A client gives thirty days' notice. In the first week of the wind-down the account lead discovers that two of the client's LinkedIn senders were connected through an operator's personal login, the opt-out list lives inside a campaign tool the client has never seen, and eleven open conversations have no named owner on the client's side. The gap was easy to miss during the engagement. The stack never recorded who owned what, and the exit is where that shows.
That scenario is invented, like the agency used throughout this guide, but it explains why an agency stack should be designed around boundaries before features. An agency that can answer four questions for every client (what do we know that belongs to them, who controls each sender, what counts as qualified for them, and what leaves when they do) can run on fairly ordinary software.
What the Agency Knows and What the Client Owns
An agency learns two kinds of things. The first is craft: how to structure a handoff note, which review step catches bad lists, how long to wait before following up a referral. That travels between clients and is what the agency is paid for. The second is client knowledge: pricing, approved claims, named customers, exclusions, what buyers said in replies. That belongs to one client and should be stored where it cannot reach another client's work by accident.
The test is whether an operator working on Client B can see, search or paste from Client A's material without the permissions and deliberate context switch appropriate to that client. If a shared snippet folder, prompt library or enrichment table spans clients, the boundary exists only in people's heads. Separate workspaces cost more in seats and setup, and they earn it once clients sell into overlapping markets.
Sender Ownership
Every sending identity needs a recorded owner, and my default is the client. Domains and mailboxes sit in the client's accounts; LinkedIn profiles belong to the client's people, because a profile is a person and leaves with that person. The agency works through delegated access and keeps a short register for each sender: who owns it, who may reconnect it, who may pause it, and who removes agency access at the end. A pilot may use agency-owned infrastructure, but agree on access, history export and the exit process before it starts.
Judge tooling against the register. HeyReach presents multiple-sender LinkedIn outreach for agencies and teams with a shared inbox, and lemlist markets campaigns across email, LinkedIn, calls and messaging with a unified inbox. Neither description tells you whether a client can see only their own senders and conversations, or what happens to a connected account when an operator's seat is removed. Ask to be shown both.
Four Clients, Counted in Operator Minutes
Take a hypothetical agency with four clients and two campaign operators. The weekly work for each client is set out below, using planning figures chosen for illustration.
| Task | Weekly Minutes for One Client | What Would Reduce It |
|---|---|---|
| Reply triage and assignment | 125 (25 each weekday) | One inbox that keeps clients separate |
| List review before launch | 90 | Fewer rejects from better sourcing rules |
| Chasing and recording approvals | 30 | A standing approval record |
| Writing handoff notes | 60 (six notes at ten minutes) | Context captured during the conversation |
| Weekly client report | 45 | Reporting from the records operators already use |
That is 350 minutes per client, or 5 hours 50 minutes. Across four clients it is 1,400 minutes: 23 hours 20 minutes a week, or 11 hours 40 minutes for each operator. Reply triage alone is 500 of those minutes, a little over a third.
Two judgments follow. A tool that claims to simplify reply handling should be tested against those 500 minutes. A shared inbox that leaves an operator working out which client a thread belongs to, then copying the outcome elsewhere, has moved the work without removing it. And at the same workload, a fifth client adds another 5 hours 50 minutes a week. At an illustrative loaded rate of $45 an hour, that is roughly 25.3 hours and $1,138 of operator time per average month (52 weeks divided by 12) for each client, before seats, data, infrastructure and strategy. Set it against the retainer to see whether a small client pays. Seats scale with senders and labor scales with clients, which is why the cheapest stack at two clients can be the most expensive at eight.
Approval Records
“The client approved it on a call” is a memory. An approval record holds the exact wording or claim, the named person who approved it, the date, the campaign or segment it covers, and what triggers re-approval, such as a price change. Operators then need one rule about edits: changing a greeting is free; changing a claim, a number or a named customer needs a new record. It provides a traceable answer when a buyer forwards a message to the client's chief executive and asks who wrote it.
Qualification Belongs to the Client, and So Does the Clock
Suppose one client sells payroll software and wants every positive reply passed over the same business day, because its reps qualify on the call. Another sells maintenance contracts for industrial equipment and wants site count and current provider confirmed first. Both definitions are reasonable. Left in operators' memories, they drift toward whichever client was handled last, so each belongs in that client's brief.
The handoff that fails most quietly is the one that was sent and never picked up. A buyer asks a pricing question, the operator forwards it, the client's sales owner is on leave, and five days pass. The agency reported a qualified conversation; the buyer experienced silence. Give every handoff an acknowledgment state, a fallback owner named by the client, and a time limit after which the agency escalates. Report time to acknowledgment for each client every week, because it is a number the client can fix and the agency cannot.
Decide the Exit in Week One
At onboarding, agree what the client receives when the engagement ends: opt-outs and exclusions, open conversations with their current state, approved content with its records, qualification notes, and removal of agency access from every sender on the register. Then run the export on sample data in the first week, because an untested export is only a line on a vendor's help page. What the agency may retain afterward is a matter for the contract and for counsel; the stack's job is to make the agreed answer possible to carry out.
For a one-person agency with a single client, most of this can live in one document. The register, the approval record and the tested export begin to pay at the second client, when the answer to “whose is this?” stops being obvious.