Cold Email Infrastructure for Agencies: Multiclient Setup Done Right

Agencies live and die by throughput. You promise pipeline, not theory, and every client expects leads without hurting their brand. That pressure turns technical decisions into business outcomes. Configure your cold email infrastructure poorly and campaigns stall under spam filters, addresses get burned, and renewal conversations become awkward. Configure it well and you can run reliable, scalable programs that keep inbox deliverability healthy month after month.

What follows pulls from years of building and rescuing multiclient setups. It is not a one size fits all recipe. Constraints differ across industries, geographies, and client risk tolerance. The goal is to surface the important trade offs, explain how mailbox providers really police your traffic, and give you an operating model that works when you are running ten, fifty, or two hundred domains at once.

What inbox providers reward and punish

Gmail, Microsoft, Yahoo, and the regional providers do not care about your intentions. They score behavior at the domain and mailbox level and make fast, automated decisions. The core signals are consistent:

Positive engagement outweighs volume. Replies, not opens, drive trust. A smaller volume of thoughtful, targeted campaigns is safer than a shotgun.
Consistency beats spikes. Big jumps in send volume, domain count, or daily cadence look like automation trying to game filters.
Alignment across DNS, headers, and content matters. SPF, DKIM, and DMARC alignment reduce suspicion. Mismatched envelope domains, bare link shorteners, and URL redirects raise flags.
Complaint rate is king. Even a 0.2 to 0.5 percent spam complaint rate can tank cold email deliverability for a domain. Keep it near zero.

Behind the scenes, providers maintain reputation graphs. A user marks one message as spam, that signal bleeds into the sender domain, the sending IP or shared IP pool, the linked domains, and sometimes the template itself. If your agency runs many clients through shared tools or trackers, reputation entanglement becomes your hidden enemy.

The multiclient challenge

Running a single brand is simple compared to an agency footprint. You juggle:

Multiple root brands, each with different legal and security constraints.
Unique buyer personas that require different sending styles, content density, and follow up sequences.
Divergent tech stacks. Some clients insist on using their Microsoft 365 tenant, others let you provision new domains, some need SSO and central IT review.
Scale. You might need 2 to 6 sending mailboxes per client, with daily volumes ranging from 40 to 200 messages each, plus follow ups.

The multiclient pattern that wins uses isolation where it matters and shared services where it is safe. The trick is knowing which is which.

Domain strategy that respects brand and risk

For outbound, your choice is simple in concept, contentious in practice. Should you email from the main brand domain or from a sibling domain? I rarely use the main root domain for cold outreach. Too much executive email, customer support, and transactional traffic rides on it. A reputation hit there is expensive. Instead, I use branded sibling domains that still pass the sniff test for prospects.

For a hypothetical brand acme.com selling B2B services, options that often work:

acmehq.com or acme-team.com for a general business voice.
getacme.com or tryacme.com for offers and trials, if the brand skews SaaS.
acmeconsulting.com for services arms, when relevant.

Avoid awkward mashups or long strings. Keep it pronounceable and natural. Ideally, the website for the sibling domain resolves with a lightweight brand page or at least a redirect to the main site. A cold recipient who hovers or Googles should not think you are faking the brand.

Under the sibling domain, create the sending mailboxes that reflect real humans, not “no-reply” or “sales”. Firstname variations work best. If the client has multiple sellers, mirror them, and consider role flavors like research@ or partnerships@ only when a campaign genuinely matches that persona.

Authentication that actually aligns

You need SPF, DKIM, and DMARC, aligned end to end. Basic SPF include records are table stakes. What trips teams up is alignment. The visible From domain, the DKIM d= domain, and the bounce or Return Path domain should all share the sibling domain. If your email infrastructure platform routes through a shared envelope domain, correct it or use a custom bounce domain. Alignment mistakes do not always show in preflight tools, but mailbox providers notice.

DMARC at p=none is fine during warm up. Move to p=quarantine, then p=reject, only once traffic is stable and your authentication passes consistently. Do not rush DMARC enforcement while you are still validating toolchains, especially if the client also uses the sibling domain for web forms or CRM notifications.

If the brand has an established BIMI record with a verified mark, resist the urge to reuse it on the sibling domain unless legal and brand teams approve. BIMI can help with brand trust, but it does not fix reputation problems.

Mailbox provisioning and warm up

Under each sibling domain, provision 2 to 6 mailboxes depending on ramp targets. I rarely go higher than 6, because coordination complexity grows faster than throughput. More mailboxes also tempt teams into blasting. Keep the daily target per box sane. For new boxes, 20 to 40 first touches per day is a safe launch after warm up, with automated warm ups handling 2 to 4 weeks of low friction exchanges beforehand.

Automated warm up still helps, even after the 2024 changes that limited obvious warm up exchanges. Use network warm up tools that simulate enterprise email infrastructure real threads with varied topics and timings. Pair that with manual engagement from internal accounts where possible. Forward a couple of industry articles. Ask polite questions. Let replies sit for a bit before responding, then close the loop. The goal is to show normal human behavior, not to rack up fake opens.

Mailbox settings that reduce friction:

Set reasonable display names, like “Amy Chen, Acme” rather than “Amy - Acme”.
Configure signatures with full contact details and a real address for compliance.
Keep inboxes tidy. Archive marketing blasts and update filters so internal newsletters do not drown replies.

Sending volumes, cadences, and follow ups

Cold outreach patterns that look spammy are blunt and repetitive. The safe lane is narrower and more patient. Across clients, my rule of thumb:

Cap first touch volume at 20 to 50 per mailbox per day for the first month, rising to a ceiling of 75 to 100 only when reply rates are healthy and complaint rates are near zero.
Use 2 to 3 follow ups at most for cold, spaced 3 to 7 business days apart. High frequency sequences are predictable and drive complaints.
Pause sequences on weekends for B2B. Monday mornings also tend to be noisy. Midweek mornings or early afternoons often perform best, but test this per segment.

Track replies, not clicks, as your primary success metric. Opens are dynamic and noisy. A campaign with a 3 to 8 percent positive reply rate can usually sustain itself without pushing volume into risky territory.

Content, links, and tracking choices

The most expensive mistakes in cold content are subtle. A single tracking decision can tank inbox deliverability across many clients if you share infrastructure. Link shorteners, especially branded ones tied to a central domain, collect reputation debts fast. If one spam trap or angry recipient dings the shortener domain, every campaign using it inherits the stain.

Safer patterns:

Use the client’s sibling domain for links where possible, either with vanity paths that redirect via the client’s web server or direct links to the main domain. Keep redirects minimal.
If you must track clicks, consider per client tracking domains, not a shared global one. Host them on subdomains of the sibling root with proper DNS and SSL.
Reduce link count. One relevant link is plenty in a short cold email. A calendar link is fine if the offer justifies it, but test variants that request a simple reply first.

On content itself, plain text with a clear, specific reason for outreach still wins. Write as if you were asking for directions, not pitching a product brochure. Make it short, rooted in the recipient’s world, and honest about the ask. If your agency does personalization at scale, maintain quality controls: spot check 10 to 20 emails daily for odd merges, awkward phrasing, or hallucinated facts. It takes ten minutes and prevents dozens of complaints.

Data hygiene and list sources

Your reputation depends on who you email as much as what you say. The fastest route to spam traps and high bounce rates is a sloppy list. Even good data vendors need filtering. Before import:

Validate emails with a reputable verifier, but treat “catch-all” results cautiously. Send to them only with highly relevant targeting and at low initial volumes.
Segment by recency and source. Freshly scraped data typically performs worse than relationship adjacent targets like webinar attendees or mutual communities. Send gentler, smaller batches to cold scraped segments.
Cull hard bounces immediately, and do not retry them across other mailboxes or domains. Repeated attempts to the same dead address look like bot behavior.

For agencies that inherit legacy lists, quarantine them. Test 100 to 300 contacts in a controlled batch with enhanced monitoring and a gentle message. Do not plug a legacy list into a new domain at full volume, however tempting the quick win may look.

Infrastructure choices: in house, hybrid, or platform

You can run cold email from native Google Workspace or Microsoft 365 with careful tuning, or from an email infrastructure platform that manages routing, reputation isolation, and reporting. There is no universal best choice. Consider:

In house with native providers gives you maximum authenticity and the same sender reputation model as normal business mail, but you shoulder the routing care and cannot easily scale past a few dozen boxes per provider tenant without complex quotas and governance.
A dedicated email infrastructure platform can provide per domain envelope domains, per client tracking isolation, and smart throttling. The risk is shared reputation if the platform mixes your traffic with noisy neighbors. Look for platforms that offer private pools or per account isolation at the IP and bounce domain level.
Hybrid models work well for agencies with a few high touch accounts and many lighter ones. Run VIP clients through native Workspace or 365 for maximum trust, and run the rest on a platform that gives you controls and reporting.

Ask platform vendors blunt questions about shared components. Which domains touch my mail? Are envelope domains unique per sender domain? Are IP pools private? Can I bring a custom bounce domain? How do you detect and isolate bad actors in your network? Vague answers predict pain later.

Monitoring that catches trouble early

Deliverability lives in the trend lines. Watch what matters and act before the floor falls out:

Domain level reputation via Google Postmaster Tools for any domain that sends meaningful volume to Gmail. Track spam rate, domain reputation, and feedback loop data.
Seed testing helps, but treat it as directional. Seeds can be too clean. Pair them with live metrics like inbox placement inferred from reply rates and from occasional test sends to real, consenting addresses across providers.
Blocklist monitoring for your sibling domains and any tracking or bounce domains. If a domain lands on Spamhaus or SURBL, stop sending and remediate. Most minor blocklists do not correlate strongly with inbox placement, but you still want a clean bill of health.
Complaint and bounce alerts at the mailbox level. If a single mailbox has a sudden bounce spike, pause it immediately and test from another mailbox on the same domain. If complaints rise, pause the whole domain’s outreach and rework the target and message.

When something feels off, it usually is. A sudden drop in positive replies, open rates collapsing at Microsoft domains, or calendar links showing fewer clicks, all hint at reputation drift. Do not ride it out. Short pauses and targeted fixes beat plowing ahead and burning a domain.

Isolation boundaries that prevent cross contamination

The multiclient setup excels when clients cannot hurt each other. Strong isolation means:

Separate sibling domains per client, never shared. Even companies owned by the same holding group deserve their own sender identities.
Per client tracking domains and web assets. If you run scheduling links, use the client’s scheduling domain or a per client subdomain, not a central agency link.
Distinct routing. If a platform allows it, use per client envelope domains and dedicated IP pools. If you send from native Workspace or 365, use separate tenants or at least separate organizational units with clear quotas.
Segregated suppression lists. A global opt out is legally wise when the same recipient appears across multiple client campaigns, but do not let a single client’s internal suppression spill into another without consent. Handle it with a shared “do not contact” registry keyed on the recipient address, with per client exceptions documented.

With these walls in place, when one client experiments with harder messaging or a riskier list, the blast radius stays small.

Playbooks that scale beyond the hero operator

What breaks at 20 clients is not the DNS, it is the human workflow. Create simple playbooks so the team can run campaigns with confidence:

Preflight checklist before any new domain goes live: DNS alignment double check, Postmaster provisioning, forward and reply testing, unsubscribe and postal address verified, warm up in progress, seed tests across Gmail, Microsoft, Yahoo, and the top regional provider for the target segment.
Daily health routine: quick look at new bounces and complaints, spot check sequences for awkward personalization, confirm follow ups are throttled correctly, skim Postmaster graphs.
Weekly review: per client reply rate trends, complaint rate, domain reputation grades, and any seeds showing Promotions or Spam. Adjust volumes and content based on these signals.
Incident response: when reputation drops, pause, diagnose, and route around. Use a backup domain only after the root cause is fixed. Spinning up new domains without fixing targeting or content just pushes the problem down the road.

Document these as short, actionable notes the team can follow. Fancy SOPs gather dust. The best playbooks fit on a single page and match how work actually flows in your tools.

Legal and ethical guardrails

Compliance is not optional. For B2B cold outreach in many jurisdictions, a legitimate interest argument holds if you offer something relevant and provide clear opt out. Still, your agency brand benefits from gold standard behavior:

Include a simple, working unsubscribe in every message. A reply to opt out should also be honored, quickly.
Use a real physical address in the signature, typically the client’s business address.
Respect regional rules. Canada’s CASL and Germany’s UWG are stricter than the United States CAN SPAM. When in doubt, narrow your target and use softer messaging to validated contacts.

Playing fair keeps complaint rates low and reputation high. It also makes it easier to defend your program if a client’s legal team asks hard questions later.

A practical multiclient architecture

A workable baseline architecture for a 25 client agency looks like this:

For each client, register one sibling domain that passes brand review. Set up DNS at a provider you control so you can manage SPF, DKIM, DMARC, tracking CNAMEs, and any custom bounce domains without waiting on corporate IT for every change.
Create 2 to 4 mailboxes per domain to start, using Workspace or a reputable email infrastructure platform with per client isolation. Warm them for 3 to 4 weeks.
Configure SPF to include only the senders you truly use for that domain. Publish DKIM keys unique to the domain. Set DMARC to p=none with rua and ruf reporting to an inbox you actually monitor.
Stand up per client tracking on a subdomain, like link.client-sibling.com, with SSL and domain alignment. If using scheduling tools, use a branded scheduling domain as well.
Build campaigns in smaller, targeted segments of 200 to 600 contacts. Throttle initial sends to 20 to 40 per mailbox per day. Watch reply and complaint rates. Scale only when the numbers justify it.

This setup keeps your agency nimble. You can bring a new client live in 10 to 14 days, with warm up overlapping copy and data work, and you are not waiting for a massive IT change window.

Troubleshooting stubborn inbox placement issues

Every agency ends up with a client who sits mostly in Microsoft spam, while Gmail looks fine, or vice versa. Diagnosis needs patience:

Compare engagement by provider. If Gmail replies are healthy but Outlook replies vanish, consider content filters. Outlook is stricter about links and attachments. Remove tracking, send pure plain text, and test again.
Inspect error codes on soft bounces. Microsoft’s SCL scores and 550 5.7.x hints point to policy blocks, not transient errors.
Test from a fresh mailbox on the same domain with a brand new, short message. If it lands in inboxes, your template or thread history is the likely culprit.
If domain reputation is low in Google Postmaster, slow sends for a week, prune sequences, and push for higher intent segments. It can take 2 to 4 weeks to climb back from “low” to “medium.”

Avoid the reflex to swap domains instantly. That buys short term relief but delays real learning. Only rotate when the business needs continuity while you fix root causes.

When to use dedicated IPs, and when not to

Cold email at B2B volumes rarely needs dedicated IPs. Mailbox providers key on domain and content behavior more than IP for the volumes most agencies send. Dedicated IPs make sense if you send hundreds of thousands of messages per month with steady patterns and you have the expertise to manage warm up and pool health. For typical agency volumes, shared IPs at reputable providers, paired with strong domain controls, perform better and require less maintenance.

If you do go dedicated, run separate pools per client or at least per risk tier. Do not mix experimental campaigns with your steadiest performers.

Tooling that pays for itself

You do not need a sprawling tool stack to excel. You do need a few reliable instruments and habits:

DNS and domain management you can control without tickets. Bulk editing and templated records save hours.
An email infrastructure platform or sending tool that exposes throttling, per mailbox scheduling, and clear reporting. Vanity metrics do not help you debug.
Deliverability monitoring that aggregates Postmaster, blocklist, and seed data into a daily snapshot.
A CRM or tracker that records replies and meetings against campaigns, so you can defend your impact and adjust spend.

Be skeptical inbox deliverability rate of magic scores and black box “health” bars. You want raw signals and the ability to test changes, not guesses wrapped in a glossy dashboard.

A short checklist of non negotiables

Unique sibling domain per client, with SPF, DKIM, and DMARC aligned.
Per client tracking and scheduling domains, no shared agency links.
Warm up for new mailboxes, then ramp slowly based on real reply and complaint data.
Short, relevant emails with minimal links, personalized responsibly.
Daily monitoring and fast pauses when metrics slide.

Rapid triage when deliverability dips

Pause new sends on the affected domain. Let follow ups finish if they show healthy replies, otherwise pause sequences entirely.
Test plain text, no tracking, from a fresh mailbox on the same domain to a small, high intent segment. Watch provider specific behavior.
Review complaints, bounces, and recent list sources. Remove any risky segments and update suppression lists.
Reduce volumes for a week and reintroduce sends gradually once replies normalize and Postmaster trends improve.

A brief story from the field

A services client came in with two burned root domains after months of high frequency cadences and a global tracking link shared across three brands. Gmail Postmaster showed “bad”, and Outlook was worse. We rebuilt using a new sibling domain, moved to per client tracking on a subdomain, cut link count to one or none, and tightened the target to 300 accounts with buying signals. For the first two weeks, we capped at 30 first touches per mailbox per day. Positive replies climbed to 7 percent, complaint rate fell under 0.1 percent, and Postmaster moved from “bad” to “medium” in 17 days. By week six, the program ran at 70 first touches per mailbox per day across four boxes, with steady meetings and no new blocklist hits. The win was not the domain swap. It was isolation, restraint, and content that warranted a reply.

Where agencies over optimize

It is easy to chase the wrong levers. Over rotating domains without fixing targeting burns goodwill. Endless A B tests on subject lines rarely matter if the audience fit is off. Aggressive follow ups look busy but depress brand value and drive complaints. Focus on the pieces that compound: clean data, precise relevance, domain alignment, and consistent sending behavior.

The quiet advantage of process

Great cold programs feel boring in a good way. The domains are healthy. The mailboxes warm slowly. The copy is calm and specific. Volumes climb when the audience responds, not because a dashboard flashed green. That rhythm lets an agency handle twenty or two hundred client programs without daily fires. It keeps inbox deliverability intact and positions you to scale on your own terms.

Cold email is not a dark art. It is systems work wrapped around human conversations. With the right infrastructure, you can give every client a fair shot at the inbox, protect their brand, and show results that stand up in any review.