outbound-framework

Outbound framework

This skill turns a brief into cold emails through a fixed chain of tables. Every stage outputs a table before any copy gets written. The tables are the thinking: they force one industry, one title, one pain, one proof set and one free asset per email.

That's what makes an email relevant, and relevance is the only criterion. A cold email gets a positive reply when the reader thinks it was written for their job, in their industry, this quarter. Copy quality matters less than people think. The CTA matters more than anything.

A full worked example (hypothetical company, 10 ICPs, every table filled) lives in references/worked-example.md. Read it when unsure about the expected depth or format.

---

Table rules (apply to every table you output)

1. State rows and columns. Put one line above each table: Rows: ... Columns: .... The reader should never guess what a row is.
2. Fill every cell. If a value is unknown, write MISSING. Never guess a name, a number or a client. Every MISSING becomes a question for the user.
3. Use PASS / FAIL words in check columns, not symbols. They survive every renderer and copy-paste into sheets.
4. Tables first, prose second. Prose is one or two connecting lines between tables, never a paragraph restating a table.
5. Don't skip stages the routing table says to run. A skipped table is where irrelevant emails come from.
6. Emails go in fenced text blocks so they paste into Instantly or Smartlead without formatting damage.
7. No em dashes anywhere, in tables or copy. Use commas, periods or colons.

---

Stage 0: Route the request

Rows: request types. Columns: which stages to run, and what the final output is.

Request	Stages to run	Final output
Full campaign from a brief	1, 2 (if a list is involved), 3, 4, 5, 6, 7, 8	ICP matrix, Variant A + B per ICP, sequence, QA table
One email for one ICP	1, 3 (single row), 4, 5, 6, 8	Variant A + B
Subject lines only	1, 3, 5	Subject check table with winners marked
Follow-ups for an existing opener	1, 5 (new-thread subjects), 7, 8	Sequence table + full copy
Review or fix an existing email	8 on their draft, then 5 and 6 to rewrite	Audit table + rewritten email
List building or segmentation	2, 3	Filter spec table + ICP matrix
Weekly stats review	9	Decision table + new hooks table

---

Stage 1: Brief intake

Rows: the inputs every campaign needs. Columns: the input, what the brief says, status, and why it matters.

Input	Value from brief	Status	Why it matters
Product in one sentence		Have / MISSING	Feeds the How it works
How it works, 3 to 4 steps		Have / MISSING	The default way to explain the product
Sender name and title		Have / MISSING	Self-intro line
Named customers, by industry		Have / MISSING	Authority and subject name-drops
Exact numbers (volumes, results, ratings, counts)		Have / MISSING	Authority and subject numbers
Backers, certifications, partners		Have / MISSING	Extra authority
Free assets the client can actually deliver		Have / MISSING	The CTA, the most important element
Target industries and titles (if given)		Have / MISSING	Seeds the ICP matrix
Language and register (e.g. Italian Lei)		Have / MISSING	Tone and structure
Size split rules (if any)		Have / MISSING	Only case where a meeting CTA is allowed on touch 1

If named customers, exact numbers or deliverable assets are MISSING, ask for them in one message before writing copy. An email without real names and numbers, or without an asset the client can deliver, fails the two elements that drive replies.

---

Stage 2: Data reality check (Apollo, Lusha, Hunter)

Treat all three as big LinkedIn scrapers with an email guesser bolted on. Their data comes largely from LinkedIn profiles and the public web, plus contributor networks. The who-does-what is there and it's good enough to segment on. The email is a guess: the tool applies the company's naming pattern (first.last@, f.last@, first@) and pings the mail server. On catch-all domains the server accepts everything, so the check proves nothing.

Rows: the fields these tools return. Columns: where the field really comes from, how far to trust it, what to do.

Field	Real source	Trust	Action
Name, title, company	LinkedIn profiles	High, lags job changes	Segment on it. Check LinkedIn for top accounts before sending
Seniority, department	Derived from title text	Medium	Filter on title keywords too
Industry	Company's LinkedIn category	Medium, broad buckets	Combine with keywords, spot-check 20 rows
Headcount	LinkedIn employee count	Medium	Use ranges
Email	Pattern guess + server ping, or public pages	Low until verified	Send only to verified. Catch-alls in small batches
Phone	Contributor networks	Variable	Not used in email outbound

When building a list, output a filter spec. Rows: filters. Columns: filter, value, why, trust note.

Filter	Value	Why	Trust note

---

Stage 3: ICP matrix and relevance test

One ICP is one industry plus one title, and it gets one campaign and its own email. Two titles in the same company never share a row, because they don't share a pain. For a full campaign aim for 8 to 12 ICPs unless the brief sets a number: enough to learn fast, few enough to review weekly.

Rows: one ICP each. Columns: the fields that make the email relevant.

#	Industry	Title	Pain point (internal)	Peer proof names	Weighted How it works step	Free asset	Campaign name

Column rules:

• Pain point is something this title owns and gets measured on. It's internal: it shapes the email, it never appears as "you struggle with X".
• Peer proof names come from the same industry as the row. A hotel group wants to see hotel groups.
• Weighted step is the How it works step that answers this row's pain.
• Free asset is what the CTA offers this specific ICP (see Stage 6).
• Campaign name format: [client]-[industry]-[title]-[hook]-[yyww], e.g. NWAP-3PL-HeadFin-Overbilling-2639.

Then run the relevance test. Rows: one ICP each. Columns: the five layer tests, then the verdict.

#	Peer names recognizable to this title	Pain owned by this title	Asset answers the pain	Proof from same industry	Vocabulary specific to this industry	Verdict

Any FAIL: fix the ICP row before moving on. Don't carry a failing row into copy.

---

Stage 4: Proof inventory

Generic authority reads as no authority. Every claim carries a recognizable name or an exact figure, ideally both. Prefer non-round numbers where the real figure is non-round: 2,066% reads as true because it is.

Rows: one proof point each. Columns: what it is, where it fits, and whether it can be used.

Proof point	Type	Fits which ICPs	Exact figure	Source	Status

Type is one of: named customer, volume, result, rating, backer, partner. Status is one of: Verified, Needs client confirmation, MISSING.

Banned versus required. Rows: claim types. Columns: the vague version, the required version.

Claim type	Banned	Required
Customer base	"trusted by thousands"	"10,000+ Amazon and Shopify sellers"
Rating	"top-rated app"	"rated 4.9 on the Xero App Store"
Named customers	"used by leading studios"	"[named studio], [named studio] and 4,000+ architects in 135 countries"
Volume	"processing significant volumes"	"over €500M a year on [regulator] infrastructure, with [bank] and [bank]"
Results	"great results for clients"	"[named brand] moved 2,066% more units off a single feature"
Backers	"backed by top investors"	"backed by [named fund]"

Never invent a name or a number for a real campaign. If an ICP needs proof that doesn't exist, mark it MISSING and ask. If the user explicitly wants a draft anyway, insert a visible slot such as [PROOF: named 3PL customer + monthly invoice volume]. Never a made-up name.

---

Stage 5: Subject lines

Every subject is an impersonal BuzzFeed-style trade headline about the prospect's peers. Two elements are mandatory on top of the style: a number and a recognizable name. The number makes it concrete. The name makes it relevant, because it's a company the prospect benchmarks against.

Rows: subject patterns. Columns: the pattern and an example.

Pattern	Example
How [Name] and [N] other [peers] [behavior]	How GXO and 37 other 3PLs catch carrier overbilling before it gets paid
Why [N] [peers] like [Name] stopped [old behavior]	Why 22 auto parts makers like Brembo stopped matching partial deliveries by hand
How [Name] went from [X] to [Y]	How Philips and 6 shared service centers pushed touchless invoices from 38% to 81%
Why [Name] does [X] instead of [Y]	Why DHL Supply Chain audits 100% of fuel surcharges instead of a 5% sample
[N] things [Name]'s [team] stopped doing [when]	4 things Kuehne+Nagel's finance team stopped doing at month-end
Come [peers] [behavior] (Italian)	Come i concessionari auto spediscono in tutta Italia senza rischio sul pagamento

For each ICP, write 3 candidates and run the check. Rows: candidate subjects. Columns: the five mandatory checks, then the verdict.

Candidate	Number	Recognizable peer name	Peer group	No "you" / "your"	No target, sender, product or offer	Verdict

Pick one candidate with five PASS. Rules behind the checks:

• The peer name is never the prospect's own employer and never the sender. At send time, if the prospect works at the name-dropped company, swap the name.
• The subject never states the offer or the benefit as a pitch. It reads like industry news, not marketing.
• Name and number in the subject must match the authority line. "GXO and 37 other 3PLs" in the subject means the body adds up to 38.
• Specific beats clever: name the concrete behavior, not "working smarter".
• Only deviate from this style if the user explicitly asks for a different format.

---

Stage 6: Email build

The four elements, always in this order

Rows: the four elements. Columns: job and rules.

Element	Job	Rules
1. Kairos (opener)	Why this person, why now	"I'm contacting you as you [run/lead X] at {{companyName}}..." or "Reaching out as I saw you're [title] at {{companyName}}...". Always first. No greeting, no congrats, no flattery
2. Authority	Why believe it	"I'm [Sender], [Title] at [Company]." then 1 to 2 sentences of names from their industry and exact numbers
3. How it works	What it is and why it's better	3 to 4 steps, then one value line with a number, strongest word last
4. CTA	The reply	"Would you mind if I share..." plus a free asset built for this ICP

Never open with the pitch and never open with authority. In campaigns sent from a company alias with a full signature (e.g. formal Italian Lei), the self-intro can fold into the authority line. The authority itself is never optional.

How it works is the default explanation

In most situations, How it works step by step is the best way to explain a product. A prospect can picture four steps inside their own workflow. They can't picture "AI-powered automation". A mechanism they can follow reads as true, a claim reads as marketing.

• 3 to 4 steps, one action each, under 12 words, in their vocabulary (rate card, goods receipt, cost center, site sign-off).
• Describe their workflow with the product, not the product's features.
• Put the weight on the step from the ICP matrix that answers this ICP's pain.
• Close with one value line carrying a number.
• In Variant A, compress to one inline flow: "you set a CPC budget, publishers pitch you story ideas, you approve, they publish, you pay only for the clicks".

The CTA decides the campaign

Syntax: "Would you mind if I share..." / "Would you mind if I sent you..." / Italian: "Le dispiace se le mando...". A yes must unlock a mini action that creates real value for this prospect. The closer to their own material, the better.

Rows: asset types, strongest first. Columns: what the prospect gets, when it fits.

Asset	What they get	When it fits
Audit of their data	Their real data run through the product, with the result	Highest intent. Needs them to share a sample
Sample from their material	Output built from what's public: their site, their project	Zero effort for them, strongest first impression
Benchmark	How their peer group performs, in numbers	Senior titles who manage by KPI
Template or case	The document peers use, ready to adapt	Ops roles and follow-ups

• Never on touch 1: a call, a calendar link, "15 minutes", a free trial of an expensive service, a physical sample.
• Banned on any touch: "Worth a quick chat?", "Worth a call to discuss?", "Got 20 mins this week?", "Open to a call?", and any "[value summary] + worth a call?" construction.
• Sanctioned exception: when the user defines a size split, the over-threshold segment may get a meeting CTA paired with a free trial account. Only when the user sets it up.

Build table, then render

For each ICP, fill the build table first. Rows: the parts of the email. Columns: the content, and the rule check.

Part	Content	Rule check
Kairos opener		PASS / FAIL
Self-intro + authority		PASS / FAIL
How it works steps		PASS / FAIL
Value line		PASS / FAIL
CTA		PASS / FAIL
Sign-off	Best, [Sender]	PASS / FAIL

Then render two variants:

• Variant A, super short: 5 to 10 rows, under 80 words, inline How it works.
• Variant B, mid size: 10 to 20 rows, 100 to 150 words, numbered How it works.

Variant B skeleton:

Subject: [Stage 5 winner]

Hi {{firstName}},

I'm contacting you as you [run/lead X] at {{companyName}}, [one clause of context from their world].

I'm [Sender], [Title] at [Company]. [Named peers from their industry] and [N others] [do X at exact volume] with us.

How it works:

1. [Step in their vocabulary]
2. [Step, weighted for this ICP's pain]
3. [Step]
4. [Step]

[Value line with an exact number, strongest word last].

Would you mind if I [share the free asset built for this ICP]?

Best,
[Sender]

Tone rules

• Decisive, professional, friendly. A human operator, not a marketing bot.
• Sentences under 20 words. Contractions. Lead with the point.
• Max two merge tags: {{firstName}}, {{companyName}}. Heavy templating reads fake.
• No greetings or congrats ("Hope you're well", "Congrats on the raise").
• Never lecture the prospect about their industry or describe their problem to them.
• Banned words: synergy, leverage, unlock, game-changer, revolutionary, delve, streamline, empower.
• No em dashes or double dashes. Sign off "Best,".
• If a voice-profile skill for the sender is available, apply it. If ai-tells is available, run a light pass.

---

Stage 7: Follow-up sequence

The opener is followed by 5 follow-ups. They alternate: same subject as a reply in the thread, then a new subject in a new thread, then the same subject again. Same-subject touches keep the offer alive. New-subject touches are a second first impression for someone who ignored the first: new subject, new named proof, new asset.

Rows: the 6 touches. Columns: send day, subject, thread, job, length, CTA.

Touch	Day	Subject	Thread	Job	Length	CTA
Opener	1	A	New thread A	Full email: Kairos, authority, How it works, CTA	100 to 150 words	Asset 1
FU1	3	Re: A	Reply in A	Bump with one new detail on asset 1	Under 50	Asset 1
FU2	7	B	New thread B	New angle, new named proof, short How it works	60 to 100	Asset 2
FU3	10	Re: B	Reply in B	One number from asset 2	Under 40	Asset 2
FU4	15	C	New thread C	Listicle or case. First touch where a call may appear, as an alternative	60 to 100	Asset 3 or a 15-minute walkthrough on their data
FU5	21	Re: C	Reply in C	Close the loop, ask for the right person, keep asset 1 open	Under 50	Referral + asset 1

• Subjects B and C pass the full Stage 5 check: number, peer name, peer group, no "you", no target, sender or pitch.
• Never "just checking in", "bumping this" or "did you see my email". Every bump adds one new fact.
• A call never appears before touch 3, and always as an alternative to the asset, never instead of it.
• Stop on any reply. Out-of-office with a return date: pause and resume after it.

Output the sequence table filled for the ICP, then the full copy of all 6 touches in text blocks.

---

Stage 8: Pre-send QA

Run on every email, every variant, every follow-up. Rows: the checks. Columns: result, and the fix if it failed. Deliver only emails where every row is PASS.

Check	Result	Fix if FAIL
Opener says why this person: what they do or run
2+ recognizable names from their industry in the authority line
At least 1 exact figure, non-round where the real one is
How it works in 3 to 4 steps, in their vocabulary
One value line with a number
CTA is "Would you mind if I..." + a free asset built for this ICP
No call, calendar link or "15 minutes" on touch 1
Subject has a number and a recognizable peer name
Subject never names the target, the sender or the offer, never says "you"
Subject name and number match the authority line
Max 2 merge tags
No em dashes, no buzzwords, no greeting or congrats
Under 80 words (A) or 150 words (B)
Every name and number is Verified in the Stage 4 inventory

When reviewing someone else's draft, run this table on their draft first, then rewrite through Stages 5 and 6.

---

Stage 9: Weekly review

Every campaign is an experiment. Measure on delivered emails, never opens: Apple Mail Privacy Protection preloads images, so open rates don't mean much.

Rows: one campaign each. Columns: the ICP, the metrics, and the decision.

Campaign	Industry	Title	Subject hook	CTA asset	Delivered	Bounce	Reply	Positive	Signal	Action

Decision rules. Rows: signals. Columns: starting threshold (calibrate after 4 weeks of real data), usual meaning, action.

Signal	Starting threshold	Usually means	Action
Bounces	Above 2%	Guessed emails, catch-all domains	Pause. Re-verify, drop or batch catch-alls
Few replies of any kind	Under 1.5% after 300 delivered	Subject or opener not landing	New subject with a different peer name and number. Keep the body
Replies, few positive	Under 1% positive after 300 delivered	Relevance holds, offer is weak	Change the CTA asset. Move it closer to their data
"Not me" replies	More than a third of replies	Wrong title for this pain	Move the title filter one level up or down
"Send more info" replies	Recurring	Asset too vague	Make the asset personal: their data, their site
Strong positives	Above 3% positive	ICP and hook both work	Clone the hook to the adjacent industry, same title

Change one variable at a time, otherwise nothing is learned. Close every review with a new hooks table. Rows: ICPs getting a new hook. Columns: new peer name, new number, new asset, what it replaces.

ICP	New peer name	New number	New asset	Replaces

---

Output order for a full campaign

1. Stage 1 intake table (and questions, if anything is MISSING)
2. Stage 3 ICP matrix, then the relevance test
3. Stage 4 proof inventory
4. Per ICP: Stage 5 subject check, Stage 6 build table, Variant A and Variant B
5. Stage 7 sequence table and full copy (for each ICP, or for the first ICP as the pattern when the user asks for a sample)
6. Stage 8 QA table