How to Run a Chatting Agency: The Day-to-Day Operation

Running a chatting agency is an operations job. The selling happens inside conversations, but almost nothing you do as an operator is selling: you set up accounts, you fill a rota, you read a sample of what your team wrote, and you pay people correctly and on time. Four functions, held at once, on a business that runs seven days a week and does not pause at night. Here is what each one involves, and where it breaks.

What does running a chatting agency actually involve?

Four functions, running in parallel, every day. Get one wrong and it shows up disguised as one of the others. A payout structure that pushes people to pitch early looks exactly like a quality problem when you read the transcripts.

• Accounts. Onboarding a creator, agreeing what gets sold and at what price, setting access, and keeping her rules current as she changes her mind.
• Shifts. Deciding which hours are staffed, by whom, and what happens at the handover.
• Quality. Reading a sample of the work, scoring it against named rules, correcting.
• Money. Tracking what each thread earned, paying chatters on the right base, paying creators on schedule.

Everything else (hiring, tooling, reporting) exists to serve those four.

How do you build the shift schedule?

Off your own sales-by-hour, never off equal blocks. The single most common mistake is a rota built like a call centre: three even shifts covering the clock. That spends the same money on hours that behave completely differently.

• Pull sales by hour per creator over several weeks. The peak does not land at the same hour for every account, and a rota built on the agency average is wrong for most of them.
• Overstaff the peak, thin everything else. Evenings maximise conversion in our data. That is where a second person pays for themselves.
• Treat the night as continuity, not revenue. The 2am-6am window collapses conversion. Push offers into it, at a night premium, and you pay more to sell less.
• Name the person who covers a no-show before it happens, not during.

Reading those hours properly (where the peak actually sits, and why it moves between accounts) is the subject of what time fans buy.

What has to exist in writing for each creator?

Four living documents per account, or two chatters will contradict each other inside a week. The contradiction almost always takes one of two forms: a promise one made and the other ignored, or content re-offered after a refusal.

Document	What it holds	Who updates it	Read when
Voice guide	How she writes, five real examples in her own words, hard nos	Creator, with the manager	Before the first message, then on change
Price grid	Tiers, what is bundled, what is never discounted	Manager, creator signs off	At every pitch
Handover log	Open promises, what was refused, where each live thread stands	Chatter leaving the shift	At the start of every shift
Escalation note	What goes straight to the creator: refunds, disputes, anything legal	Manager	On any exception

The voice guide is the one people fake. A guide written by the manager on the creator’s behalf produces chatters who all sound like the manager. See voice guide and handover for what belongs in each.

How do you keep quality up without reading everything?

By sampling deliberately and scoring against named rules, while accepting that you are reading a small sample of the work. That limit is structural. No agency reads its own inbox, and any process that pretends otherwise is theatre.

A workable loop, per chatter, per week:

1. Pull a small fixed set of threads: one that closed, one that did not, one that went quiet.
2. Score against rules that are known to move money, not against taste.
3. Return one correction, with the thread attached. One, not five.
4. Track whether the same correction comes back next week.

The rules worth scoring on are the ones our data actually measured across 38,879 sales: no pitch before the sixth message, with the optimum after about ten exchanges; never end a sales message with an ellipsis, the worst-performing closer measured; never close on a closed question, which costs several points of conversion. One more is worth naming because it feels wrong: a personal callback placed at the exact moment of the pitch costs points, even though every chatter believes it helps.

How should chatters be paid?

On tracked sales in threads they handled, never on activity. Paying per message sent, or per offer sent, manufactures the early pitch, and our data puts a pitch before the sixth message at roughly a third of conversion lost. You are paying people to destroy your own conversion rate.

The hard part is attribution across a handover, since the chatter who built the thread is often not the one who closed it. Two defensible options: split credit between the shifts that touched the thread before the sale, or credit the closer and pay a smaller flat rate on coverage hours. Pick one, write it down, do not change it mid-month.

Work out your own numbers rather than trusting anyone’s benchmark. Fill this in for one week, one creator:

Line	Your figure
Hours staffed this week	___
Base cost per staffed hour	___
Commission paid on tracked sales	___
Net revenue through the inbox	___
Inbox cost as a share of net revenue	___

Run it for four consecutive weeks before you conclude anything. See chatting team cost for the cost lines this worksheet leaves out.

Where does the operation break as it grows?

At three predictable points, and each has a different fix.

Breaking point	What actually broke	What fixes it
The second creator	Voice bleed: both accounts start sounding the same	Separate voice guides with examples written by each creator
The first unstaffed night	Threads go cold between shifts; nobody owns the gap	Written handover, and a named owner for the quiet hours
One manager, too many accounts	Nobody reads transcripts any more; drift is invisible	Hire the reader before the next chatter

There is a fourth, quieter one: the inbox outgrows the people. Past a certain volume the team stops choosing which threads to open and simply answers whatever is on top, which means the older half of the inbox is never touched again. That is a different problem with a different answer, covered in managing fan conversations.

What do you look at every week?

Five lines, and no dashboard beyond them until they are stable: net revenue per creator, sales by hour, cost per staffed hour, share of threads opened, and the spread between your best and worst chatter on the same account. That last one is the diagnostic. A chatter who is weak everywhere is a people problem; a chatter who is excellent on one creator and weak on another tells you the second creator’s voice guide was never really written by her.