, replies, a decision it triggered, an API call, a downstream job. Ninety days is the window. A distribution list is not a reader.
Is there a rule that says this must exist?
A regulation, a standard, a contract clause or an audit requirement you can point at. An internal policy is not an anchor unless it cites one.
If it stopped tomorrow, would any decision or outcome change?
Name the decision and the person who makes it. “People like seeing it” is not a decision.
Do the approvals inside it ever change the answer?
Count how many instances were approved unchanged. Past 95% you are looking at a rubber stamp with a latency cost.
Does the work actually follow the documented route?
Count the exceptions over a month. Past 40% the documented path is fiction and the real one is somewhere else.
Does the same output already exist somewhere else?
Search for the number or the table it contains, not its filename. Two reports with one set of figures are one process in two costumes.
When did anyone last ask whether this should exist?
A dated answer on the record. “We have always done it” is a no.
Do you know why it started, and is that reason still true?
The person or the event that created it. A reason that has expired counts the same as a reason nobody knows.
How much time does it take each week, across everyone who touches it?
Hours per week for all of them, not just the owner. This sets priority and whether automation could ever pay back.
Could somebody show you all of the above in ten minutes?
If nobody can, it is not scored at all: it goes on a 14-day notice while an owner is found.
What this tool is not
- Not process discovery. Nothing here mines your logs or finds processes you have not thought of. That is process mining, a heavier tool for a different job. This assumes a shortlist your team can name from memory.
- Not a decision. A verdict here is a first pass on ten answers you gave yourself, in private, with nobody checking them. The version a sponsor signs needs the evidence attached, the contested verdicts settled in a room, and a memo somebody puts their name on.
- Never a safety authorisation. Standing exclusion: safety verification steps (permits, isolations, confined-space checks, inspection sign-offs) never enter triage and never reach this gate. They are outside this ladder by design, not by oversight: AI prepares, people decide, and nothing on this page has an opinion about a safety sign-off.
- Not a graveyard. A process nobody can evidence in ten minutes is not killed, it goes on a 14-day notice. An owner who surfaces inside the window re-enters scoring like any other candidate.
The 24-process portfolio inside the tool is Halvard Marine, the kit’s invented worked example. It is labelled as invented wherever it appears. For verdicts real organisations reached, with the evidence and the hours, read the Kill Log.
The Kill This Process Kit
This tool runs one process for one person. The kit runs a backlog for an organisation: an eight-tab triage scorecard that scores asynchronously, a 90-minute session your own sponsor facilitates to settle only the contested verdicts, a register, the kill-or-keep memo, and the automation applicability gate that routes every survivor to deterministic workflow, AI-assisted, agent-assisted, or manual by design.
Two calendar weeks of edge-of-day effort. Self-serve, one organisation, unlimited seats inside it, nobody attends anything from our side.
See the kit, $249Not ready for the kit? Should This Process Exist? The 5-Minute Test is free, and it is the one page you can take into a meeting on Thursday.
One useful thing every other Tuesday
AI at Work: what actually changed, what it costs, and what to do about it. Checked against the vendor’s own pages before it ships.
Where to next
The Kill This Process Kit
The version a sponsor signs. Scorecard, workshop guide, register, memo and the applicability gate. $249, one organisation.
Go→The Kill Log
Processes organisations discovered they did not need: the verdict, the evidence, the hours released, and whether a build was avoided.
Go→More free tools
The Copilot cost calculator, the allocation method checklist, and the five rounding checks where the decimals decide.
Go→Kesslernity. Independent practitioner, no vendor sponsorship. The ladder here is the one the shipped workbook runs, and a build-time check recomputes the workbook’s own verdict distribution over its 24 worked rows on every deploy. Halvard Marine is invented and labelled as invented. Output is a first pass for planning, not a decision, and not legal, audit or employment advice.