On paying people
The unwritable part
You can write an SOP for how to run a Google Ads account. You cannot write one for whether somebody helps the person next to them. Here is how I try to pay for both — and the model I got wrong.
Two kinds of work, one paycheck
Every company runs on two kinds of work. Only one of them can be written down.
The first kind is technical execution. Build the campaign, fix the listing, publish the page, file the report on Friday. This kind of work should be written down, in detail, as an SOP — because a documented process is how a job stops depending on the person who happens to be holding it, and how somebody new gets to competence in weeks rather than years. We are aggressive about this. If a task is done twice, it should be a checklist by the third time.
The second kind has no SOP and never will. Somebody notices that a teammate is stuck and stops to unblock them. Somebody records the walkthrough once so the next five people never have to ask. Somebody says “this isn’t ready yet” to a colleague, kindly, when it would have been easier to let it ship. Somebody fixes the thing that is nobody’s job.
Here is the trap, and I have walked into it: the more precisely you specify and compensate the first kind, the more reliably you get only the first kind. People are not being cynical when this happens. They are reading the system correctly. If every hour is attached to a rate and every deliverable to a line item, then helping a colleague is, structurally, unpaid overtime — and a rational person does less of it over time.
You can mandate rules. You cannot mandate that somebody be helpful, or be a good person, or care whether the thing actually works for the customer.
So the design question is not “how do we write a better SOP for teamwork.” There is no such document. The question is how much structure you need before more structure starts destroying the thing you wanted.
Structure the technical, leave the human alone
My working split, after getting it wrong in both directions:
- Fully specified: technical execution and anything with a safety or money consequence. Access and credentials. What gets published and where. How a client is onboarded and deboarded. What has to be true before we spend somebody’s ad budget. The weekly report and its date. These get SOPs, checklists, and gates, and I do not want creativity in them.
- Guardrails only: how you communicate, how you handle a client who is upset, when to escalate. A few bright lines — never go silent, never guess on camera, never let a client hear bad news from someone else first — and then judgment.
- Deliberately unspecified: how you help each other. How much you teach. Whether you care. I will not write a policy for this, and I will not attach a bonus to it, because the moment I price it I have converted it into a transaction and killed the thing I wanted.
That third category is not me being soft on accountability. It is the opposite: it is an admission that this part cannot be enforced, so it has to be selected for instead — at the door, before somebody joins, when it is still cheap to say no.
Every hour spent trying to compel behaviour that only comes from character is an hour you could have spent finding somebody who already has it. Call it fit, culture, values, whatever you like. The practical version is simpler: some people ask “what am I supposed to do?” and some people ask “what needs doing?” You cannot convert the first into the second with a compensation plan. You can only pick.
I paid people ahead of what they had shown
For a stretch I offered new operators a very large share of the revenue on the clients they ran — the kind of share that belongs to somebody who owns the whole business: finds the clients, carries the risk, trains their own replacement.
I did it on purpose and for reasons I still think were decent. I wanted people to have a real reason to go all in. I would rather be generous early than lose somebody good over a percentage point. And I genuinely believed the money would pull the performance up behind it.
It does not. Paying at the top of the ladder for work at the middle of it means the top never has to be reached. The money arrives first, and after that there is no particular reason to climb. That is not a character flaw in the person who accepted it — anybody would. It is a design error, and it was mine.
There was a second half to my mistake, and it is the one I think about more. The implied deal was that the big share came with me personally available: training, unblocking, reviewing, usually one to one. If personal training and support were my only job, I could do that properly for maybe two or three people. Instead I did it in the gaps for many more than three. Everybody got a thinner version than they were promised, and I became the bottleneck on a hundred small things that only I was doing because only I had been asked.
That is backwards twice over. The most expensive person in the company should not be the default help desk — and an answer that only ever exists in a private message is worth nothing to the eleventh person who needs it.
Pay the role, not the person
Every dollar of a client retainer is doing one of five jobs. Write down what each job is worth and most compensation arguments stop being arguments.
Move the sliders. The point of this is not the exact percentages — argue with those — it is that a share should be traceable to work somebody actually did this month, and that the arithmetic has to leave the business standing.
What the work is worth
Pick the roles you actually perform, then see what that pays and what it leaves behind.
Assumes the cost of actually running a client — tools, ops, unbilled work, bad debt, overhead — is 30.4% of the retainer, and that roles you do not perform are paid to whoever does. Full derivation in Where the Retainer Goes.
Take every role yourself and pay nobody else, and the business breaks even at a 69.6% share. Above that it loses money on every client, every month, no matter how good the work is. That is not a negotiating position — it is subtraction. It is also exactly why the very generous deals I wrote could never have worked, however well anyone performed.
Four ways to pay, four sets of problems
Anyone who tells you one compensation model is simply correct is selling something, usually the one that suits them.
Here is the same work — the clients and retainer you set above — run through four models. The numbers move together. The trade-offs do not.
The same work, four ways
Adjust the assumptions each model needs. Everything else comes from the sliders above.
Equity is valued as your share of a sale at the chosen multiple on annualised profit, plus your share of profit distributed. It pays nothing this month, which is the entire point of the comparison.
What each one is actually good at
Revenue share aligns you with the client’s success and scales without a conversation. It also transfers real risk to the person least able to carry it: a client cancels and their income drops that month through no fault of theirs. And it quietly pays people for accounts they have stopped touching, which is how a book of business becomes an annuity for whoever got there first.
Hourly is the fairest model for work whose value is hard to see, and the only one that protects somebody on a difficult account from being punished for it. It also pays for presence rather than outcome, penalises the person who gets faster, and — this is the one nobody says out loud — becomes actively perverse the moment AI makes the same output take a fifth of the time. The better you get, the less you earn.
Per task is beautifully clear and settles most disputes before they start. It is also the model that most reliably produces exactly what was listed and nothing else. Every unlisted thing — the phone call that saved the account, the teammate you unblocked — becomes visibly unpaid, and behaviour follows within a month or two.
Equity is the only model that pays for building something rather than operating it, and the only one where the upside is uncapped. It is also worth nothing to somebody who needs to pay rent this month, it depends on an exit that may never happen, and offering it to somebody who wanted cash is not generosity — it is asking them to fund the company out of their own wages.
Revenue share for people running client accounts, because it tracks the thing the client is paying for. Hourly or per task for defined project work with a clear edge. Equity only for people building the business itself, and only when they want it and understand what they are taking instead of cash. The mix is not indecision — it is an admission that the models fail in different places, so you match the model to the failure you can least afford.
The gate, not the promise
A share should go up when something checkable happens — not when somebody asks at a good moment, and not because a year went by.
We describe careers on a nine-level ladder. Four of those levels touch a client retainer, and each one is a set of behaviours somebody can point at. Pick one to see what it pays and what has to be true first.
What each level pays, and what opens it
Not seniority. Not tenure. Demonstrated, and checkable by somebody other than me.
The reason to write the gate down is not bureaucracy. It is that the alternative — “show me you’re ready” — gets relitigated every month forever, and the person who is best at asking gets promoted over the person who is best at the work. A gate you can check yourself means nobody has to catch me in a good mood.
What we do not pay for, on purpose
Everything above is an attempt to price work precisely. This is the part where I admit the attempt only ever gets you most of the way.
Nobody here is paid to answer a question they already solved eighteen months ago. Nobody is paid to record the walkthrough once so that five people never have to ask. Nobody is paid to tell a colleague their work is not ready, or to fix the thing that is nobody’s job. It is not in anyone’s contract, it is not tracked, and there is no bonus attached to it.
I am not paid for it either, and I do more of it than anybody.
The reason we can ask at all is that the investment ran the other way first. Before anybody here calculated a revenue share, somebody handed them a method that was already written down, clients they did not have to find, a stack of tools they did not have to buy, a brand that makes the call get answered, and somebody senior who reads their work before a client sees it. The first client trusted them because they trusted the person who introduced them. That is a real balance and it was extended on credit.
Mentoring the next person is how it gets settled. Not because an agreement says so — because it is the only currency the debt is denominated in.
All of that is only honest if the investment is real, specific, and came first. “We invested in you” is a sentence that can justify almost anything, and plenty of companies use it to extract work they never earned the right to ask for.
The test is whether you can name the investment — this training, these clients, this tooling, these hours of somebody’s time reviewing your work — and whether the person receiving it would recognise the list. If you cannot produce that list, you do not have a mentorship culture. You have a story you tell yourself while underpaying people.
And it has a second edge: an unwritten expectation cannot be allowed to grow without limit. If “help each other” quietly becomes fifteen hours a week, it has stopped being culture and become an unpaid job. When that happens the answer is to hire, or to pay for it, or to stop asking — not to appeal harder to the mission.
Some things are meant for loss. The hour spent with somebody who then leaves. The training that ends up helping a competitor. The article that helps an agency owner I will never meet and never invoice. Carrying that on purpose, as a line you accept rather than a leak you plug, is most of what separates a company that compounds from one that merely operates.
What I am actually asking
- Specify the technical, and leave the human unspecified. SOPs and gates for execution, access, money and publishing. Guardrails for judgment. Nothing at all for whether you help each other — that gets selected for, not enforced.
- Pay the role, not the person. A share should trace to work done this month. Roles you stop performing, you stop earning.
- Never pay ahead of demonstrated work again. That was my error, it was expensive, and it was unfair to the people who received it — because it removed the reason to grow and then made the correction feel personal.
- Never reach backwards. Money already earned is earned. New terms start on the first of a month, with notice, and they apply to me before they apply to anybody else.
- Ask in the open. If the answer only exists in a message to me, it helped one person. In a thread it helps eleven, and it becomes a checklist that helps everybody after that.
None of these percentages are sacred. If you think execution is worth more than fifteen points, make the case — I would rather argue now than have somebody quietly conclude in six months that it is unfair. The card is a starting position that survives contact with the arithmetic. That is all a compensation model can honestly claim to be.
Get the structure right and you remove the excuses. Get the people right and you barely need the structure.