Why size is the wrong ordering
Because a ranked list by size is a ranked list by difficulty, and nobody notices until the first build is four months late. The correlation is not an accident. Big leaks tend to sit at structural boundaries — between two systems of record, between two departments, between a process and the workaround the team invented to survive it — and those are exactly the places where closing anything requires a decision nobody currently owns.
Smaller leaks tend to be local: one missing trigger, one unrouted channel, one step that depends on somebody remembering. They are cheap, fast, and they produce a number you can put in front of a board within a quarter. Starting there is not timidity. It is how you buy the political capital to attempt the structural one.
This is why a report that ranks by recoverable dollars against effort is worth more than one that ranks by magnitude, and why the order the report is written in should be the order the work happens in.
The four-question test for any candidate
A leak needs a yes on the first two to be a candidate at all.
- 01Can it be measured in a system you already own, before and after? If not, you cannot prove the build worked, and an improvement you cannot prove is one you will be asked to justify in a budget review you then lose.
- 02Is the logic writable? If you can describe the rule in a paragraph, it can be automated. If the honest answer is "it depends, you get a feel for it", this is a process to document before it is a process to automate.
- 03Does it recur often enough to matter? A monthly failure costing $2,000 is $24,000 a year against a build that might take three weeks. The arithmetic has to clear on its own.
- 04Is the process stable? Automating something you are about to redesign means building it twice, and the second build is more expensive than the first because it has to be unpicked.
The ordering, for most operating companies
First: response latency on inbound enquiries. First for nearly everyone and it is not close. The loss is silent, the mechanism is well understood, the fix is a routing and automation problem rather than a staffing one, and the signal is visible in the first week. It is also the leak most often misdiagnosed, which means fixing it corrects a belief as well as a number.
Second: follow-up behind a first response. Cheaper to build than intake routing and frequently comparable in value. Most companies reply once and stop. A conversation that stalls and is never picked back up closes itself, and the CRM records it as open indefinitely, which is why nobody sees it.
Third: routing and assignment. Enquiries that arrived, were answered, and reached the wrong person, territory or location. Invisible until somebody measures time-to-first-contact segmented by source, at which point one channel is obviously worse than the others and nobody knew.
Fourth: renewal and expansion triggers. Existing revenue lapsing because the prompt was a person remembering rather than a system firing. Usually the cheapest of all to close and almost always the last one anybody looks at, because it is not a new-business number and therefore has no natural owner.
Fifth: quote and proposal turnaround, where quoting is formulaic. High delay cost per instance and the logic is usually already written down in a pricing sheet, which satisfies question two immediately.
Sixth: handover between sales and delivery. Lower urgency, real retention impact, and it is the moment a customer decides whether the thing they bought is well run.
The three that feel urgent and are not
Internal reporting. Genuinely annoying, highly visible to leadership, and what it produces is hours saved — which get absorbed into the working day rather than reallocated, and therefore never reach a P&L. Worth doing. Not worth doing first.
Data entry between two systems that should not both exist. Automating the transfer entrenches the duplication and makes the eventual consolidation harder. The better move is usually to remove one system, and an automation project is frequently how a company avoids having that conversation.
Whatever the loudest person complains about most. Sometimes that is the right target. Sometimes it is a process affecting one person for twenty minutes a week. Run it through the four questions before it sets the roadmap by volume of complaint.
Sequencing rules that survive contact with reality
- Capture the baseline before touching anything. Current latency, conversion by latency, volume by channel. After go-live that data is gone, and without it you cannot prove what changed — which is the most common reason a successful build gets no credit.
- Pick something measurable within one sales cycle for the first build. A first project whose result is only visible in nine months will not survive a budget conversation.
- Do one intake path properly rather than four at once. What you learn doing the first usually changes the scope of the second, and doing them in parallel forfeits that.
- Write the failure path before the automation. Every automated step needs a defined answer for the case it cannot handle. Deciding that afterwards means deciding it during an incident.
- Name an owner for each closed leak, and attach a visible number to it. A leak that was closed and is not watched reopens, and a tool nobody owns leaks again within a quarter.
- Let the second build wait until the first has run unattended for a month. That month is where you find out what you actually built.
The failure this ordering exists to prevent
The common failure is not choosing the wrong leak. It is automating a genuinely bad process extremely well, and thereby making a fragmented operation faster at being fragmented. Automation amplifies whatever it is pointed at: point it at a coherent process and it compounds what works, point it at a broken one and it industrialises the breakage at a speed no human could sustain.
Which is why measurement comes before building, and why the honest version of this exercise sometimes concludes that a process should be deleted rather than automated. That conclusion is a good outcome and it is considerably cheaper than the alternative.