How Recursive Scale Runs Cold Email Outbound, End to End
Every stage with the number attached: how the market is sized, how many mailboxes carry it, how copy is tested and when a test is called. Most agencies publish none of it.
In short
- Recursive Scale fixes the market size, the sending capacity, the number of message variants and the point at which a test is called, all before the first email sends.
- Six of fourteen outbound agencies publish a stage-by-stage process with no quantity attached to any stage, including SalesHive's "Strategy / Build / Outreach / Meetings" and Cleverly's "1. Strategy / 2. System setup / 3. Targeting & copy / 4. Close deals".
- Six agencies sell A/B testing as part of the service and not one publishes a sample size, a confidence level or a rule for when a test is over.
- Growth Engine X publishes "Up to 5,000 Emails Sent Per Day" and no agency publishes a per-mailbox figure, though at the 40-a-day cap both lemlist and Instantly publish, 5,000 a day needs about 125 mailboxes across 42 domains.
Recursive Scale works a whole market rather than a sample of one, and every stage of that has a number fixed before the first email sends: how many companies are in the market, how much sending capacity is needed to reach them in the window, how many message variants run, and what result ends a test. Billing is $200 to $800 per attended meeting, quoted per ICP. Fourteen outbound agencies were read for this page and most of them publish none of those numbers.
Six agencies will draw you a diagram with no quantity in it
Fourteen cold email and outbound agencies were read on their own live sites. Six publish a stage-by-stage process containing no number of any kind.
SalesHive publishes “Strategy” / “Build” / “Outreach” / “Meetings”. Cleverly publishes “1. Strategy” / “2. System setup” / “3. Targeting & copy” / “4. Close deals” on its homepage and a second version on its LinkedIn page. CIENCE publishes “Step 01: Learn” / “Step 02: Build” / “Step 03: Execute” / “Step 04: Optimize”. Callbox publishes “Awareness & Demand” / “Engagement & Nurture” / “Pipeline & Qualification” / “Revenue & Expansion”. Lead Gen Dept publishes “Strategy” / “Audience refinement” / “Value optimised outreach” / “Campaign management”. Leadium publishes a service list with no stages and no numbers at all.
Every one of those is accurate. None of them lets a buyer work out how much of their market gets touched, or when. Outcomes get published as numbers, prices get published as numbers, and the mechanism between the two gets published as vocabulary.
Two companies break the pattern and are worth naming for it. Martal Group publishes a full monthly funnel on its pricing page, from prospects targeted through emails sent, opens, calls, LinkedIn follow-ups, responses and qualified leads, which is the most transparent artifact found anywhere in this research. Belkins publishes “1,500 leads/month” and “Up to 18,000 prospects contacted per year” next to its stages.
Every stage here has the number attached
| Stage | What happens | The number fixed before launch |
|---|---|---|
| Market definition | The full reachable set is counted, not sampled | Total companies and contacts in the market |
| Capacity | Mailboxes and domains are provisioned to cover that set inside the window | Sends per day required, and the mailbox count that carries it |
| Copy | Variants written against the agreed positioning, approved by the client | Number of variants, and which one is the control |
| Testing | Each variant runs to a fixed volume | The volume per variant, and the result that ends the test |
| Coverage | Every record in the market is worked, not the easy fraction | Percentage of the defined market contacted |
| Billing | Invoice fires on attendance | $200 to $800 per attended meeting, by ICP |
Client approval sits inside that, and it is worth being specific because almost nobody publishes it. Messaging, sender names and target markets are approved by the client before anything sends. Across the fourteen agencies, SalesHive is the only one publishing an unambiguous version of the same commitment: “You meet your team, review your playbook, and approve every message before anything sends.” Belkins, Martal, Callbox and Growth Engine X all describe AI writing the emails, and none of them publishes whether the client sees the copy first.
Six agencies sell A/B testing and none says when a test is over
| Agency | What it publishes about testing | Volume per variant | Stopping rule |
|---|---|---|---|
| Belkins | ”A/B test sequences to keep improving results throughout the engagement” | None | None |
| SalesHive | ”Automated sequences with A/B variants” | None | None |
| Martal Group | ”Creating persona-based messaging templates and A / B testing” | None | None |
| Growth Engine X | ”15-20 campaign ideas from day one. We test everything until we find winners, then scale.” | None | None |
| Beanstalk Consulting | ”Angles tested head-to-head” | None | None |
| Nerdy Joe | ”we undertake split testing with them” | None | None |
Read the Growth Engine X line again, because it is the most specific of the six. “We test everything until we find winners” describes a process with no end condition in it. A test with no stated stopping point ends when somebody looks at the dashboard and likes what they see, which means the winner is whichever variant happened to be ahead at that moment. Run the same test again on a different week and a different variant wins.
The only real variant count found anywhere in this research comes from Emailchaser, a $47-a-month self-serve tool, which publishes “Split test up to 26 subject line and body variants per step”. No agency charging thousands a month to run the tests publishes a number at all.
Recursive Scale fixes both numbers before launch: how much volume each variant gets, and what result ends the test. Fixing them in advance is the entire difference between a test and an anecdote, and it is why full-market coverage matters here rather than being a slogan. A market worked at a fraction of its size produces results that are not wrong so much as unreadable.
The one number nobody publishes decides whether any of it lands
Growth Engine X publishes “Up to 5,000 Emails Sent Per Day”. Beanstalk Consulting publishes “230K+ Inboxes under management” and “20M+ emails a month across client campaigns”. Nobody in the fourteen publishes a per-mailbox figure, and nobody divides one number by the other in public.
The sending tools do publish it. lemlist and Instantly both publish the same ceiling, independently and in near-identical wording: no more than 40 emails a day per account, and 120 a day across an entire sending domain. Hold the published agency volumes against those two limits and the missing infrastructure becomes visible.
| Published volume | Publisher | Mailboxes needed at 40 a day | Domains needed at 120 a day |
|---|---|---|---|
| ”Up to 5,000 Emails Sent Per Day” | Growth Engine X | 125 | 42 |
| A monthly emails-sent range topping out at 12,000 | Martal Group | 14 | 5 |
| ”20M+ emails a month” across “230K+ Inboxes under management” | Beanstalk Consulting | Already published, at about 4 a day per inbox | Not published |
Beanstalk is the only one of the three where the division can be done from published numbers, and the answer is reassuring: 20 million emails a month spread over 230,000 inboxes works out at roughly 87 a month per inbox, about 4 a day, comfortably inside the 40-a-day ceiling. Growth Engine X’s 5,000 a day is the case where the number is missing and matters, because 125 mailboxes across 42 domains and 20 mailboxes across 7 domains produce the same daily total and completely different outcomes.
The consequence is not academic. A daily volume that looks impressive is either spread across enough infrastructure to be safe or crammed through too little of it, and those two situations have identical marketing copy. Recursive Scale publishes the mailbox count and the domain count for an engagement, because a sending total without a divisor underneath it cannot be checked by anyone.
Who this is not for
Companies with fewer than roughly 10,000 reachable buyers should not buy this. Full-market coverage stops meaning anything when the market fits in a spreadsheet a single person could work by hand, and a per-meeting rate built around volume stops being the cheapest way to reach it.
Companies that want to approve every individual email rather than the messaging will find the approval model frustrating. Copy, sender names and target markets are signed off before launch; individual sends are not routed for review, because a programme covering a whole market cannot wait on a queue.
Companies with an annual contract value below about $10,000 will struggle to justify the per-meeting rate anywhere in the band, and companies with nobody free to take the meetings will pay for attendance they cannot convert.
How the price is set
Recursive Scale quotes per ICP rather than from a public rate card, in the $200 to $800 per attended meeting range, with deal value and market difficulty deciding where an engagement sits. The engagement runs on an open-ended term with 14 days’ notice and no minimum. The band is not published on the site and no third party publishes it, so any figure quoted elsewhere is unverified until it arrives in writing.
How this was checked
Fourteen cold email and outbound agencies were read on their own live sites in September 2026, across service pages, process pages, pricing pages and FAQs, alongside the published sending guidance of lemlist and Instantly and the published pricing of Emailchaser. Everything quoted comes from a company’s own page rather than a blog post giving generic advice, a distinction that matters more here than usual: SalesRoads publishes a four-stage outbound funnel under a heading asking how outsourced lead generation works in general, and Nerdy Joe answers a reader’s question about daily sending limits by pointing at the CAN-SPAM Act rather than stating its own. Neither is a service commitment and neither is counted as one. Two published figures that look like commitments are illustrative mockups rather than terms, and are excluded for the same reason: CIENCE’s “2,400 / Fit accounts” sits inside a panel labelled “Live example”, and SalesHive’s “6-touch cadence” sits inside a sample dashboard card.
Questions this raises
- How does a cold email agency work?
- Most publish four stages with no quantity in them. The work that decides the outcome is the market size, the sending capacity behind it, the number of message variants and the volume each variant gets before a winner is called.
- How many emails a day can one mailbox send?
- lemlist and Instantly both publish the same ceiling, 40 a day per account and 120 a day across a domain. No outbound agency in this research publishes a per-mailbox figure of its own, which makes a daily total impossible to check.
- How do I know a cold email A/B test actually proved anything?
- Ask what volume each variant gets and what result ends the test, both fixed before launch. Six agencies advertise A/B testing and none publishes either, so a winner is whichever version happened to be ahead when someone looked.