Field note 01 · Operating efficiency
Most small businesses respond to a growth problem by buying more demand. In two and a half years of consulting I almost never found demand to be the binding constraint. The constraint was the delivery system — and nobody could see it, because the waste was distributed one email, one approval, one re-do at a time.
None of these are intuitive. All four are why the obvious move is usually the wrong one.
If 30% of your delivery is rework, marketing spend buys you 30% more rework — at full price, with a customer attached. Demand poured into a leaking system converts into churn and reputation damage, which are far more expensive than the ad spend that caused them.
The intuitive use is generating content. The high-return use is triage, extraction, summarisation and drafting — the coordination layer nobody bills for and everybody pays for. That layer is where a small team quietly loses its week.
An ad impression has a half-life measured in hours. A fixed defect, a written process, a senior hire — these keep paying out for years and reduce the cost of everything downstream. Expense versus asset is the whole decision.
Communication paths scale as n(n−1)/2. Going from 6 to 12 people quadruples the paths. This is exactly the size at which most founders hire a marketer instead of fixing how decisions get made — and then wonder why the team feels slower at twice the size.
Each one is a move I watched intelligent operators make, the mechanism that makes it backfire, and what works instead. Expand any of them.
We're at capacity, so add a person. Capacity goes up by one person's worth.
Output falls for one to two quarters. The new hire produces at a fraction of full rate while consuming a meaningful share of your best person's time, and every existing communication path multiplies.
Mechanism: onboarding is a transfer from your highest-value capacity to your lowest. If the process isn't written down, the transfer happens verbally — so the cost lands on whoever holds the knowledge, who is invariably the person you could least afford to slow down. Do first: document the role before you fill it. The document is what makes the hire cheap, and writing it frequently reveals the role wasn't needed.
Find whoever is most overloaded and help them. That's clearly where things are stuck.
You optimise a step that wasn't constraining anything and the lead time barely moves. The real bottleneck is a queue — work sitting still, waiting on a decision, invisible because nobody is busy with it.
Mechanism: busyness is a measure of utilisation; bottlenecks are a property of flow. By Little's Law, lead time = work-in-progress ÷ throughput, so the fastest way to shorten delivery is usually to reduce how much is in flight at once — not to speed anyone up. Do first: measure where work waits, not where people work. Waiting is nearly always the majority of elapsed time.
Push utilisation toward 100%. Idle time is waste, and we're paying for the hours either way.
Past roughly 85%, lead times don't rise gently — they rise sharply and then explode. Every small disruption now propagates, because there is no slack anywhere to absorb it.
Mechanism: queueing theory. Waiting time scales with ρ/(1−ρ) — at 80% utilisation the ratio is 4; at 95% it is 19. The last 15% of utilisation you squeeze out costs you multiples in delivery time and blows up variability, which customers experience as unreliability. Do first: treat 80–85% as the target, and protect the slack deliberately. Slack is not idleness; it is the shock absorber that keeps promises credible.
This task eats four hours a week. Automate it — or point AI at it — and reclaim four hours.
You've now spent money making an unnecessary step fast, cheap, and permanent. Automation confers legitimacy: nobody questions a step that runs itself, so the step outlives every reason for its existence.
Mechanism: the cheapest efficiency gain is deletion, and deletion has to come first — the order is eliminate, simplify, standardise, then automate. Skipping to the last step is what produces an expensive tool nobody can explain. Do first: ask who consumes the output and what decision it changes. A startling share of recurring reports have no reader and change no decision.
People aren't aligned, so book a recurring sync. Everyone hears the same thing at the same time.
A weekly hour with seven people is not one hour — it is seven hours a week, roughly 320 a year, and it fragments two working days for everyone attending. Misalignment usually persists, because the actual failure was that no decision was ever written down.
Mechanism: meetings are a synchronous fix for what is nearly always an artefact problem. If a decision isn't recorded with an owner and a date, it will be re-litigated — and a recurring meeting is simply a scheduled place to re-litigate it. Do first: require every recurring meeting to output a written decision with a named owner. Meetings that can't produce one get deleted, and the calendar audits itself.
Discount 10% to win more volume. We'll make it back on scale.
On a 40% gross margin, a 10% discount removes a quarter of your contribution per unit — so you need roughly a third more volume just to stand still, delivered by the same team that was already near its ceiling. Volume you can't deliver well converts to churn.
Mechanism: price cuts hit contribution margin directly, while efficiency gains raise it. The two are the same lever pulled in opposite directions, but only one of them also increases your capacity to serve. Do first: compute the volume required to break even on the discount, and ask honestly whether your delivery system could absorb it. Usually the answer settles the question.
Arguments are cheap. Drop in real material — inbox exports, meeting notes, an ops log, a project tracker — and this page reads it for the signatures of drag, credits what you're already doing well, and prices the gap.
Nothing leaves this page — and it can't. Your files are read into your browser's memory and analysed by JavaScript running on this page. There is no server, no upload, no account, no cookie, and no analytics. Nothing is stored: close the tab and the contents are gone.
This is enforced, not just promised. The page ships a Content Security Policy that blocks all outbound connections and all external scripts, so no code here — mine or anyone's — is able to transmit your data. You can verify it: save the file, disconnect from the internet, and it still works. Because no personal data ever reaches me or any third party, nothing you drop in here creates a processor relationship under GDPR — you remain the sole controller of your own material throughout.
One caution. The audit quotes real lines from your files on screen, including names. If you export this page to PDF or screenshot it, those quotes travel with it — so treat the output as confidential to the same degree as the input.
Plain-text formats: .txt .md .csv .tsv .log .eml .json .vtt .srt — email exports, meeting notes, Slack exports, task lists, transcripts. Multiple files welcome; larger samples give better estimates.
This tool deliberately cannot open .docx or .pdf: parsing them needs an external code library, and loading one would break the guarantee that nothing on this page can reach the network. Selecting all and pasting is the honest workaround. Meeting notes work best when each block keeps its Attendees: and Duration: lines — that is what lets the audit price meetings from real numbers rather than estimates.
Every number below traces to an input you can change. The coefficients are estimates, and they are all visible — open the assumptions panel and argue with them. That is the point.
Why coverage matters. A sample of your inbox shows only a sample of the waste. Detected hours are divided by the period, then grossed up by this figure to estimate the whole. It ships at 100% deliberately — that reports only what is literally present in the files you provided, which is the defensible floor and the number to quote to anyone who will push back. Lower it only if you know these files are a partial view, and say so when you cite the result. Every percentage point below 100 makes the headline figure larger and harder to defend.
Meeting hours are not coefficient-based — they are computed from stated attendee counts and durations. Tool sprawl is computed as a context-switching share of total team hours, not per instance.
What this is and isn't. Detection is keyword- and structure-based. It finds the linguistic signatures of drag — a person asking for the fourth time, a version numbered v4, a meeting with seven attendees and no recorded decision. A negation guard drops lines that carry the keywords but mean the opposite ("nobody has to do that manually now"), and meeting blocks are bounded so a later decision cannot be credited to an earlier meeting that recorded nothing. It still cannot know whether a given instance was justified, and it will miss waste that nobody wrote down. Treat the output as a directional map of where to look and a defensible order of operations, not a diagnosis. Every serious engagement starts here and then goes and watches the actual work.
Same budget, three strategies, 24 months. The model gives all three the same customers, the same margin, and the same team — the only difference is where the money goes and what that does to the ceiling on what you can deliver.
A line chart comparing marketing-first, efficiency-first, and sequenced strategies. Full values are available in the table view below.
Load files or the worked example to run the comparison.
A thesis that always wins isn't a thesis, it's a sales pitch. The argument above is about sequencing, not about marketing being bad. Three cases where the order flips.
If utilisation is comfortably under the ceiling and delivery quality holds, you have paid-for capacity sitting idle. Filling it is the highest-return thing available and efficiency work can wait — the model reflects this, and will say so.
Low churn is evidence the delivery system already works. Efficiency work then competes against a compounding acquisition machine, and usually loses. Fix-first is for businesses whose churn or rework is telling them something.
Seasonality, a competitor's exit, a regulatory or platform shift. Timing value is real and process work is rarely urgent. Take the window — but book the debt honestly, because you will pay it during the growth.
Where this goes next
A keyword pass over your inbox is a starting map, not an answer. The engagement is watching the actual work, ranking fixes by effort against recovered capacity, and deleting before automating — in that order.