Running an Agency With 20 Creators: Delegate and Steer

Twenty creators is not four times five. At five, the agency fits in your head and what you do all day is the work itself. At twenty, the work is done by people you are not watching, in inboxes you will never fully read, during hours you are not awake for. You stop reading conversations and start reading numbers about conversations. Everything below exists to make those numbers worth trusting.

What actually changes between five and twenty creators?

You lose direct evidence. At five creators, every problem is something you notice: a fan who sounds confused, a chatter whose messages read differently this week. At twenty, none of those reach you, and the failure signals that worked at five creators stop firing.

What breaks first, in order:

• Reading. You can no longer sample the work yourself, so quality becomes something you are told about rather than something you have seen.
• Memory. Rules that lived in your head (what this creator refuses, what that one discounts) now have to live in documents someone else maintains.
• Averages. With twenty accounts, the agency total goes flat while one account halves and another doubles. The number stops describing anything.
• Escalation. At five, everything reaches you. At twenty, only what someone chooses to escalate reaches you, which means you are steering on a filtered feed.

How do you split twenty creators into portfolios?

Into portfolios balanced on peak hour, with revenue spread across them rather than concentrated. The split you choose determines what your managers can see, so it is worth one deliberate decision rather than a historical accident.

How you split	What it groups	What it makes easy	What it hides
By revenue rank	Big accounts together, small together	Rewarding your best manager	Small accounts get no attention and never grow
By peak hour	Accounts whose fans buy at the same time	Rostering, because one shift covers the whole portfolio	Nothing structural: this is the default to prefer
By content type or voice	Similar tone, similar offers	Chatters switching between accounts	Fragile: one bad practice spreads across the whole group
By who signed them	The manager who won the account keeps it	Creator relationships	Wildly uneven portfolios, and no comparison between managers

Peak hour wins because it is the only split that makes the roster cheaper to staff. A portfolio whose accounts peak in the same block can be covered by one shift instead of three, and our data is unambiguous about where that block sits: evenings maximise conversion, and the 2am-6am window collapses it. Balance revenue across the portfolios afterwards, so that comparing two managers means something.

Who reviews whom, and how often?

Review crosses portfolios, always. A manager who scores their own accounts is scoring their own decisions, and the scores drift upward long before the quality does.

A workable chain, weekly:

1. The manager reads their own portfolio daily. Unscored, informal, for correction in the moment. This is coaching, not assessment.
2. A manager from another portfolio scores a fixed sample. Three threads per chatter: one that went quiet, one that did not close, one that did. Same three every week, so the comparison is stable.
3. You read the reviewers, not the chatters. Three scored threads a week, chosen at random from what was already reviewed. You are checking whether the scoring is honest, not whether the writing is good.
4. Each creator gets a sample, not the inbox. A handful of threads on a schedule tells her more about the team than the whole inbox, which she will open once.

Score on rules that have been shown to move money, never on taste. That distinction is what quality control means in practice. Publish the grid before anyone’s first shift. A score nobody can see is not correction, and chatters write defensively around it.

What do you keep, and what do you genuinely delegate?

Keep three things, delegate the rest, and mean it. The most common way a twenty-creator agency stalls is an owner who has delegated the work but not the decisions, so every portfolio queues behind one person’s inbox.

Decision	Who holds it	Why
Price floor and what is never discounted	You	Reversing a price collapse across twenty accounts takes months
Contracts and the commission base	You	The base a percentage applies to recurs every month, in the same direction
Escalation list: refunds, disputes, anything legal	You	These carry consequences outside the agency
Rota and coverage	Portfolio manager	They know their own peak hours
Hiring shortlist and testing	Portfolio manager	The person who lives with the hire should choose it
Voice guides and price grids	Manager, written by the creator	A guide written on her behalf produces chatters who sound like the manager
Weekly quality sample	Reviewer from another portfolio	Independence is the whole point

Note what is not on the keep list: chatter selection. Owners hold onto hiring longer than anything else and it is rarely the right call: the protocol matters more than the picker, which is the argument in how to hire chatters.

Which numbers do you steer on at twenty creators?

Spreads, not totals. A flat agency total is the least informative number you own, because it is the sum of movements pointing in opposite directions.

Six lines, read weekly:

• Each creator against her own trailing baseline, never against the roster. Accounts are not comparable to each other; each is comparable to herself a month ago.
• The spread between your best and worst chatter on the same account. Same voice, same price grid, same fans: the gap is a people signal, cleanly isolated.
• The spread between portfolios on the same metric. If one portfolio is consistently better, find out what its manager does differently before you promote them.
• Coverage gap: staffed hours against sales-by-hour. The question is not how many hours you staff, it is whether they are the right ones.
• Share of threads opened. The hours nobody covers appear in no cost line, which is why nobody counts them. That arithmetic is set out in what a chatting team costs.
• Net revenue per creator, on a base written down in words. Gross and net revenue are different numbers, and at twenty accounts the difference is no longer something anyone reconciles by eye.

How do you spot a drop before it costs you?

By watching the conversation, not the revenue line, because a badly handled fan stops replying weeks before he stops appearing in the numbers. Five signals move first, and all five are countable from transcripts.

1. Where the pitch lands. Sales attempts drifting earlier in threads is the clearest early warning there is. Our data, across 1.5M messages, puts a pitch before the sixth message at roughly a third of conversion lost against one placed after about ten exchanges. It drifts when chatters are overloaded or paid on activity.
2. Reply latency inside the peak block. Slow replies at the hour that converts best cost more than slow replies anywhere else.
3. Ellipsis closers reappearing. The worst-performing closer measured, and the easiest habit to slide back into.
4. Closed questions at the moment of closing. Costs several points of conversion.
5. A personal callback placed at the exact moment of the pitch. It costs points even though it feels like it should help, the most counter-intuitive result in the corpus.

Count each one as a rate per chatter per week. A signal you cannot count is a signal you will argue about.

What does structure not fix?

Two things, and both survive any reorganisation. The first is the hours nobody covers: threads never opened, subscribers nobody has spoken to, lapsed fans nobody has hours for. Splitting twenty accounts into portfolios does not create hours; it only allocates the ones you already pay for.

The second is what a chatter writes at the moment of the pitch. Reviewing, scoring and portfolio structure raise the floor and make drift visible; none of them puts the right sentence in the message. That is a training and tooling question, not an org chart question, and confusing the two is how an agency ends up with three layers of management and the conversion rate it had at five creators.

Where do you start if you restructure one thing this quarter?

Move the scored sample out of the portfolio that produced it. Same people, same grid, different reader. It is the change that makes every other number trustworthy. Until review crosses portfolios, your quality data is a self-assessment, and steering on numbers you cannot trust is worse than steering on feel, because it feels like evidence.