Case study: 100 Discord accounts moderating client servers

this is a case study about a moderation agency project i ran from late 2024 into early 2026. the client was a boutique community management shop that handled Discord servers for small gaming studios, NFT projects, and a few DeFi protocols. their problem was simple: they had signed contracts for 23 active servers, each needing dedicated mod accounts, backup accounts, and a few “community participant” accounts to keep conversation moving during dead hours. one human identity could not cover all of that cleanly, and they did not want every moderator’s personal account tied to a client server permanently.

my job was to build and maintain the account infrastructure. by the time we hit peak operations we were running 100 live Discord accounts across those servers, all routed through residential proxies, managed through an antidetect browser setup, and warm enough that Discord’s trust scoring treated them as regular users. the headline result: we kept a 94% account survival rate over a 90-day rolling window, served 23 client servers without a single client-visible outage, and the agency charged clients between SGD 800 and SGD 2,200 per month per server depending on activity requirements.

i want to be direct about something upfront: operating multiple accounts on Discord sits in a gray area under Discord’s Terms of Service, which prohibits using self-bots and automation to misrepresent identity. we did not use self-bots. every account was operated by a human moderator during active hours. the accounts existed to separate client work from personal identities and to provide redundancy, not to fake engagement through automation. even so, Discord can and does ban accounts it flags as coordinated. know the risk before you replicate this.

the setup

antidetect browser. we used AdsPower on the team plan at roughly USD 30 per month for the base tier, then scaled to their growth plan at around USD 80/month once we needed more profile slots. each Discord account lived in its own browser profile with a unique canvas fingerprint, WebGL hash, and timezone. this is the baseline, non-negotiable layer. i’ve written more about how antidetect browsers compare over at antidetectreview.org/blog/ if you want a deeper breakdown.

proxies. we used residential rotating proxies from two providers, IPRoyal for the bulk of accounts (roughly 70) and Bright Data for the 30 accounts that needed sticky IPs tied to a consistent city. IPRoyal’s residential plan was running about USD 3.50 per GB in early 2025. Bright Data’s residential is more expensive, around USD 8-10 per GB depending on your plan, but the IP quality and geo-targeting precision is noticeably better. total proxy spend averaged USD 160-220 per month across both providers depending on session volume. see the proxy sourcing guide on this site for how we evaluated providers.

account creation and phone verification. we did not use virtual number services for phone verification. this matters. Discord increasingly flags accounts verified on SIM bank numbers, and verified-with-real-SIM accounts have measurably better trust scores based on what we observed in ban rate patterns. we used physical SIMs routed through a GSM modem pool. cost per SIM (prepaid, local SG and MY numbers) was roughly SGD 5-8 each. for accounts that needed non-SEA phone numbers we used staff in those regions to verify on their own devices. this added friction but the survival rate difference was real.

account aging. newly created accounts went through a 21-day warm period before being handed to client servers. during that period each account joined 5-8 unrelated public servers, posted occasionally, and added a few friends. this is not revolutionary advice but it is often skipped. we used a shared Google Sheet to track each account’s age, verification status, current server load, and last activity date. no fancy tooling, just discipline.

scale numbers at peak: - 100 active accounts - 23 client servers - 4-6 accounts per server (1 primary mod, 1 backup mod, 2-3 community participant accounts) - 2 human operators managing the infrastructure side - 6 part-time moderators doing actual community work across shifts

total infrastructure cost (antidetect + proxies + SIMs, excluding labor): approximately USD 420-480 per month.

what worked

sticky IP assignment per account. early on we rotated IPs per session and saw elevated account flags during the first month. switching to sticky residential IPs, where the same account always connects from the same /24 subnet if not the exact same IP, dropped our flag rate significantly. Discord’s trust system weights login location consistency. a Bright Data sticky session for a US account always appearing from the same Chicago residential range looks very different to their systems than an account hopping between Dallas, Frankfurt, and Singapore every day.

aging past the 30-day mark before any server admin action. accounts that performed admin-level actions (banning users, deleting messages, managing roles) before hitting 30 days old were banned at roughly 3x the rate of aged accounts. we enforced a hard rule: no admin permissions for any account under 30 days. community participant accounts had no restrictions because they were not being given elevated roles.

rotating primary mods every 90 days. we kept backup accounts warm on every server by having them participate regularly, not just sit idle. every 90 days we would rotate which account held the primary moderator role. this meant backup accounts never went cold and we always had a tested fallback. the first time a primary account got banned mid-shift, this saved us. the backup was already trusted and known to the server community.

keeping each account to a maximum of 5 servers. Discord does not publish a hard limit for server membership (the user limit is 100 servers per account), but operationally we found that accounts spread across more than 5 active client servers showed higher ban rates, probably because the activity pattern looks unnatural. keeping accounts focused also made operator workloads manageable.

fingerprint hygiene in AdsPower. we audited every profile quarterly using browserleaks.com to check canvas, WebGL, font, and timezone consistency. profiles where the antidetect configuration had drifted (e.g., timezone mismatch with the proxy’s geo) got rebuilt. this took about 2 hours per quarter and probably saved us a dozen accounts.

what broke

the March 2025 Discord wave. in March 2025 Discord ran what appeared to be a coordinated account review sweep. we lost 14 accounts in 72 hours, all of them accounts that shared a subnet overlap with two flagged accounts. the likely cause: two of our accounts had been used for something outside the client scope (a moderator had logged into their personal account on the wrong AdsPower profile, mixing fingerprints). the lesson was brutal, direct, and obvious in hindsight: physical isolation between client-scope accounts and any personal use. we rebuilt the 14 accounts and implemented a rule that no personal Discord use happens within the AdsPower environment under any circumstances.

IPRoyal residential IP churn in SEA. around mid-2025 the SEA residential IP pool at IPRoyal degraded noticeably. we were getting the same IPs cycling back within 48 hours on supposedly rotating sessions. Discord had clearly seen some of these IPs before. we moved our highest-value accounts off IPRoyal to Bright Data and used IPRoyal only for the lower-risk community participant accounts. the lesson: proxy provider pool quality is not static. check your IP reputation periodically using a tool like ipqualityscore.com and be ready to migrate.

client expectation mismatch on account persona depth. two clients expected the community participant accounts to have elaborate backstories, consistent posting personalities, and post histories going back months. we had not scoped this into the contract. building convincing long-term personas at scale requires significantly more labor than a moderation-only account. we ended up hiring a freelancer through Upwork for two weeks to build out the histories at cost, and updated our intake questionnaire so every new client specifies whether they need “functional” accounts or “persona” accounts. this is now a line item in every proposal.

the numbers

these are real numbers from the project, not projections.

monthly infrastructure cost: USD 420-480 (antidetect browser, proxies, SIM pool maintenance)

agency revenue (my client’s revenue, not my direct revenue): the agency billed clients SGD 800-2,200 per server per month. across 23 servers in the final quarter, that was roughly SGD 24,000-28,000 gross per month. my fee was a flat monthly retainer for infrastructure management, not a revenue share.

account replacement cost: when an account was banned, rebuilding it (new SIM if needed, aging period, fingerprint setup) cost roughly USD 8-15 in direct costs and about 45 minutes of operator time. with ~6 account losses per month on average after the March wave, that was approximately USD 60-90 per month in direct replacement cost plus about 4.5 hours of labor.

scaling ratio: one infrastructure operator can comfortably manage 80-120 accounts with the setup described here. above 120, you need a second operator or better tooling for the audit and tracking side.

client acquisition cost for the agency: not something i tracked directly, but the agency told me most clients came through referrals from crypto project communities. no paid ads. the multiaccountops use case for discord is discussed more in the multi-account management overview on this site.

lessons

your account survival rate is a function of your IP quality, not your fingerprint quality. i spent a lot of time early on obsessing over canvas hash tuning and very little time auditing IP reputation. it should have been the reverse. start with IP quality.

separate environments for separate contexts, no exceptions. the March 2025 incident happened because one moderator blurred the line between personal and client use. the only robust fix is a policy with zero tolerance and a setup that makes the violation require deliberate extra effort, not an accidental click.

charge for account infrastructure explicitly. the agency initially bundled infrastructure costs into the moderation fee. this created pressure to cut corners when proxy costs spiked. once infrastructure was a separately itemized cost that passed through to clients, the quality of the stack improved because the budget was visible and defensible.

account aging is a tax you pay upfront or a ban rate you pay forever. 21 days feels like a long time when you have a client who wants to go live immediately. every time we skipped or shortened the aging process under pressure we regretted it within 60 days.

document your account inventory like it is production infrastructure. we used a Google Sheet but the principle is what matters. every account should have a creation date, phone number used, proxy assignment, current server list, last health check, and flag history. when something goes wrong at 2am, you need this information immediately.

proxies from the proxyscraping.org blog community surfaced a useful rule of thumb i started applying: if you can’t tell whether a residential IP has been used for spam in the last 90 days, assume it has. verify before assigning to a high-trust account. this sounds paranoid but it’s operationally correct.

you can read more about our broader multi-account tooling choices in the antidetect browser comparison guide on this site.

would I do it again

yes, with one significant change: i would not take on this kind of engagement without a written agreement that explicitly covers account loss risk and defines the agency’s liability when Discord bans accounts through no fault of the infrastructure. we had a good working relationship and the agency was reasonable when things broke, but the March incident put real strain on the engagement because nothing in the contract addressed it.

the core of the operation works. residential proxies, a clean antidetect setup, physical SIM verification, and disciplined aging will get you a reliable account inventory for Discord moderation work. the operational overhead is real but manageable with two people who are paying attention.

the margins for an agency doing this are reasonable if the account infrastructure is budgeted correctly and the agency is charging rates that reflect the actual complexity. i would not do this for a client paying SGD 300 per server per month. at that price point the infrastructure cost alone makes it not worth the risk exposure.

discord moderation as a service is a legitimate business. the infrastructure to run it at scale is more demanding than most people realize going in. this case study is an honest record of what that looks like.

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.