The agency case for one harness across 30 clients
Don't build 30 harnesses. Run one connected harness across every client, one workspace per client, at a per-client cost that a from-scratch build can't match. Here's the agency math.
If you run a marketing agency, you have felt the pull in both directions. The first is toward the build: a bespoke reporting stack, custom dashboards per client, home-grown automations stitched together with scripts. The second is toward today: ship the client's work, bill it, move on, and never build the reporting your pipeline keeps asking for.
Both feel like they're winning until you multiply them by the number of clients on your roster. That's where the choice gets clear. Don't build 30 harnesses — run one harness across 30 clients. One connected instance, configured per client, at a per-client cost the build can't match.
This is the agency case for the harness, and specifically for the harness that runs one workspace per client.
The per-client build trap
The build-first approach usually starts with one anchor client. You write the queries, wire the connections, set the schedule. It works. Then client two arrives with a slightly different stack, different questions, a different reporting template. Client three has a different ad platform mix.
Pretty soon you are maintaining:
- A separate integration for every platform × every client
- A set of dashboards that drift because they were each built by hand
- Reporting that takes meaningful hours per client per month to assemble
- A growing pile of screenshots, exports and copy-paste
The costs are not one-time. They compound with every new client, every platform change, every staff departure that takes tribal knowledge with them. The build looks free because the time is your own — until you cost it.
The harness is one build, configured many times
The harness inverts the equation. You build the engine once: one connected system that knows how to talk to Google Ads, GA4, Search Console, Meta and the rest. Every client is then a configuration of that engine, not a new build.
In Cogny's terms, that's one workspace per client. Each client gets its own workspace with its own connections, its own ledger and its own plan — connected to the client's own accounts, authorised by signing in. Nothing is pooled across clients; there is no cross-client dashboard. You switch between workspaces, and each one runs the same measured loop: one concrete change, your approval, made in the client's own account, then graded in their data.
The harness is one operation, not 30. Adding a client means adding a workspace and connecting their accounts — not re-building the plumbing.
The cost math the build can't match
The honest number for the solo operator: 30 client workspaces on Solo is $270 a month ($9 each) — one person running all of them from their own AI. Solo, Cogny AI and Cogny Pro are single-seat plans, so the operator themselves carries the work; the workspace economics stay flat per client no matter how many clients you onboard.
Contrast that with the build. The per-client cost of the build isn't the tooling — it's the maintenance hours, the drift, the reporting time, and the retraining. Those don't fall as you add clients; in a hand-built stack they rise.
The harness has a different shape: the per-client cost is roughly constant and low, and the operator effort is shared across all workspaces because the loop is the same everywhere. That's the part the build can't match.
And when a client outgrows what one operator can run, the harness scales in place: a team that shares client workspaces runs on Cloud, with per-member access — no second architecture, just more seats on the same engine.
Why agencies run one, not several
Agencies that try pieces of several tools end up with the build's cost without the build's control: N different logins, N different reconciliations, client data scattered across half a dozen surfaces. The harness is only a harness if clients are on the same engine and the same loop.
One workspace per client keeps the promise of isolation — the client's data stays the client's — while the operation stays unified. You learn the loop once and apply it everywhere. Someone who has run one client workspace can run the thirtieth on the first day, because it is the same loop.
That's also what makes it defensible when you show the client. Every measured win carries a share card you can copy as an image or send as text. The verdicts are honest by design: worked, partial, didn't work, inconclusive. Misses stay on the record — that's the point. The share card sells the work with the numbers, not with promises.
Build-vs-buy, settled for the agency
If this series has a through-line, it's the build-vs-buy case for the marketing harness. For an agency the verdict is sharper than for a single brand: a brand builds one harness for its own channels; an agency that builds is committing to maintaining many derivatives of the same thing.
The buy isn't a piece of software you configure once and leave. It's a running operation measured in the same way you measure your clients' work — a change, a verdict, a share card, repeated. And because the engine is shared, the agency's own margin is protected whether you run two clients or thirty.
Run one harness across your roster
The decision is concrete and cheap to test. Open the Cogny for agencies page, stand up a workspace for one client, connect their accounts, run one loop end to end. Then decide whether the per-client cost of that — versus the maintenance ledger of a bespoke build — is the number you want governing your next ten clients.
For teams running many clients, we offer volume pricing on a walkthrough — book an agency walkthrough and we'll quote it.
Don't build 30 harnesses. Build once; configure thirty times.