Agency work on Telegram breaks in a predictable way. It starts with one account doing outreach for two clients because setting up a second one felt like overhead. Six weeks later that account is restricted, and now two clients have a problem instead of one. Or worse: a message intended for client A’s market lands in client B’s community, and the conversation you have next is not about deliverability.
Almost every operational rule below is downstream of one principle. Anything shared between clients is a channel for one client’s problem to become another’s.
Isolate the accounts
One client, one set of Telegram accounts. No exceptions, including for the small clients, including at the start.
This costs you setup time and it buys three things.
Contained restrictions. When an account gets limited, and eventually one will, the damage stops at one client’s campaign rather than crossing your whole book.
Clean handover. When a client leaves, you hand back or retire their accounts. If those accounts also carry your other clients’ history, you cannot do either, and the offboarding conversation gets uncomfortable.
Honest attribution. Reply rate per client means something only if the account was doing one job.
The corollary matters too: an account should have a consistent network location and device fingerprint, and it should keep them. Rotating a shared pool of accounts and proxies across clients looks exactly like the evasion behaviour Telegram is built to catch. We went through the mechanics in session strings, devices and blast radius.
Isolate the data
Client A’s contacts must not be visible to the person working on client B, and they certainly must not appear in a suppression check that leaks the fact that they exist.
Practically, that means a separate workspace per client rather than a tag on a shared one. Tags are a convention, and conventions fail on a Friday afternoon when someone filters wrongly. A boundary that the software enforces does not.
The same applies to your own staff. Freelancers who work on two accounts should have access to two workspaces, and losing one of those should not require an audit of the other.
Structure the week so nothing depends on memory
The operational load of an agency is not the sending. It is the fifteen simultaneous states. A weekly rhythm handles this better than any dashboard.
- Monday: account health. Every account, every client. Any restriction, any error rate above normal, any account that has drifted from its usual network location. This is the meeting that prevents the expensive ones.
- Midweek: campaign quality. Reply rate per campaign against the floor you set at launch. Anything under it stops for a rewrite rather than continuing quietly. The rules are in stop rules.
- Friday: pipeline per client. Deals without a next step, deals older than the normal cycle, and what actually converted. Fifteen minutes per client, not an hour.
- Monthly: the report the client reads. Which should be generated from the system, not assembled by hand, because the version assembled by hand is the one that goes out with last month’s numbers in it.
The unit economics nobody calculates
Agencies price Telegram outreach one of three ways, and the choice decides which clients you can afford.
Per account per month. Simple, predictable, and it aligns badly with results. It works when the client understands they are buying capacity.
Per qualified conversation. Aligns better, and requires a definition of “qualified” that both sides agree on in writing before the first campaign. Without that definition it becomes a monthly argument.
Retainer plus performance. The usual endpoint for agencies that survive. The retainer covers account warming, list work and the operational load, which are real costs that produce nothing visible in month one, and the performance component covers the part the client actually cares about.
Whichever you pick, price the boring parts explicitly. Account warming takes weeks of doing nothing productive. List research takes longer than sending. Handling replies is the largest cost of a successful campaign and the one most proposals forget entirely. An agency that prices only the sending is subsidising its own success.
What clients should ask you, and what you should be able to answer
The good clients ask these. Be ready.
- Whose accounts are these, and what happens to them when we stop working together?
- Can anyone at your agency see our contact list who is not on our account?
- Where is our data stored, and under what legal basis are you contacting people on our behalf? For European markets this is a controller and processor question, and “we use a tool” is not an answer.
- What is your stop rule, and who decides to restart?
- What is the reply rate, per campaign, per step?
If you can answer all five without preparing, you are running a real operation. If you cannot, the first client audit will be memorable.
Where Cavyro fits
Cavyro supports a workspace per client, each with its own Telegram accounts, contacts, pipeline, campaigns and reporting, and its own billable plan. Access is role-based, so a freelancer on one account cannot see another, and a client offboarding is a workspace you hand over or delete rather than a data extraction project.
Campaigns run through the client’s own accounts with per-account caps, quiet hours and automatic pauses on Telegram warning signals, and reports give you the per-client numbers that go into the monthly deck. Everything sits on EU infrastructure, which makes the processor conversation short. See security and GDPR for the specifics.
If you are currently running several clients out of one shared setup, start free for 5 days and split the riskiest one out first.