Email in Marketing Cloud is four things that have to work together: a template system marketers can use without a developer, personalization that degrades gracefully when data is missing, an audience and suppression layer that decides who receives anything, and the sender authentication and reputation work that decides whether it arrives at all. Twopir Consulting builds all four, in that order of consequence.
Deliverability gets treated as a launch checklist item and it is not one. It is a property of how you authenticate, who you send to, how often, and what happens when people stop engaging — which makes it an architecture decision that continues to be made every week after go-live.
It also compounds in both directions. Good sending behavior improves placement, which improves engagement, which improves placement further. Poor sending behavior does the same thing in reverse, slowly enough that no single week looks alarming.
Four things determine it. None are creative, and all four are fixable — and all four are set during the build, which is why they belong in the Marketing Cloud architecture rather than in a launch checklist.
SPF, DKIM and DMARC correctly configured on every domain you send from, including the ones somebody set up for a campaign two years ago and forgot. Mailbox providers increasingly treat authentication as a precondition rather than a signal, and a partially authenticated estate is a slow leak that is invisible until placement drops.
Continuing to mail addresses that have not engaged in a year is the single most common self-inflicted deliverability problem. A sunset policy feels like giving up on an audience; in practice the dormant segment is actively suppressing placement for the engaged one. Suppression, bounce hygiene and complaint handling belong to the same discipline.
Reputation problems frequently originate upstream of marketing entirely — a form with no confirmation step, a purchased list somebody imported once, an event badge scan that produced addresses nobody consented with. The sending behavior is a symptom; the acquisition practice is the cause, and it is usually owned by a different team.
Sending reputation attaches to infrastructure and domain. New infrastructure starts without the history mailbox providers use to decide where mail lands, so volume has to ramp on a schedule. This is not optional and it cannot be accelerated by resourcing — which is why it belongs in the plan rather than in the launch week.
The measure of a good email build is not how it looks in the reveal. It is whether a marketer can produce a correct, on-brand, personalized email on a Tuesday afternoon without asking anyone for help.
Reusable blocks, brand elements locked, content areas open, and a structure that holds together when somebody puts a long product name in a short space. Built once, properly, it removes most of the reason anyone would want custom code later.
The hard part of personalization is not the merge field — it is what renders when the value is missing, malformed, or in a different case from what the copy assumed. An email that greets a customer as "Dear ," has done more damage than an unpersonalized one.
Who receives the email, who must not, and how a customer changes that themselves. The suppression side is the one that matters commercially — a well-crafted email sent to someone who asked not to receive it is not a marketing success.
A repeatable pre-send process rather than a person remembering. Most email incidents are not exotic — a broken link, a merge field that rendered empty, a send to the wrong audience — and all three are caught by a checklist that runs every time.
Not aspirational best practice — the actual list, because every item on it exists because something once went out without it. Most take seconds; the expensive ones are the two nobody thinks to do.
| Check | What it catches | Frequently skipped because |
|---|---|---|
| Audience volume against expectation | A send about to go to ten times the intended audience, or to none. | The audience "looked right" in the preview, which shows a sample rather than a count. |
| Personalization with an edge-case record | Empty greetings, broken formatting, a fallback that was never configured. | Previews default to a tidy record. The problematic ones are records with missing data, which is exactly what will not be previewed. |
| Link check, including dynamic links | A tracked link pointing at a staging URL, or a link constructed at send time that resolves to nothing. | Static links get clicked in the proof. Links assembled from data do not exist until send. |
| Rendering across real clients | A layout that collapses in one mail client, or inverts unreadably in dark mode. | It looked correct in the editor, which is not one of the clients your audience uses. |
| Suppression applied | Sending to someone who opted out, complained, or has an open escalation. | Suppression is assumed to be automatic. Where it is implemented per journey rather than centrally, it frequently is not. |
| Send classification and sender profile | A commercial send going out under a transactional classification, or from the wrong sender identity. | It is set once at template level and inherited silently, so nobody re-checks it. |
| Reply handling | Replies going to an unmonitored mailbox, including out-of-office and customer questions. | Nobody replies to marketing email — except that they do, and those replies are a deliverability signal as well as a customer. |
No deliverability benchmarks on this page — see the claims note in the source header Last reviewed 23 September 2026
None of these look like mistakes while they are happening. Each is a reasonable local decision whose cost accumulates somewhere nobody is watching.
If placement has already slipped, the sequence for recovering it is on the optimization page — and it starts with the list, not the creative.
Mailing everyone because the list is an asset — The dormant segment feels like latent value and behaves like a liability. Continuing to send to people who have not engaged in a year suppresses placement for the people who do, which means the asset is actively taxing the audience that matters.
A template per campaign — Forty templates, each a slightly modified copy of another, none the acknowledged master. A brand change then requires forty edits, several get missed, and the inconsistency is visible to customers before it is visible internally.
Personalization without fallbacks — It works on every record anyone tested, because the records anyone tests are complete ones. The incomplete ones render as an empty greeting or a stray comma, and they go to real customers.
Treating replies as noise — Reply handling left pointing at an unmonitored address. Customers who reply get silence, which costs goodwill — and engagement signals, including replies, are part of how mailbox providers assess you.
Two categories here, and it is worth knowing which you have. The first three signals are an operating problem — the team cannot execute. The last two are a deliverability problem, and those get worse on their own.
Orchestration: when the email is sent, to whom, and what happens next.
Recovering inbox placement, and the order the levers have to be pulled in.
AMPscript and content assembly, and the test for whether a requirement genuinely needs it.
Adding channels, and the consent design that has to precede them.
Where the template system and sending infrastructure get built in the first place.
A worked example of the architecture-first approach on a real client engagement.
With authentication and with the list, in that order — not with the creative, which is where most teams start. Verify SPF, DKIM and DMARC on every sending domain including ones added for old campaigns, then look at who you are still sending to and how recently they engaged.
Continuing to mail long-dormant addresses is the most common self-inflicted cause, and it compounds: poor engagement signals worsen placement, which worsens engagement. Recovery is a trend over weeks rather than a fix, which is why the sequence matters more than the effort.
Far fewer than most estates carry. The right structure is a small number of layouts plus a library of reusable content blocks, rather than a template per campaign. A brand change should then be one edit rather than forty.
If your team cannot currently build a new email without copying an old one, that is the diagnostic. The copy-an-old-one habit is what produced the forty templates, and it will produce forty more.
Dynamic content blocks handle rule-based variation without code and cover most requirements. AMPscript earns its place when the rule depends on a value calculated at send time, on data outside the sending data extension, or when you need formatting and fallback logic the content editor cannot express.
The useful test: if a marketing administrator could not read the block and say roughly what it does, ask whether the requirement really needs it. Code written for convenience is a maintenance cost with no matching benefit.
It is worth enabling where you have the engagement history to support it. Einstein Send Time Optimization uses machine learning over roughly 90 days of email or push engagement data to determine the best time within the next 24 hours to send to each contact — which is a genuinely useful thing to get for free with the license.
The prerequisite is the point: it needs history. A newly migrated org, or one that has just warmed new sending infrastructure, does not have it yet. Enabling it and expecting immediate results is the common disappointment, and it is a timing problem rather than a product one.
With a repeatable checklist rather than a careful person. The list above is the one we use: audience volume against expectation, personalization against edge-case records, link checking including dynamically constructed links, rendering across the clients your audience actually uses, suppression, send classification and reply handling.
The two most valuable items are the least intuitive. Previewing with an edge-case record rather than a tidy one catches most personalization failures, and checking audience volume catches the expensive mistakes.
Directionally, at most. Mail clients pre-fetch images on the recipient's behalf, so an open can be recorded without a person reading anything — which means the metric moves for reasons unrelated to the campaign.
We use delivery and placement as the health metrics, clicks and downstream action as engagement, and goal completion as the journey measure. More on that on the reporting and analytics page.
Tell us whether the problem is that the team cannot execute or that the mail is not arriving. Those are different engagements, and the second one gets worse while it waits.
Serving: US | Canada | UK | UAE | Australia | New Zealand
