Case study: how a YouTube agency runs 60 channels with one stack
Case study: how a YouTube agency runs 60 channels with one stack
Running one YouTube channel is hard enough. Running 60, for paying clients, across different niches, with a team spread across three time zones, is an operational problem most people don’t survive. I’ve been tracking a mid-size agency, based out of Eastern Europe, that figured out a repeatable system for doing exactly this. I’ll call them Cascade here, since they haven’t gone public with their methods. What I’ll describe is what I’ve observed and verified over about eight months of direct contact.
Cascade started in 2022 with four channels under management. by early 2026 they were at 63 channels, about 18 clients, two full-time editors, three part-time VAs, and one ops manager. monthly retainers run between $1,200 and $3,500 per client depending on output volume. the business works. it is not glamorous and it is not passive, but the unit economics are defensible and the ops layer is the reason why.
the headline result: they cut per-channel management overhead from roughly 11 hours per month to under 3, and they did it without a proprietary SaaS build. everything runs on tools you can buy today.
the setup
the stack has five layers: account isolation, content production, scheduling and publishing, analytics aggregation, and team coordination.
account isolation. this is where most agencies fall over before they even start. when you manage YouTube channels for multiple clients, you’re working inside multiple Google accounts. Google’s Terms of Service do not prohibit agencies from managing client channels via delegated Brand Account access, and that’s the legitimate path they use by default. but some clients want full channel separation, different Google accounts entirely, no cross-contamination in the event one account gets flagged for a policy dispute. for those cases, Cascade uses AdsPower at the team plan tier, which runs about $99/month for 10 browser profiles and scales to $199/month for 100 profiles. each profile gets its own persistent fingerprint and is paired with a residential proxy. they use Smartproxy on a traffic-based plan, averaging about $90-120/month at their current volume.
the antidetect setup is not about evading YouTube, it’s about preventing accidental account linkage when a VA logs into the wrong session or a shared machine auto-fills a credential. for a full walkthrough of how to configure browser profiles for multi-account ops, see our antidetect browser setup guide. if you want a deeper comparison of antidetect tools, antidetectreview.org/blog/ has done more systematic testing than I have.
content production. scripts are generated with the Claude API (Anthropic’s claude-sonnet tier, roughly $0.003 per 1k output tokens). a template system in Notion defines the format per niche, talking points, CTA structure, thumbnail brief. editors work from there. CapCut handles most editing via template chains for faceless content; Descript handles anything with a talking head. raw footage from clients goes into a shared Google Drive folder per client, organized by month.
scheduling and publishing. YouTube’s native scheduler handles most of this. for bulk operations, they use the YouTube Data API v3, which allows programmatic video uploads, metadata writes, and playlist management. they built a small internal Python script that takes a CSV of video paths, titles, descriptions, and scheduled times and uploads in batch. the script took about two days to write and has been running without significant changes for over a year.
analytics aggregation. YouTube Studio’s per-channel dashboard is fine for individual channels but useless for portfolio views. they pull data into a Google Looker Studio dashboard via the YouTube Analytics API. it shows RPM, watch time, subscriber velocity, and CTR per channel, updated daily. setup time was about one day. ongoing cost is zero beyond the API quota, which they’ve never come close to hitting.
team coordination. Airtable runs the content calendar. each row is a video: client, channel, script status, edit status, publish date, review flag. automations in Make (formerly Integromat) ping a Telegram group when a video moves from “edit done” to “needs review,” and again when it goes live. the ops manager reviews the Airtable board once per morning, takes about 25 minutes for the full 60-channel portfolio.
total stack cost at 60 channels: roughly $550-650/month including proxies, AdsPower, Airtable, Make, and API costs. at $1,200 minimum retainer across 18 clients, the overhead is under 2.5% of revenue.
what worked
templated scripting per niche. the biggest time save wasn’t AI, it was the template. each niche (finance, fitness, home improvement, personal development) has a Notion doc defining the opening hook formula, segment structure, CTA wording, and things to avoid. the Claude API prompt pulls from this template. output quality is consistent enough that editors rarely send scripts back for rewrites. without the template, AI-generated scripts for 60 channels per month would be chaos.
the CSV upload pipeline. manually publishing 15-20 videos per week across 60 channels would take a full-time person. the batch upload script reduced this to about 90 minutes per week total, mostly spent reviewing metadata before the script runs. the YouTube Data API has a daily upload quota of 10,000 units; a standard upload costs 1,600 units, meaning you can upload about 6 videos per day per API project. Cascade registered multiple API projects under different Google Cloud accounts to stay within quota. this is documented in the API quota documentation and is a known pattern.
client-side Brand Account delegation. for clients who want Cascade to manage an existing channel rather than a new one, YouTube’s Brand Account system lets you add a manager email without transferring ownership. the client keeps full owner access; Cascade gets manager-level access, which is enough for uploads, scheduling, and analytics. no credential sharing, no account risk for either party. this is how the relationship with about 70% of their clients works.
niche-specific thumbnail templates. Canva templates per niche, locked so VAs can only change the text and swap the image. CTR variance between good and bad thumbnails in a portfolio this size is very visible in aggregate. by standardizing the format and only varying the text, they keep CTR in a predictable band. finance channels average around 4-5% CTR, fitness higher at 6-8%, based on what they’ve shared with me. these are not outlier numbers; they track with what YouTube itself reports as typical for established channels in those categories.
weekly performance reviews, not daily. checking analytics daily for 60 channels is a trap. instead, the ops manager reviews the Looker Studio dashboard every Monday morning and flags any channel where watch time or CTR has dropped more than 20% week-over-week. everything else runs on autopilot until the next review. this single change reduced “noise anxiety” in the team significantly.
what broke
the proxy rotation problem. in the first six months, they used datacenter proxies from a budget provider. three channels got their sessions flagged, not banned, but prompted to re-verify via phone. the fix was moving to residential proxies. residential IPs cost more (Smartproxy’s residential plan starts around $7/GB, compared to $1-2/GB for datacenter), but the flag rate dropped to zero across the next eight months. for any operation involving persistent Google account sessions, datacenter proxies are a false economy.
VA access control and credential leaks. in the early setup, VAs had direct login credentials for client accounts stored in a shared LastPass vault. one VA left, didn’t terminate cleanly, and two client accounts had unauthorized session activity within 48 hours. the fix was strict: all access goes through AdsPower profiles. VAs log into the antidetect browser, which has a profile pre-loaded with the correct account. they never see the underlying credentials. session data lives in the profile, not on the VA’s machine. this is now a hard rule and it has held.
API quota exhaustion during a upload push. a client delivered 30 videos at once before a product launch. the batch script hit quota mid-run and stopped. half the videos went up, half didn’t, and the scheduled dates were already embedded in the metadata for the failed uploads. the fix was a queue system with retry logic and an alert when uploads hit 80% of the daily quota. Make handles the alerting; the script now writes failed uploads to a retry queue and picks up the next day automatically. this took a day to build and has never been an issue since.
the numbers
at 60 channels across 18 clients, the business produces roughly $25,000-45,000 in monthly retainer revenue depending on the client mix and output commitments. i’m not publishing their exact revenue at their request, so i’m giving the range based on what I know of their pricing tiers.
ops cost breakdown monthly: AdsPower $199, Smartproxy $110, Airtable $40, Make $29, Claude API ~$80, Google Workspace ~$72 (12 seats at $6), miscellaneous tools ~$50. total: approximately $580.
labor: two full-time editors at local market rates, one ops manager part-time, three VAs at roughly 10 hours each per week. total labor cost is approximately $6,500-8,000/month depending on project load.
margin after labor and tools at the midpoint of revenue: roughly 50-60% gross. this is consistent with what I’d expect for a productized service agency at this scale that has invested in ops infrastructure.
they added 15 new channels in the last 6 months without hiring additional staff. that’s the indicator that the ops layer is working.
lessons
the bottleneck is almost never content quality. for most niches, “good enough” content published consistently outperforms “great” content published sporadically. build the ops layer first. content quality is easier to improve than ops debt.
separate the client relationship layer from the channel management layer. clients want to know their channel is growing. they do not want to understand your proxy setup. keep those conversations completely separate. document your ops stack internally but never lead with it in client calls.
residential proxies are not optional for persistent Google sessions. if you’re managing Google or YouTube accounts at any real scale, datacenter proxies will eventually cause problems. the price difference is real but so is the risk. for more on proxy types and their tradeoffs, see our guide to residential proxies for multi-account work or proxyscraping.org/blog/ for more systematic comparisons.
standardize before you automate. Cascade spent about three months standardizing their templates, workflow naming conventions, and folder structures before they wrote a single automation. operators who automate a chaotic process get automated chaos.
brand account delegation is underused. most agencies I talk to either manage clients by sharing credentials or by building on new accounts. the delegation path through Brand Accounts is the right answer for most client relationships. it’s documented, it respects account ownership, and it doesn’t require any workarounds.
the reporting layer is a retention tool. the Looker Studio dashboard is shared with every client on a read-only link. clients can check their channel stats any time without asking. this eliminated about 80% of “how are we doing” check-in requests and materially reduced churn. if you’re running any kind of content service, a shared live dashboard is worth more than any monthly report PDF.
for more on building multi-account ops stacks, see the /blog/ index for related guides on account management and automation tooling.
would I do it again
if I were building a YouTube agency from scratch today, this is roughly the stack I would use. the tools are not exotic: Google Workspace, Airtable, Make, the YouTube Data API, AdsPower, and residential proxies from an established provider. nothing here is proprietary or hard to replicate.
the harder part is the discipline. Cascade’s ops manager told me that the biggest threat to their efficiency isn’t tool failure, it’s scope creep. a client asks for “just one extra video this month” outside the retainer, you say yes, the template doesn’t cover the format, a VA improvises, quality drops, now you’re debugging a one-off instead of running the system. the system works because they protect it.
the 60-channel number is also not a limit. they’ve told me they think they could reach 100 channels on the current stack with one additional VA and no new tools. i believe them. the infrastructure is already there. the bottleneck at that point is editor capacity and client onboarding throughput, both of which are linear problems, not systems problems.
one thing I’d flag for anyone starting out: be careful about the YouTube Terms of Service around automated actions. bulk uploads via the Data API are explicitly permitted. automated engagement actions like view inflation or comment automation are not. Cascade does not do any engagement manipulation, and I’d recommend against it regardless of what a client asks for. the downside risk is not worth it when the legitimate stack already works.
if you’re managing five or more channels and still doing everything manually, this case study is a blueprint. you don’t need to build anything custom. you need to wire together tools that already exist and enforce process discipline around them.
Written by Xavier Fok
disclosure: this article may contain affiliate links. if you buy through them we may earn a commission at no extra cost to you. verdicts are independent of payouts. last reviewed by Xavier Fok on 2026-05-19.