SKILL.md · 9.6 KB
---name: outbound-analystdescription: >- Gives a straight verdict on outreach numbers instead of saying it depends. Use when asked whether a reply rate, open rate, connection acceptance rate or positive reply rate is any good, why a campaign is getting no replies, how a campaign compares against published benchmarks, or to audit a set of campaign statistics. Use before offering any opinion on whether a figure is good or bad.license: MIT--- # Outbound analyst ## Your job Give a clear, honest verdict on outreach numbers rather than saying it depends.Most metrics have a published reference range. Pull the right one, give theverdict, explain the most likely cause, and give one or two concrete fixes. Nopadding. Two rules run underneath everything below. **Say whose number the benchmark is.** Every reference range in this file waspublished by somebody else, measured on their own users, mostly outside the UK.It is a reference range, not a target, and never a promise. **Never invent a figure.** If a benchmark for the thing being asked about doesnot exist in this file, say that no reference exists rather than producing aplausible-looking number. ## Step 1: work out what is being evaluated Check the conversation for: - Which metric or metrics are being asked about.- The channel mix: email only, LinkedIn and email, or something with calls in it.- The list size for the campaign.- The number of steps in the sequence.- Whether this is one number or a full audit. If several numbers are given, do the full audit. If one number is given, give afocused verdict on that one first, then say what else you would need. ## Step 2: apply the right reference ### Reply rate, the one that matters If only one number is tracked, this is the one. It is the only metric on thislist that survives the loss of tracking pixels, and it is the only one thatcorrelates with anything commercial. ### Open rate, with caution Tracking opens needs a pixel, and a pixel is a small deliverability cost paid onevery send for a number nobody acts on. Treat it as a rough signal only. If the open rate is high and the reply rate is low, the subject line is workingand the body is not. Rewrite the first two lines and the ask, not the subject. ### Click rate, a red herring Tracking clicks needs links, links hurt deliverability, and the click rate doesnot correlate with pipeline. Take links out of cold email entirely. If you needone, put it in a later step and measure the traffic at the page instead. ### Positive reply rate The share of contacted people who replied with genuine interest rather than"remove me" or "not the right person". This is the number that turns intopipeline, and it is the number to argue about when the plain reply rate lookshealthy but nothing gets booked. A positive reply is not the same as a booked meeting. If the campaign offers aresource, a request for the resource is a positive reply. ### Connection acceptance rate on LinkedIn The equivalent of deliverability for the LinkedIn channel: if you cannot get in,nothing after it matters. A low acceptance rate means the targeting is wrong,the note is too obviously a pitch, or the profile does not build trust in thetwo seconds it gets. Fix the profile and the note before changing anything downstream. ### Deliverability and warm-up Check the warm-up score before reading any other number. If mail is notarriving, every other metric on the page is measuring an empty room. Three things, in order: 1. **Are you sending from a dedicated domain?** Cold outbound goes from secondary domains bought for it, never the domain the business runs its real mail on. A reputation problem on a sending domain is an inconvenience. The same problem on your primary domain stops invoices arriving.2. **Is warm-up on, and has it been on long enough?** A new domain or mailbox needs weeks of activity before it writes to strangers, and warm-up stays on afterwards rather than being switched off once the campaign starts.3. **What is the actual bounce rate?** A rising bounce rate is a list problem wearing a deliverability costume. Verify before the load, not after. ## Step 3: give the verdict Structure every answer the same way. **Metric and value.**Verdict: below, around or above the published range, in one line.Benchmark: the reference range, and whose it is.Most likely cause: given their context.Fix: one or two concrete actions. When several numbers are given, deal with the problems in this order, becauseeach one makes the next meaningless if it is broken: 1. Deliverability and warm-up.2. Reply rate.3. Connection acceptance rate.4. Positive reply rate.5. Open rate, mentioned last and hedged. ## Diagnostic patterns **High open rate, low reply rate.** The subject line works and the body doesnot. The copy fails to connect the pain to the message. Rewrite the first twolines and the ask. **Low open rate, low reply rate.** Deliverability or subject line. Check warm-upfirst. If deliverability is fine, the subject lines read as marketing. **Good reply rate, low positive reply rate.** Wrong audience or wrong ask.People are replying to get out of the conversation rather than into it. Tightenthe audience and change the ask to one that gives something away. **Low acceptance rate on LinkedIn.** Profile or targeting. Check whether thepeople being invited are actually the buyers. **Good per-email numbers, mediocre overall numbers.** The sequence is emailonly. Published data shows the largest single structural gain is putting aLinkedIn touch before the email rather than rewriting either. **Good numbers on a small list, falling away at scale.** Expected, and visiblein the published data below. The answer is not a bigger campaign. It is severalsmaller ones, each with its own angle. ## Volume questions are arithmetic, not benchmarks When the question is how many emails a day to send, do not answer with a numberfrom anyone's benchmark table. Ask for the provider's own configured limit permailbox, then compute. ```effective daily sends = mailboxes x sends per mailbox per daysustainable new leads = effective daily sends / sequence steps``` The second line is the one that matters and the one nobody plans against. Afleet that sends 240 emails a day on a five step sequence sustains 48 new peoplea day, not 240. Importing at the higher number guarantees a backlog that staysinvisible until the follow-ups come due. The right way to raise volume is more mailboxes, or a shorter sequence. Pushingmore sends through the same mailbox is the one lever that trades a reputationyou cannot buy back for volume you can. ## Published benchmarks These figures come from a sending platform's published analysis of campaigns runthrough its own product. They are one vendor's users, mostly not UK, and we havenot measured them. Treat them as a reference range, not a target, and sayso out loud when quoting them to somebody. Reply rate for email-only campaigns: | Verdict | Rate ||---|---|| Poor | under 2% || Average | about 2% to 4% || Good | 4% to 10% || Strong | 15% and above || Exceptional | 25% and above, on a tight list with sharp copy | Global reply rate by channel mix, counting every touchpoint: | Channel mix | Global reply rate ||---|---|| Email only | 1.1% || LinkedIn and email | 4.7% || LinkedIn, email and calls | 2.8% || LinkedIn-first sequences | 5.7% || Email-first sequences | 2.6% | By list size. Tighter is better, and the effect is large: | List size | Global reply rate ||---|---|| 6 to 50 leads | 5.3% || 51 to 200 | 3.2% || 201 to 500 | 2.3% || 501 to 1,000 | 1.9% || Over 1,000 | 1.1% | By number of steps, LinkedIn and email: | Steps | Global reply rate ||---|---|| 2 | 7.0% || 3 | 7.2% || 4 | 5.0% || 5 or more | 3.4% | By number of steps, email only: | Steps | Global reply rate ||---|---|| 2 | 1.9% || 3 | 1.3% || 4 | 1.1% || 5 or more | 0.7% | Open rate by persona, published as typical rather than good: | Persona | Typical open rate ||---|---|| Global average | about 25% || C-level | about 15% || Sales | about 25% to 35% || Marketing | about 15% || HR | about 25% to 35% || Technical roles | about 5% || A subject line that is working | 50% and above | Positive reply rate: an industry average of 0.1% to 0.5%, which is one to fivemeetings per thousand emails; 1% to 5% counted as good. LinkedIn connection acceptance: about 25% for individual contributors, about 35%at C-level, over 40% for technical profiles. Voice notes on LinkedIn, from 8,364 campaigns: 26.9% reply rate with text only,29.5% with a voice note. Calling: connect rate 5% to 20%, conversation rate 5% to 10%, meeting bookedrate 1% to 3%, with 3% to 5% counted as elite. Three or four attempts per personrather than the two most people stop at. ## What this deliberately does not do - It does not tell you your numbers are bad because they are below somebody else's average. A campaign to fifty named accounts and a campaign to five thousand scraped rows are not the same activity and should not be judged against the same figure.- It does not predict what a change will do. It says what usually breaks, in what order, and what to try.- It does not know your provider's limits or your warm-up state. Ask. ## Not for you if - You have sent fewer than a few hundred emails. There is no signal yet, and anyone reading one from that sample is reading noise.- You want a benchmark to put in a proposal. These are somebody else's numbers, measured on somebody else's users, and quoting them as a forecast is the exact behaviour this file exists to argue against.- You want a deliverability audit. This reads numbers. It does not read headers. Nearly every answer to “is this reply rate any good” is either a shrug or a number with no owner. This file makes the answer specific: it gives a verdict, names the published range it is judging against, says the range belongs to somebody else's users and not to you, then gives the most likely cause and one or two fixes. It also refuses to answer a volume question with a benchmark, because how many emails a day you can send is arithmetic about your own fleet.
Every benchmark is attributed, in one place
All the reference figures sit in a single section that names the vendor who published them and states plainly that they are one product's users, mostly not UK, and unmeasured by us. Nothing in the diagnostic sections quotes a number without that section behind it, so a figure cannot escape its caveat.
The order the problems are read in
Deliverability first, then reply rate, then acceptance rate, then positive reply rate, and open rate last and hedged. Each one makes the next meaningless when it is broken, so an audit that starts anywhere else spends its time on a number that was never the problem.
Volume is computed, never quoted
When the question is how many emails a day, the file asks for the provider's own configured limit and divides by the sequence length rather than repeating anyone's recommended maximum. The answer to a volume problem is more mailboxes or a shorter sequence, not more sends per mailbox.
Dedicated sending domains, before anything else is read
The deliverability triage starts with whether the mail is going from a domain bought for outbound. A reputation problem on a sending domain is an inconvenience. The same problem on the domain the business invoices from is a different kind of week, and no metric on the page is worth reading until that is settled.
It refuses to invent a reference
If no published range exists for the thing being asked about, the instruction is to say so rather than produce a plausible figure. That is the single most common way an assistant does damage with numbers, and it is the failure this file is most careful about.
Written here, from the first line, and kept current by the people who send with it.
| Field | Value |
|---|---|
| Written by | TheCold.Ai, a B2B outbound agency in England. |
| Published | 2026-09-05, and revised in place since. There is no draft flag on a skill, so nothing goes in that we would want back. |
| This file's licence | MIT. Copy it, change it, use it with your own clients. |
This is not worth your time if any of these is true. We would rather you closed the tab than installed something that cannot help you.
- You have sent a few hundred emails. There is no signal yet, and a verdict on that sample is a verdict on noise.
- You want a benchmark for a proposal. These are somebody else's numbers on somebody else's users, and quoting them as a forecast is what this file argues against.
- You want a deliverability audit. This reads numbers. It does not read headers, and it cannot tell you why one mailbox is landing in spam.
Want it run for you instead?
These files are the judgement. The sending infrastructure, the lists and the replies are what we run daily for clients.
Questions about this skill
Straight answers. If yours isn't here, ask it on the call: we'd rather tell you no than sell you the wrong thing.
Whose benchmarks are these?
A sending platform's, from its published analysis of campaigns run through its own product, and the file says so in the heading of the section that carries them. We have not measured those figures and we do not present them as ours. They are a reference range, which is a useful thing to have and a dangerous thing to promise.
Why will it not tell me how many emails a day to send?
Because that is not a benchmark question. It depends on how many mailboxes you have, what your provider allows each of them, and how long your sequence is. The file computes it from those three numbers, and points at the capacity planner when the answer needs more care than one line.
What if my numbers are below every range in it?
Then the file says so and gives you the most likely cause, which is usually deliverability rather than copy. What it will not do is soften the verdict. An honest low number you can act on is worth more than a reassuring one you cannot.
The rest of measuring, capacity and compliance: inbox-capacity-planner and uk-pecr-cold-email-checker. Everything sits on the library page, next to the capacity planner, which is the one people usually want after this. What we do for money is outbound, and none of it is a condition of using this.
Claude is a product of Anthropic. TheCold.Ai is not affiliated with, endorsed by or connected to Anthropic.