What Cold Email Costs, Itemised from Published Rates

Here is the trap. You compare sequencers, notice that Mailshake starts at $29 and lemlist starts at $69, and conclude that Mailshake is the cheap one. Read the send caps on the same two pages and Mailshake's entry plan allows 1,500 emails a month while lemlist's allows 50,000. Per email actually sent, the cheap one costs 14 times the expensive one.

At 1,000 sends a month the all-in floor is $126.44, and at 50,000 it is $2,941.55.

The sequencer is 30.8% of the first number and 3.2% of the second. Contact data is 39.6% and 85.0%. Every figure on this page is computed from rates read off vendor pages on 25 August 2026 and every one of those rates is linked.

Nothing here is a measurement. No invoice was read, no campaign was run, and no reply rate is claimed. This is published-rate arithmetic at stated volumes, on stated assumptions, and where an assumption changes the answer the page says so and gives you the lever.

What follows is the itemised floor, the same stack costed at three volumes, all twelve sequencers converted to a cost per send, the verification line at three list sizes, what Google and Microsoft actually require in their own words, whether an assistant writing your emails changes any of it, a procedure for measuring your own deliverability before you scale, and what breaks in month three.

What does cold email cost, in one number?

Between six and thirteen cents an email, and the range is set almost entirely by what you pay for the contact rather than by what you pay to send to them.

At 1,000 sends a month the all-in cost is $0.1264 an email. At 50,000 it falls to $0.0588. It falls because the fixed lines get spread, not because sending gets cheaper.

Strip the data out and the picture changes completely. Software plus mailboxes plus domains plus verification is $0.0763 an email at 1,000 sends and $0.0088 at 50,000, a fall of 88%. The sending stack has enormous economies of scale.

The list feeding it has almost none, because every contact is bought once at roughly the same unit price. That is the whole argument of this page in two sentences.

Teams spend weeks choosing between a $39 tool and a $47 tool, and the difference between them is worth $8 a month against a bill that is mostly somewhere else. The decision that moves the number is how many contacts you buy and how many times you email each one.

What are you actually buying when you buy a sequencer?

Four separate products, sold by four separate kinds of company, and only one of them is the thing with the price you quoted to your founder. Each of the other three has its own unit, its own billing basis and its own scaling curve, and the curves point in different directions, which is the reason the sum surprises people who priced only the first one.

• The sequencer. Software that holds a campaign, decides who gets step two on which day, stops when someone replies, and reports. Smartlead, Instantly and lemlist are this. It is the cheapest layer at every volume above about 3,000 sends.
• The identities. Mailboxes and the domains they live on. Sold per mailbox per month by Mailforge, Maildoso, Primeforge and, at a higher unit price, by Google and Microsoft. This line grows in a straight line with your sending volume because volume is spread across identities on purpose.
• The list. Contact records with a work email on them, sold per record by LeadMagic, Apollo, Clay and a dozen others. This is the largest line on the bill at every volume above roughly 500 sends a month, and it is the one nobody puts in a cold email cost roundup.
• The hygiene. Verification before you send, warmup before and during, and placement testing while you run. NeverBounce and Clearout bill per address checked. MailReach bills per mailbox warmed, which means it scales the same way the identity line does.

The reason nobody publishes the sum is structural rather than sinister. Each vendor prices its own layer accurately, and the addition is nobody's job. Woodpecker is the exception worth naming: its pricing page carries a calculator with mailboxes at $4 or $6, domains at about $1 a month, dedicated servers at $59 and Azure domain hosting at $20, which is more of the real stack than any competitor prints.

The itemised floor at 6,000 sends a month

Smartlead's Base plan is $39 a month on monthly billing and caps sending at 6,000 emails against 2,000 active leads. Mailforge sells mailbox slots at $3 each on monthly billing, in a minimum block of ten, and .com domains at $14 a year. Both pages were read on 25 August 2026.

A correction to our own number

This site published $73.67 before, describing the mailbox line as $3 each billed yearly.

The total was right and the label was wrong. Mailforge's calculator shows 25 slots at $75 a month billed monthly and $60 billed yearly, so $3 is the monthly-billed unit and $2.40 is the yearly one. Ten slots on annual billing is $24, not $30.

Paying everything annually takes the same stack to $61.17 a month: Smartlead Base at $32.50, ten slots at $24, four domains at $4.67. That is 17.0% off, and it is the discount you forfeit by keeping the flexibility to stop. Whether that trade is worth it depends entirely on whether you expect to still be sending in month four.

Ten mailboxes is not a preference. Mailforge states plainly that it sells a minimum of ten slots and that a slot is charged whether a mailbox exists in it or not: "If you purchase 10 slots but only create 5 mailboxes, you are still charged for all 10 slots." That is the floor for anyone using this provider, at any volume, including zero.

Why the mailbox line sets the bill and the software line does not

Because a mailbox has a daily ceiling and software does not. Mailforge publishes its own guidance on the pricing page: "We recommend sending 30 emails per mailbox per day (max 100) and using 2-3 mailboxes per domain for the best deliverability results." Those two sentences determine the entire shape of the bill. At 30 sends a day across 22 working days, one mailbox carries 660 emails a month. So mailbox count is your monthly volume divided by 660, and domain count is mailbox count divided by 2.5. Both scale linearly and forever. The sequencer, meanwhile, moves in three or four steps across its whole ladder and then stops mattering.

Smartlead's own ladder shows it. Base is $39 for 6,000 sends, Pro is $94 for 90,000, Unlimited Smart is $174 for 150,000 and Unlimited Prime is $379 for 500,000. Sending goes up 83 times between the first and last rung while the price goes up 9.7 times.

The identity line has no such curve.

At 6,000 sends a month you need 10 mailboxes and the software is 53% of the sending stack. At 50,000 you need 76 mailboxes and 31 domains, the identity line is $264.17 a month, and the software is still $94.

Software share of the sending stack falls from 53% to 21% while your volume goes up 8.3 times.

All-in at 1,000, 10,000 and 50,000 sends a month

Same stack, three volumes, every rate published and dated. The model assumes a three-step sequence, so contacts are sends divided by three, mailboxes are sends divided by 660, domains are mailboxes divided by 2.5, and the cheapest Smartlead plan whose cap clears the volume. Data is priced at LeadMagic's $0.150, the floor of the seven enrichment vendors this site has costed.

The sending stack per email falls 88% between the first column and the third. The all-in figure falls 53%. That gap is the data line refusing to get cheaper, and it is the single most useful thing in this table if you are forecasting a year of outbound rather than a month of it.

Swap Smartlead Pro for lemlist's Email plan at $69, which publishes a 50,000-send allowance, and the 10,000-send column falls from $676.94 to $651.94 while the 50,000 column falls from $2,941.55 to $2,916.55. A 27% saving on the software line moves the total bill by 0.8%. That is the ratio worth internalising before your next vendor comparison.

The band on the data line is wider than the entire rest of the stack. At 50,000 sends, 16,667 contacts cost $2,500.05 at LeadMagic's rate and $13,750.27 at BetterContact's. That 5.5-times spread, computed on this site from seven vendors' own pricing pages, is worth more attention than every sequencer feature comparison ever written.

Twelve sequencers, converted to a cost per sent email

Every price below was read off the vendor's own pricing page on 25 August 2026 in the monthly-billing toggle state, in USD, except where the row says otherwise. The fourth column is the thing no ranking page computes, and it is where the ordering falls apart.

Mailshake has the lowest sticker in the category and the highest cost per send by a factor of 14 against lemlist.

Its entry plan also grants exactly one email address, and 1,500 sends from one mailbox across 22 working days is 68 a day, more than double Mailforge's recommended 30 and two thirds of the 100 it names as the ceiling.

Instantly's pricing page states the Hypergrowth plan includes 125,000 emails monthly on the plan card, and 100,000 in the feature comparison table further down the same page, and 100,000 again on the annual version of the same card. Read on 25 August 2026. Three statements, two answers, one page, and no note explaining which governs.

Saleshandy's page opens with a currency selector offering USD, EUR, GBP, AUD, CAD, INR and BRL, and defaults to the annual price with the monthly price in smaller type underneath. Starter reads $34 a month billed annually at $408, or $41 if paid monthly. Our own record said $36. The correct monthly-billing figure is $41, and it is now what this table shows.

MailReach moved too. This site had it at $25; the page today reads $19.50 per mailbox per month with a worked example of nine mailboxes at $175 a month, and 20% off on annual. That is a fall of 22% since 7 August 2026. Prices in this category move inside three weeks, which is the argument for reading them rather than citing them.

Warmup is either free or the largest line you have

Seven of the twelve sequencers bundle warmup at the entry tier, which makes it feel like a solved problem. It is not solved, it is cross-subsidised, and the moment you buy it standalone the pricing shape becomes obvious: MailReach charges $19.50 per mailbox per month, which is 6.5 times what Mailforge charges to supply the mailbox in the first place.

At 50,000 sends a month, standalone warmup at $1,482 is larger than the entire rest of the sending stack at $441.50.

It would take the all-in bill from $2,941.55 to $4,423.55, a rise of 50.4%. Whether your sequencer bundles warmup is therefore a more consequential question than its headline price, and almost nobody asks it that way.

Verification is the one line that bills by outcome

Every other line on this bill charges you for capacity. Verification charges you per address checked, which means it tracks how aggressively you prospect rather than how much you send, and it is the only line where the vendor's definition of a billable event is genuinely contested.

Clearout is the only one of the three specialists that publishes what happens to an undecidable address. Its documented rule charges a credit for a result of "valid or invalid or catch-all" and leaves unknown free. NeverBounce publishes no equivalent rule, and separately excludes accept-all, unknown, invalid and disposable results from its 3% bounce guarantee.

That matters more than the per-unit price, because catch-all domains are a large and unpredictable share of any B2B list, and a vendor that neither tells you the rule nor covers the result under its guarantee has moved the whole risk onto you. On our numbers verification is 2.1% of the bill at 1,000 sends and 2.8% at 50,000, so this is a governance question rather than a budget one.

One live error is worth flagging if you are pricing from NeverBounce's own help centre rather than its pricing page. Its billing article prices five verifications at $1.20, which is $0.24 each and 30 times its own $0.008 list rate, in the same article that correctly prices 5,000 credits at $400. Read the rate card, not the worked example.

What the mailbox providers actually require, in their own words

This is the section most cold email guides get from other cold email guides. The primary documents are short, public and specific, and they say less than the folklore around them does. Google's bulk sender page and Microsoft's Outlook.com postmaster announcement were both read on 25 August 2026.

Google, from 1 February 2024

Google's threshold for the enhanced requirements is "5,000 or more messages per day" to Gmail accounts. Above it you need SPF and DKIM both, DMARC with a policy that may be set to p=none, valid forward and reverse DNS, a TLS connection, RFC 5322 message format with a valid Message-ID, and one-click unsubscribe implemented to RFC 8058 and RFC 2369.

The spam-rate sentence is the one people misquote. Google's page says to "keep spam rates reported in Postmaster Tools below 0.10% and avoid ever reaching a spam rate of 0.30% or higher." Those are two different numbers doing two different jobs: 0.10% is the target you run at, 0.30% is the line you never cross.

Microsoft, from 5 May 2025

The Outlook.com postmaster site carries the announcement verbatim: "Starting May 5th, 2025, Outlook.com is enforcing stricter email authentication standards (SPF, DKIM, DMARC) for domains sending over 5,000 emails per day." The requirement is SPF pass, DKIM pass, and DMARC at "at least p=none and align with either SPF or DKIM (preferably both)."

Microsoft's own blog post, by Puneeth on 2 April 2025 with an update dated 29 April, contradicts itself about the consequence inside a single section. It states the decision "to reject messages that don't pass the required authentication" with the error 550; 5.7.515 Access denied, and two paragraphs later says non-compliant mail goes to Junk with rejection coming at a "date to be announced". Microsoft publishes no complaint-rate threshold at all. Its list is authentication plus four hygiene recommendations: a valid reply-capable From address, functional unsubscribe links, list hygiene and bounce management, and transparent mailing practices. No percentage appears anywhere in it.

Does AI-written email get filtered for being AI-written?

No. Read the two requirement lists above again and notice what is absent from both. There is no content-authorship criterion, no AI-detection step, and no published mechanism by which a filter would establish who wrote a ninety-word business email. Everything either provider requires is either a DNS record or a behavioural rate. That does not mean nothing changed when you started using an assistant. Something almost certainly did, and being precise about what matters because the remedy is different. The causal chain runs through volume, not through prose, and every link in it is something a provider measures.

An assistant removes the writing constraint, and the writing constraint was accidentally also the list constraint. When an email cost ten minutes nobody sent to a segment they had not thought about. When it costs ten seconds the list gets longer before it gets better, a longer list produces more complaints, and complaint rate is a published threshold with a number attached.

There is a duller possibility worth ruling out first, and it is the one experienced senders reach for. On Hacker News thread 39888383 (350 points, 227 comments, 31 March 2024), user snowwrestler put it flatly: "SPF, DKIM, and DMARC are important to set up correctly, but doing so is NOT sufficient for good deliverability. In fact it is only a small component of success."

A small industry sells software that rewrites model output to evade AI detection. AI detection is real in academic and publishing contexts, where the question is whether a person wrote a long document. Nothing in any mailbox provider's published requirements suggests detection runs on short business email, and short text is where such classifiers are least reliable anyway.

If a humanising step improves your results, the plausible explanation is that it made the email less generic and therefore less likely to be reported. That is the same mechanism as writing it better, available at no extra subscription, and the next section is how to get it out of the prompt rather than out of a tool.

What the writing actually costs, and what it buys

Recipients are the filter that matters, and in 2026 they are unusually well calibrated. On thread 49103089 (302 points, 126 comments, 29 July 2026) user areoform described reading a dozen inbound cold emails a day and finding that "every other single email uses The Sentence. Or, The Paragraph. Whatever model these people use, it always always puts in The Sentence."

The framing to take from that is not that AI writing is forbidden. It is that a shared model plus a shared prompt produces a shared output, and the person reading it sees the shared part several thousand times a year. areoform's own instruction was "Please use models to research people and learn about them. But please write the email yourself." The research step is where the leverage sits, and it is the step almost every published prompt library skips in favour of optimising the drafting. The drafting is the thing these models are best at and the least valuable part of the job.

The third item is the one to keep. A list of what your prospect does not care about removes the two sentences you were going to waste, which matters under a hard word ceiling. The final instruction matters as much: without it you get confident inference presented as research, and a confidently wrong sentence about someone's business ends the conversation faster than a boring one.

The constraint block does more work than the request

• No em dashes. They are the clearest single tell of unedited model output in 2026. This site bans them in its own editorial rules for exactly that reason, and the check is automated.
• No "not just X, it's Y" constructions, and no sentence that opens by naming a problem the recipient has not told you they have.
• No adjectives in front of your own product. Powerful, effortless and best-in-class all mean the sender could not name a specific benefit and hoped an intensifier would cover it.
• No claim the research does not support. This prevents the most damage, because the recipient knows their own business better than your enrichment vendor does.
• Name the register. British or American English, and one real person's writing to imitate if you have one. Otherwise you get the average of everything, which is the definition of generic.
• Hard word ceilings. Under 90 words for a first touch, 60 for a follow-up, 40 for a breakup. Length is where generic material accumulates, so a ceiling forces the model to spend its budget on the specific observation.

Now cost it. Real research pasted into a real prompt is about five minutes a prospect even with an assistant doing the reading, so fifty a week is achievable and five hundred is not. At a fully loaded SDR cost of $134,000 to $154,000 a year, five minutes is roughly $5.40 to $6.20 of labour, against $0.15 for the contact and $0.0123 to send to them.

That is the actual trade in cold email, and it is not visible anywhere in the software comparison.

Human attention costs 36 to 41 times the contact record and 440 to 500 times the send. Volume is cheap and consideration is expensive, which is precisely why the market keeps buying more volume.

Measure your own deliverability before you scale

Everything above is arithmetic on published rates. This section is the part you run on your own account, and it takes about ninety minutes spread over two weeks. Do it before you buy the second block of mailboxes, not after.

Step one, Google Postmaster Tools, 20 minutes plus DNS propagation

Add every sending domain at postmaster.google.com, copy the TXT or CNAME record it gives you into that domain's DNS, and wait. Google's own help page states it "verifies domains right away, but it can take up to 10 minutes before your domain's verification status is updated." Do this per domain, which at 31 domains is a real afternoon.

Two limits to know before you trust the output. The data "only applies to messages sent to personal Gmail accounts" ending @gmail.com or @googlemail.com, so a B2B list of corporate Workspace addresses will populate almost nothing. And Google suppresses low-volume days outright: "Data might be missing if the total number of messages for a given day is too low."

Step two, read the spam rate against two thresholds, 5 minutes a week

Pass is under 0.10%. Warning is 0.10% to 0.29%. Fail is 0.30% or above, which is Google's own "avoid ever reaching" line. On a 10,000-send month, 0.30% is 30 complaints, which is three people out of every thousand pressing a button. That is a smaller number than most senders picture when they hear a percentage.

A failure here means stop sending, not send better. Complaint rate is computed on recent mail, so the only lever that moves it fast is volume down. Cutting the list to the segment you can justify person by person is the intervention that works; rewriting the copy while continuing to send at the same volume is not.

Step three, check alignment on a real header, 15 minutes

Send one email from each sending domain to a Gmail address you control, open Show Original, and read the Authentication-Results line. You want spf=pass, dkim=pass and dmarc=pass, and you want the domain in the DKIM signature to match the domain in the From header. Passing SPF while misaligned still fails DMARC.

This is the check people skip because they configured the records and assume configuration equals result. On thread 43468995 (429 points, 253 comments, 25 March 2025) user jeroenhd described an Outlook server that "checked the DKIM signature on the message it REWROTE ON ITS OWN and decided that the signature didn't match, and rejected my email." One person's account, and a failure mode no amount of correct DNS prevents.

Step four, seed test across providers, about $1 a test

A seed test sends your live campaign copy to a set of addresses across Gmail, Outlook.com, Yahoo and corporate Microsoft 365, then reports which folder each one landed in. MailReach's standalone spam tester prices 200 credits at $210 a month, which is $1.05 a test, and its All-In-One plan bundles 20 credits. Pass is inbox placement above 90% on Gmail and above 80% on Microsoft, which is the harder of the two by consensus and by every practitioner account in the threads cited here. A Microsoft-only failure with a clean Gmail result points at authentication or reputation rather than content, because content would fail both.

What breaks in month three

The stack above is priced as though it runs unchanged. It does not. Five things go wrong on a fairly predictable schedule, four of them cost money that appears nowhere in the table, and the fifth costs something you cannot buy back at any price. None of the five is a surprise to anyone who has run outbound for a year, and all five are absent from every vendor's pricing page.

Domains burn, and the replacement is not free

A domain that accumulates complaints stops delivering, and the standard remedy is to retire it and register another. At $14 a year that reads cheap. The real cost is that the replacement starts with no reputation, which puts it back into a warmup cycle, and Saleshandy prices the consequence explicitly: its Annual Flexi tier offers "free mailbox replacement (requires new domain, $14/yr)" at $3.49 a mailbox against $2.99 for the fixed tier.

That 16.7% premium is the market price of expecting to burn domains. It is the only place in this category where a vendor has put a number on the practice, and it tells you the practice is common enough to price a product around.

The warmup period nobody budgets for

Amazon's SES documentation is the clearest published statement of how long a new sending identity takes: "For some email providers, you can establish a positive reputation in around two weeks, while for others it may take up to six weeks." Its automatic warmup "steadily increases over 45 days, independently from your sending volume."

Forty-five days at full mailbox cost before full sending capacity is available.

On the 50,000-send stack that is $264.17 a month of identities and $94 of software running at partial output for six weeks. Nobody puts that in a first-quarter forecast, and it is the single most common reason a cold email budget overruns in month two.

The maintenance requirement is worse and almost never quoted. SES states that after warmup "you should send around 1,000 emails every day to each email provider that you want to maintain a positive reputation with." That is a floor on volume imposed by reputation itself, and it points in the opposite direction to every piece of advice about sending less.

Crossing a complaint threshold, and how long the way back takes

Nobody publishes a recovery time. What is published is the mechanism, and it is asymmetric: reputation is built by sustained volume over weeks and lost by a bad week. On thread 43468995, user csomar described the bootstrap problem precisely: "My head was spinning as to how that reputation will be built if your email just goes straight to spam."

User bityard, in the same thread, named the reason new domains get treated harshly: "Buying a new domain, setting up the SPF, DKIM, and MARC, and then immediately spamming from it until it's banned everywhere a week later is standard spammer MO." That is one practitioner's characterisation, and it describes the exact behaviour a cold email stack is built to perform.

Shared infrastructure moves without telling you

Every mailbox provider at $3 a slot is putting you on shared sending infrastructure, and its status page is the only honest record of what that means. SendGrid's incident feed lists 50 incidents between 22 January and 4 August 2026, one every 3.9 days, and 25 of the 50 are named for a delay, a block or a delivery failure.

Two of those name a provider directly: "Increased Microsoft Inbox Provider Blocks" on 22 January 2026 and "Microsoft Feedback Loop (Spam Reports) Processing Delays" on 29 June 2026. Mailgun's feed independently logs "Email Delivery Failures to Microsoft 365 Domains" on 17 August 2026. Two unrelated platforms, the same counterparty, inside seven months. None of that is visible from your sequencer's dashboard, which will report the message as sent. If your reply rate falls off a cliff on a specific day, check the infrastructure provider's status history for that date before you rewrite anything.

Switching vendors resets the part you cannot buy

Moving sequencer is easy, because campaigns are data and data exports. Moving mailboxes is not, because the reputation lives on the domain and the sending path, and a new path starts cold. The asset outbound builds is warmed identities, and it is the only line on this bill you cannot repurchase at any price. Which is the argument for registering your own domains rather than letting a managed vendor register them for you, even where the managed price is lower. Settle who owns the domain before the first invoice, because that question gets very expensive to ask in month nine.

The number that actually decides this

Contacts per month, not sends per month and certainly not which sequencer.

Contacts drive the data line, which is 39.6% of the bill at the bottom of the range and 85.0% at the top, and they drive the verification line too. Sends follow from contacts by whatever step count your sequence uses.

So the order of decisions runs backwards from how the category sells itself. Decide how many people you can genuinely justify contacting this month. Multiply by your step count to get sends. Divide sends by 660 to get mailboxes, mailboxes by 2.5 to get domains. Only then pick the sequencer whose published cap clears your number, which takes five minutes.

Two thresholds are worth memorising. Below about 3,000 sends a month the ten-mailbox minimum dominates and the stack is structurally inefficient, so a manual motion at fifty a week is cheaper and works better. Above about 30,000 the identity line passes the software line and keeps pulling away. And if a vendor will not publish a send cap, treat the price as incomparable rather than as competitive. Three of the twelve tools in the table above cannot be converted to a cost per send from their own published pages, and a price you cannot convert to a unit is a number rather than an offer.

What this page does not know

Three things, and the third is the one that would change the answer.

Google Workspace mailbox pricing is absent on purpose. workspace.google.com/pricing serves visitors in India prices in rupees with no US business tier visible, and converting a foreign-currency figure into a dollar price would be a guess wearing a currency symbol. Plenty of teams run cold email on Workspace mailboxes, and this page cannot tell you what that costs.

Reply.io, Klenty and Meetz publish no send cap that our fetches could read, so three of twelve rows in the comparison table carry "Not computable" rather than an inferred number. Klenty also offers no monthly billing at all: the shortest term you can buy is quarterly at $60 a month, against $50 a month on annual. And the deliverability rate is entirely absent from every figure here. Cost per email sent is knowable from rate cards. Cost per email delivered to an inbox is not, because inbox placement is the one number no mailbox provider and no sending vendor publishes for cold traffic.

So what does cold email cost?

$126.44 a month at 1,000 sends and $2,941.55 at 50,000, on published rates read on 25 August 2026, with contact data at the cheapest of seven vendors.

Per email that is $0.1264 falling to $0.0588. Take the data out and it is $0.0763 falling to $0.0088, which is the number the category advertises against.

The useful reframing is that you are not buying software with some setup attached. You are buying a list, and everything else is the cost of delivering to it. That is why the $39 and $47 tools sit so close together and why the choice between them is worth less than one afternoon of list work.

Three things to do with this. Convert every sequencer price to a cost per send before comparing, because the number on a pricing page is frequently not an offer. Check whether warmup is bundled, because standalone it is 50.4% on top at volume. And read the catch-all rule before buying verification, because only one vendor tells you whether an undecidable result costs a credit.

Then budget the identities and the list first, and treat the software as the rounding error it becomes above 10,000 sends a month. If you are not there yet, the manual motion at fifty a week costs less and reads better, and the arithmetic on this page is the reason why.