You ran last quarter's numbers and the margin came in soft. Not catastrophic. Just lower than the quarter before, which was lower than the one before that. When you ask what happened, you get four different explanations from four different people. One points at a client that kept asking for "small tweaks." One points at the designer who left in June. One points at the retainer you knew was thin when you signed it. One just shrugs.

They're all right. And they're all a symptom of the same thing: nobody is actually running your margin. It's running itself.

Six decisions, every single day

In AgencyHabits podcast, episode 17, Peter and Sei-Wook walked through the six levers of client profitability: pricing, utilization, delivery design, scope, reusable IP, and AI. Six places where gross margin gets made or lost. Six decisions your agency is making every day.

In most agencies we've seen, two of them are being made on purpose. Maybe three in a good quarter.

The other four are happening by default. And default always has an owner. It's just not you.

When a lever doesn't have a real owner, it doesn't stop moving. It just gets moved by whoever is closest to it in the moment. Pricing becomes whatever the salesperson felt good about quoting on a Thursday afternoon. Utilization becomes whatever the PMs happened to staff when capacity felt tight. Delivery design becomes whatever the senior IC had time to draft on the first call. Scope becomes whatever the client pushed for after the third round of feedback. Reusable IP becomes whatever somebody thought to save in the right Notion page. AI becomes whatever individual people happened to try last week.

None of that is wrong in the moment. All of it is wrong in aggregate.

So here's the exercise: go through the six levers, one at a time, and figure out which ones you're actually deciding on purpose. For each, we'll name what the default looks like, what deciding on purpose looks like, who should own it, and what the conversation is supposed to sound like in the room.

Lever 1: Pricing and scoping

Default mode: the salesperson has a range in their head, reads the client's reaction, and lands somewhere in the middle. Scope is drafted to justify the number, not the other way around.

Decided on purpose: you price backwards from a target gross margin. Peter walks through the arithmetic on the episode: for a $100k retainer at a 70% target margin, your cost of delivery is locked at $30k. That's the question the team now has to answer: can we actually deliver the outcome for $30k of labor? If the answer is no, the price is wrong or the scope is. You find out before you sign, not in week six.

Who owns it: the founder or head of sales, in partnership with the delivery lead who has to staff it.

Cadence: every proposal over a threshold (we've seen $25k work well) gets a margin check before it goes out.

What the conversation sounds like: "What's the cost of delivery we're building this around? Can the team actually hit it, or are we pricing a project we can't deliver?"

Lever 2: Staffing and utilization

Default mode: PMs staff by who's free. Senior people get loaded up because they're trusted. Junior people sit 40% utilized because nobody's sure what to hand them. When a project wraps, the team stays on payroll and the next engagement carries the cost of their idle two weeks.

Decided on purpose: someone is looking at utilization every week, planned or actual, with full context on what's coming down the pipeline. Sei-Wook makes a sharp point in the episode: a project can look profitable on its own and still lose you money, because the two unbilled weeks after it wraps aren't in that number. That's a utilization problem, not a pricing one. You never catch it if you only ever look at project-level margin.

Who owns it: the head of delivery or ops lead, not individual PMs.

Cadence: weekly, in a 30-minute resourcing meeting where leads, PMs, and biz dev compare live work against pipeline.

What the conversation sounds like: "Who's going to be below 60% two weeks from now? What are we doing about it before it hits?"

Lever 3: Engagement design and approach

Default mode: the senior IC on the project sketches out a plan on the first internal call. It works if they've done this exact kind of engagement before. If they haven't, the team builds the plane while flying it. Everyone gets paid while that's happening.

Decided on purpose: the engagement approach is designed up front: start with the end date, work backwards, name the milestones, decide what's async versus sync before anyone's on a Zoom. Peter's pet peeve on the episode is worth calling out here: don't let perfect be the enemy of good on the project plan. A rough shape built in 90 minutes is more useful than a meticulous plan that shows up in week three, two months behind the launch date you already promised.

Who owns it: the delivery lead for that practice area, with the assigned PM.

Cadence: locked in the first week of the engagement, revisited at each major milestone.

What the conversation sounds like: "What's the shape of this project? Where are the budgeted feedback windows? What are we doing async that we used to do on a call?"

Lever 4: Scope management and change orders

Default mode: the designer on the weekly client call hears a new request and says "yeah, we can probably squeeze that in." Multiply by three roles and twelve weeks and you've delivered a phase-two project inside a phase-one budget.

Decided on purpose: every scope change goes through a named process, even if it's a zero-dollar change order. Especially if it's a zero-dollar change order. You trade scope for scope, you document what you traded, and you get it in writing. Sei-Wook's line on this is blunt: treat the zero-dollar trade like a new contract, signed. Because if you don't have the paper trail, three months later nobody remembers what was agreed, and the argument happens in an email thread you're going to lose.

Who owns it: the PM on the engagement, backed by a delivery lead.

Cadence: reviewed weekly during the project check-in.

What the conversation sounds like: "That's a new request. Do we swap it for something in-scope, defer it to a phase two, or write a change order? What's the client saying yes to?"

Lever 5: Reusable components and IP

Default mode: whoever remembers to save something, saves it. The Notion page exists. Nobody new has ever found it. Senior people rebuild the same slide template four times a year.

Decided on purpose: someone owns the library. There's time explicitly budgeted for packaging what a project just produced so the next team benefits from it. Unbillable, deliberately. Peter's point here is the one to sit with: specialization pays off exactly here, because the more similar your engagements are, the more leverage each reusable component gives you. A firm that does the same type of project twenty times has a completely different cost structure than a firm that does twenty different projects once.

Who owns it: a practice lead or ops role. Not "the team," which in practice means no one.

Cadence: 30 minutes at the close of every engagement, protected on the calendar.

What the conversation sounds like: "What came out of this project that saves the next one time? Where does it live? Who's responsible for it being findable?"

Lever 6: AI and automation

Default mode: individual ICs experiment on their own. One developer is vibe-coding scaffolds. One designer is using AI for moodboards. One account manager has a personal GPT that drafts status emails. Nobody is measuring what it's saving, and none of it is making its way into how the agency quotes or staffs.

Decided on purpose: someone is looking at the delivery workflow end-to-end (contract signed, project kicked off, QA, invoice sent) and asking where the administrative drag is. They're also watching how the execution itself is getting faster. The savings are real, but they only show up in margin if they show up in your pricing or your utilization targets. Otherwise they just get absorbed by extra polish, extra revisions, or quietly expanded scope.

Who owns it: the ops lead, or a designated person with a real mandate. Not "everyone."

Cadence: monthly review of what's been tried, what's working, what's going into standard process.

What the conversation sounds like: "What are we doing now that three months from now an automation should be doing? What did we learn last month that should change how we scope the next one?"

What to do Monday

None of this is a new framework. You probably already own one or two of these: the ones you care about most, or the ones that burned you badly once. Go look at the other four.

Here's the exercise for Monday morning. List the six levers on a piece of paper. Next to each, write the name of the person who owns it. If you can't write a name, or if the name is "the team," or if the name is you but you know you haven't looked at it in a quarter, that lever doesn't have a real owner.

You'll probably find three or four without one. Don't try to fix all of them. Pick the two that are leaking the most right now and assign them by Friday. For this quarter, not forever. Put the cadence on someone's calendar. Write down what the conversation is supposed to sound like in the room.

You're not going to fix all six at once. You probably shouldn't. But you can change which ones are defaulting and which ones are decided, and in our experience that's the shift, not better process, not better people, that quietly pulls your margin back up.