Key takeaway
An email campaign is one message, or an ordered set of them, sent to a defined audience against a single goal. Three types, differing in one thing, which is what starts the send: broadcast, triggered, transactional. The build is six steps, and choosing a platform isn't step one. What decides the outcome is the send policy: authentication, consent, and a complaint rate under 0.30%, three in a thousand. I've watched good campaigns die on that line while nobody questioned the copy.
What an email campaign actually is.
An email campaign is one message, or an ordered set of messages, sent to a defined audience against a single goal, with a start, a rule for who receives it, and a measurable end. That's correct, and it could have come from anywhere. The part worth having is this: a campaign is a data object with a send policy attached, not a screen you sit in front of. The audience is a query, the message is a template, the schedule is a trigger, and the send policy is the set of rules the mailbox providers hold you to. Nearly every guide on this topic describes the screen instead, which is why they all open on "choose a platform" and stop at "test before sending", roughly the point at which a real campaign starts to be hard.
The definition leaves two things out, and both matter before anything else does. It says nothing about who's allowed to receive the mail, and nothing about what happens when a fraction of the people who do receive it press the spam button instead of the unsubscribe link. Both of those decide whether the campaign existed at all, and neither of them is a thing you fix in the builder.
Broadcast, triggered, and transactional.
The type lists on this topic run long: newsletter, promotion, announcement, welcome, abandoned cart, win-back, re-engagement, and on down the page. Those are use cases wearing the word "type". Sort them by the one thing that changes how a send behaves, which is what starts it, and three classes are left. Everything on those lists sits inside one of them.
- Broadcast: one send, one moment, one audience, started by a person or by a schedule. Newsletters, promotions, and announcements are all this shape, and segmenting the list is the whole of the targeting.
- Triggered: started by something a contact does rather than by a date in a calendar. Welcome series, abandoned cart, win-back, and renewal reminders are this, and the campaign is a branching automation rather than one message.
- Transactional: started by a single recipient's own action, sent to that person alone, and expected within seconds. Receipts, password resets, and order confirmations.
The split holds because it's about the trigger and nothing else. Nitrosend sends all three, and builds flows with triggers, steps, and branches. What it doesn't tell you is consent, and the third class stands on a different footing there from the first two, which the next section picks up.
What a campaign owes before it sends.
Gmail treats anyone sending more than 5,000 messages a day to Gmail accounts as a bulk sender, and since 1 February 2024 has required SPF and DKIM on the domain, DMARC on the sending domain with a policy of p=none being enough, a visible unsubscribe link in the body, and one-click unsubscribe on marketing mail, per Google's sender guidelines. Yahoo asks for the same email authentication and attaches an operational clock to the rest of it: unsubscribes honoured within two days, in its own sender best practices.
The number that actually caps a program is the complaint rate. Both providers put the threshold at 0.30%, and Yahoo calculates it against inbox-delivered mail rather than everything you handed over. That isn't a best practice with a nice-to-have attached to it. It's a limit with enforcement behind it, and it's the only campaign metric where crossing the line changes what happens to the next send rather than what happened to this one. One-click unsubscribe is the cheap side of the same trade, and it's a published standard rather than a design preference: RFC 8058 defines a List-Unsubscribe-Post header and an HTTPS POST that runs with no confirmation step, on a message carrying a valid DKIM signature. A reader who takes it has left quietly. A reader who can't find it presses the other button.
Underneath all of that sits what the sending industry considers a decent list in the first place, and it's a higher bar than any inbox rule. The standing position is verifiable, informed opt-in, with confirmed opt-in as the highest standard, per the M3AAWG sender practices. Clear the provider thresholds and you arrive. Clear this one and you keep arriving, because the complaint rate is downstream of whether the person remembers asking.
Machine learning only works if you feed it the right inputs. I described our own approach on SaaS District. We pull in "customer data, product data and the on-page stuff as well". The system processes all of this data to return recommended products. The goal is entirely mercenary. We continuously adjust the models to improve revenue per email.
Building one, in six steps.
Six steps, and choosing a platform isn't one of them. A platform is something you buy once and then use for every campaign after this one, so putting it at step one is how a build sequence turns into a product tour. Each of these produces something the next step needs, which is why the order matters more than the count.
- Name the goal, and the number that would prove it. One goal per campaign. If you can't say which number should move, you're sending because it's Tuesday.
- Define the audience as a query, not a list export. A segment that recomputes is still true next week. A CSV pulled on Monday is a photograph of a list that has changed since.
- Write the message against a single action. One call to action, and the subject line written last, once you know what the email actually says.
- Authenticate the sending domain, then check the suppression list. This is the step the guides leave out, and it's the one that decides whether anything above it gets read at all.
- Test on real clients, then send to a small slice of the list first. Rendering breaks one client at a time, and a first slice tells you what a full send would have cost you.
- Read the complaint and bounce rates before the open rate. Hold the next send if either moved, whatever the opens did.
Six steps is a first campaign. The second one is the same six run as a repeatable job, and that's where the difference between a platform and a screen starts to show.
A campaign is an object, not a screen.
Look at what those six steps consist of. A goal is a number in a document. An audience is a query. A message is a versioned template. A send is a request with an idempotency key on it. A schedule is a trigger. None of that inherently requires a person clicking, and yet every product in this category requires one, because the products were drawn as screens first and the data model arrived afterwards.
Nitrosend is MCP-first, which means the endpoint and the tool come before the screen, so a campaign lives where the rest of the work already does. BYO sending keys are available on Pro and above across Amazon SES, Resend, Postmark, Mailgun, and SendGrid, so the transport can stay exactly where it is while the platform around it changes.
I've spent a decade building email platforms, and I've watched this category answer the agent question by fitting an assistant onto a dashboard. What you get is a chat window bolted onto a builder: it'll write you a subject line and it can't run your send. That's a motor on a bicycle. The limit is worth saying flat: an agent doesn't decide whether the offer is worth sending, and it can't make an unwanted list wanted. It removes the clicking, which was never the work.
What to measure, and in what order.
A campaign produces four kinds of number: delivery, engagement, conversion, and complaint. They're normally taught in that order, with the open rate leading the engagement half, and that ordering is a leftover. Apple Mail Privacy Protection pre-fetches images for Mail users, so the tracking pixel fires whether or not a human ever looked at the message. An open rate built on that moves for reasons that have nothing to do with a reader.
So read them the other way round. The complaint rate and the email bounce rate come first, because they answer whether you get to send again, and one of them has a published threshold with enforcement behind it. Clicks and conversions come next, because they answer the question you asked when you built the thing. The open rate comes last, as a direction rather than a result, useful for spotting a subject line that died and useless as a target.
None of these numbers means much on a single send. A campaign is judged against your own previous sends, on the same list, from the same domain, which is the only comparison that holds every variable except the one you changed. Email marketing metrics published as a benchmark tell you what somebody else's audience did with somebody else's mail, and there's no version of that number that's a target for yours.
The only metric that pays wages is revenue. I made this point about our own dashboard on In the Ring with SUMO Heavy. We put the performance split front and center. A store might see that it makes "18% of your revenue from email". Tracking that baseline changes behavior. You start testing new ideas to push the figure higher. Growing that share ultimately lifts total store revenue by improving retention.
Where campaigns go wrong.
Four failures account for most of the campaigns that underperform, and all four get diagnosed as copy problems first. They aren't. Three are decisions somebody took before a word was written, and the fourth is a setting nobody checked.
- Sending to the whole list because segmenting felt like extra work. The unengaged half doesn't just fail to convert, it moves the one rate that governs the next send.
- Hiding the unsubscribe link to stop people leaving. An unsubscribe is the cheapest thing a bored reader can do to you, and the spam button is the alternative they reach for instead.
- Three calls to action in one campaign so that whatever it measures means nothing. If two goals both matter, that's two campaigns.
- Launching on an unauthenticated domain and then rewriting the subject line. The mail wasn't ignored, it never arrived.
The first three cost you performance and the fourth costs you email deliverability, and the campaign report names none of them as the cause. I watched the first one compound fastest: an unsegmented send is the quickest way to push a list past the 0.5% complaint rate we used to auto-flag accounts at, against an industry average closer to 0.02%.
A campaign is an object: a goal, an audience query, a message, a send policy, and four numbers that come back afterwards. Running one shouldn't mean an afternoon inside a builder, and none of those pieces needs a screen to exist. Nitrosend gives each of them an API endpoint and an MCP tool, so an agent can draft, segment, test, and send. Start on the free tier, 8,000 emails and then 500 a month, unlimited contacts, no card.
Go deeper
Sources
- Google, Email sender guidelines: the 5,000-a-day bulk threshold in force since 1 February 2024, the SPF, DKIM, and DMARC requirement, one-click unsubscribe, and the 0.30% spam rate.
- Yahoo Sender Hub, best practices: Yahoo's authentication and list-unsubscribe requirements, unsubscribes honoured within two days, and the spam rate measured against inbox-delivered mail.
- RFC 8058: Signaling One-Click Functionality for List Email Headers, which defines the
List-Unsubscribe-Postheader, the HTTPS POST with no confirmation step, and the DKIM requirement. - M3AAWG Sender Best Common Practices: verifiable, clear, and informed opt-in as the practice, with confirmed opt-in named the highest standard.
Common questions
Email marketing is the whole program: the list, the strategy, the calendar, and everything you send across a year. A campaign is one unit inside it, a single message or an ordered set of messages aimed at one audience against one goal, with a start and a measurable end. You run one email marketing program and dozens of campaigns inside it.
One, if it's a broadcast. A newsletter, a promotion, or an announcement is a single send to a segment, and padding it into a series doesn't make it work harder. A triggered campaign is a sequence, and it should run only while the trigger stays unresolved: a welcome or abandoned-cart flow stops the moment the contact does the thing you asked for. Every extra message needs a reason to exist beyond filling a slot.
Often enough that people remember subscribing, and not so often that they start reporting you. There's no universal cadence, because it depends on what you sell and what somebody opted into. The signal to watch isn't the calendar, it's the complaint and unsubscribe rates across consecutive sends. If those climb while clicks fall, you're sending too often for that particular list, whatever the frequency advice says.
It's a less useful number than it looks. Open tracking fires when an image loads, and Apple Mail Privacy Protection pre-fetches images for Mail users, so a share of any open rate was never a person reading anything. Treat it as a direction, judge the campaign on clicks, conversions, and the complaint rate, and use your own previous send to the same list as the benchmark rather than a published average.
An honest subject line and sender name, a real business identity in the message, a one-click unsubscribe that works on the first click, and suppression the send path checks rather than a spreadsheet somebody processes at month end. Permission sits upstream of all four: the standing position of the sending industry is verifiable, informed opt-in, and nothing downstream repairs a list that never had it.
The platform bill is usually the smallest part of it. Platforms price either on contacts stored or on emails sent, and which model you're on matters more than the headline rate: a list much larger than your monthly volume costs far more under contact pricing. Nitrosend meters sends rather than contacts, which are unlimited on every plan including Free. The larger costs are the hours spent building the campaign and the reputation damage from sending it to people who didn't want it.
Pick the number before you send, then read it against your own previous send to the same list. Delivery tells you it arrived, clicks and conversions tell you the message did its job, and the complaint and bounce rates tell you whether you get to send again. A campaign that beat its conversion target while pushing complaints up hasn't worked, it's borrowed from the next send.
Yes. Every part of a campaign is either data or a call: the audience is a query, the message is a template, the send is an API request, and the schedule is a trigger. Nitrosend is MCP-first, so every capability is an API endpoint and an MCP tool before it's a screen, and an agent can draft, segment, test, and send one without anybody clicking through a builder. You still decide whether the offer is worth sending.