Pipeline Generation · AI Scale-ups · 2026
Pipeline runs on nine engines. Outbound is one.
Pick your engines, steal the plays, run them Monday. This portfolio is complex enterprise sales. Multi-threaded deals. C-level signature.
Contents
How the portfolio sits
The nine engines
10On CACwhich engines are cheapest, and when they payThe outbound engine, in depth
The full operating program for engine 03.
Elsewhere
Part One · The Portfolio
00 · The Portfolio
Nine ways to make pipeline. One portfolio.
Most teams run two engines, usually the two their last company ran, and call the rest experiments. The portfolio view is the correction. Each engine below has a real cost curve, a real ramp time, and a real failure mode, and the mix you staff is the strategy.
Two rules before the menu. First, pick engines the way you pick markets: by where your buyer already is. An enterprise security ICP gets broken into with manual outbound, ABM, and events; a developer ICP walks in through product, community, and search; nearly everyone runs automated outbound as the baseline layer. Second, fund on two clocks: the engines you can turn on this quarter pay this year, and the slow engines you instrument now are next year's CAC advantage.
Three paths. The nine engines sit on them.
Outbound produces volume and data. Content makes that outbound land warmer, because the name is already known. Referrals close the relationships the other two produced. One path caps you. Outbound without content lands cold. Content without outbound is waiting for the market to find you. Run all three. Weight outbound while you are still learning what the market wants.
The nine engines as one system. Read the portfolio card in the open source repo. Each engine section below links its own.
The portfolio, by time to turn on
Turn on this quarter
Staffed and instrumented in weeks. Pays this year.
Months to instrument
Cheapest at scale, in this order.
Cheaper at scale
Shading steps with CAC at scale. Every engine feeds the same pipeline number, though this model does not convert engine spend into meetings or bookings; section 12 sets the mix for the outbound pair.
Staff two or three engines you can run honestly this quarter, and instrument one slow engine for next year. An engine half-staffed is a cost center wearing a strategy's clothes.
Main article: module 03 of the Sales Playbook, the stack this engine runs on.
- ICP model + campaign strategy
- Account sourcing + stakeholder mapping
- Copywriting, then campaign execution
- CRM sync + reporting
How the best run it
Infrastructure first: secondary send domains, two to three mailboxes each, four to six weeks of warmup, and send caps that live outside anyone's enthusiasm. Data runs as a waterfall: Apollo sources, Clay cascades providers until the list bounces under 2 percent, which is the bar that matters rather than a verification score (AO-2), and AI research writes one specific opening line per contact from a real signal. A tight 300-account list beats a loose 3,000, because reply rate compounds into deliverability. One operator runs 400 to 500 sends a day; the rep only touches replies.
Advanced personalization replies at 17 to 18 percent; basic or none runs 7 to 9. Platform-wide average is 3.43 percent, and campaigns over 1,000 contacts average 2.1 against 5.8 for campaigns under 50. Depth is the lever, not volume (AO-1).
Every number in this engine carries a claim ID. Read the automated outbound card in the open source repo.
- Entry model (freemium, free trial, reverse trial)
- Signup experience, self-serve or sales-assisted
- Product usage sync + segmentation
- PQL scoring on seat thresholds and feature gates
- Viral loops (referrals, upsells, community)
How the best run it
The entry model is a packaging decision, not a default. Airtable runs a reverse trial: fourteen days of the paid tier, then a real free plan, which keeps the relationship alive after the trial ends. Cursor gated nothing and let power users hit usage limits; free-to-paid ran roughly ten times the freemium norm. The PQL definition is the whole game: seat velocity, domain consolidation, an SSO or SCIM attempt. The outreach that converts references exactly what the account did inside the product.
Median free-to-paid is 8 percent. PQL-to-enterprise at 10 to 25 percent is my operating assumption, not a benchmark: no published study supports that band. The closest published finding is that trials using PQLs convert 2.8x better than trials that do not (PLG-2).
Every number in this engine carries a claim ID. Read the product-led growth card in the open source repo.
PLG in this portfolio is usage that creates a champion, then a sales motion converts it. It is not the motion for a product that closes on a card with no committee. Snyk is the conversion pattern.
Deep dive: sections 11-21 below are this engine's full program.
- ICP model + campaign strategy
- Account tiering, T1 through T3
- AI account research
- Rep assignment across phone, LinkedIn, email, video
- CRM sync + reporting
How the best run it
The tier ladder in practice: a Dream-150 worked strictly by hand, Tier 1 on calls plus semi-automated sequences, Tier 2 on automated email and LinkedIn, Tier 3 on email only. Then the below-the-line pattern: parallel-dial the practitioners first, log what they say, and call the decision maker quoting the org's own words. Connect rates are a data problem before a skill problem: about 10 percent per dial, and about 25 percent per prospect once you count every attempt (MO-1). Three attempts per prospect is the measured average. If it takes eight to reach one person, that is a data-quality problem rather than a benchmark to plan around (MO-2).
Roughly 370 dials produce one booked meeting, about 0.27 percent dial-to-meeting. Of the conversations that actually connect, 4.6 percent book (MO-3).
Every number in this engine carries a claim ID. Read the manual outbound card in the open source repo.
- ICP + TAM mapping
- Account research + tiering
- Signal architecture, 1st through 3rd party
- Awareness scoring + lead routing
- Demand gen flywheel
How the best run it
Every account on the named list sits at exactly one awareness stage, each stage has entry signals, and the stage decides the next play. First-party signals (site visits resolved to companies and, where possible, people) say who is engaging with you; third-party intent says who is researching the category somewhere else. Layered, they surface accounts in evaluation before a form fill ever happens. Two rules keep it honest: suppress customers and open opportunities before anything routes, and never advance a stage on a single signal.
I plan on 25 to 40 percent of a named list engaging within 90 days, and that is operator judgment: no published ABM benchmark reports an account-engagement rate like it (ABM-2). Judge the flywheel on stage progression, not clicks.
Every number in this engine carries a claim ID. Read the ABM card in the open source repo.
- Strategy (community-led or partner-led)
- Community and partner build
- Value engine: exclusive access, certifications, networking, co-branded content
- Signal capture + routing
- Flywheel reinforcement
How the best run it
Community-led means practitioners get real value with no purchase anywhere in sight, and the pipeline signal is who shows up, asks, and answers. Partner-led means tiers, certifications, co-marketing funds, and marketplace listings: Wiz grew its integration network past 300 partners by running it inside product rather than sales, and dbt's marketplace transactions grew 190 percent in a year once cloud-marketplace buying unlocked stuck procurement. Either lane, the value engine has to pay members before it pays you. Capture is quiet: who attended, who asked, who brought a colleague, routed to a human only when the signal stacks.
This quarter it produces nothing. Run it for the year; channel clearing 25 percent of new business is the graduation bar.
Every number in this engine carries a claim ID. Read the community and partner card in the open source repo.
- ICP + target account list
- Ads strategy: funnel plan, creative, media buying
- Nurture sequences + remarketing
- Engagement tracking
- Conversion + CRM sync
How the best run it
Matched audiences of 300 to 1,500 named accounts; the 50-account list sales loves is algorithm sabotage. Creative runs full funnel: thought-leader ads from a real person at the top (reported at 4.65 percent click-through and $0.51 CPC against 0.68 percent and $2.42 for standard ads, though those figures come from vendors selling the format and no cost-per-qualified-lead comparison is published anywhere; PM-1), proof and comparisons to engagers in the middle, and the demo ask reserved for warm retargeting pools only. Demo offers at cold audiences are the worst cell in the matrix. The other half of the engine is speed to lead: a form that books the meeting on the spot converts several times better than one that waits a day in the CRM.
Paid rarely creates category demand. Run well against a named list it accelerates cycles 15 to 30 percent and carries 5 to 15 percent of pipeline.
Every number in this engine carries a claim ID. Read the paid media card in the open source repo.
- Keyword + topic research
- AI content drafting
- Structuring it for LLM answers as well as SERPs
- Ranking + traffic capture
- Lead capture + CRM sync
How the best run it
Vanta is the reference case: topic clusters organized around what buyers research, comparison and versus pages for the shortlist prompts, and dense internal linking so a model can trace the whole argument. Versus pages are the single strongest predictor of AI-search traffic, and sites with 21 plus comparison pages see roughly nine times the median AI referrals of sites with a handful. Structure for extraction: question-shaped headings, FAQ schema, answers in the first sentences. Then measure what analytics hides: most AI-referred traffic reports as Direct, so the "how did you hear about us" field is the attribution layer and a fixed prompt set run weekly is the rank tracker.
For a B2B SaaS audience AI-referred visitors convert at about 1.05x organic search, which is parity; no universal multiplier holds across industries (SEO-2). The catch is months of instrumentation before the flywheel pays.
Every number in this engine carries a claim ID. Read the SEO and AEO card in the open source repo.
- Event strategy, 3rd-party vs self-hosted
- Multi-channel promotion
- Event + lead capture
- Qualification + CRM sync
- 48-hour follow-up
How the best run it
Build the attendee list yourself from speakers, exhibitors, and LinkedIn signals, score it against ICP (RSA's 43,000 badges reduce to under 2,000 real targets), and sequence the top tier with angles specific to their session or stack: the generic "we will be at RSA" note replies at 2 percent, the specific one near 10. The best programs pre-book half their meetings before the floor opens, and the highest-trust conversations happen off the floor: a 12 to 20 seat executive-hosted dinner outperforms the booth, and the host's title decides who accepts. Then the SLA in the flow: hot leads inside 24 hours, everything inside 48, because the inbox moves on.
Booth-first programs run 8 to 12 thousand dollars per opportunity. ICP-first with pre-booking runs 2.5 to 5.
Every number in this engine carries a claim ID. Read the events card in the open source repo.
10 · On CAC
The cheapest pipeline is the slowest.
The verdict below is the portfolio's one ranking, and it cuts against instinct: the engines that cost the least at scale are the ones that pay nothing this quarter. That is not a reason to skip them. It is the reason to start them before you need them.
PLG is the cheapest at scale, then Community + Partner Led, then SEO and AEO. Those take months to instrument. The other 6 you can turn on this quarter.
Part Two
The outbound engine, in depth.
Arriving from engine 03? Start at prospecting does not pause, then do not manufacture the meeting, then read 11 through 21.
Engine 03 gets the deep dive because it is the engine a $1M to $20M company leans on hardest, and the one most teams run worst.
Outbound pipeline generation should be the lifeblood of your sales team, tracked and managed with the same rigor as revenue, CEO included. On top of an ARR quota, every quota-carrying rep and BDR carries a pipeline gen quota, because that is what forces the right volume of input into the top of the funnel. What follows is the whole program: who owns which accounts, what each seat is held to, the plan a rep fills in on Monday, the touches that earn a reply, and the tooling that runs it.
The continuous engine
Prospecting does not pause when the pipeline looks full.
Most sales training treats prospecting as the thing you do until the calendar is busy. That is how pipelines die. The meetings you book this month were earned by the outbound you ran last month. Skip a week now and the empty week shows up about thirty days later, which is exactly late enough that the skip feels free.
I run outbound as a continuous engine for that reason. A few rules I keep on the wall:
The thesis that matters for this page: prospecting is not the warm-up before the real work. It is the work that never stops. The moment you stop, you are running down a countdown that started when the last deal closed.
That is why Engine 01 and Engine 03 sit in the portfolio as baseline layers, and why Part Two below is a full operating program, not a tip sheet.
Principles drawn from Jeb Blount, Fanatical Prospecting.
Prospecting is not the warm-up before the real work. It is the work that never stops.
Before the test
Do not manufacture the meeting.
Most teams are not failing at copy. They are booking meetings with accounts that cannot buy this quarter, then calling the ghosting a late-stage problem.
The 95
95
percent, out of market
The logo you want and none of the urgency. Firmographic ICP is not temporal readiness. Until something inside the company breaks the status quo, they are defending calendar, not evaluating you.
The 5
5
percent, in a buying window
A contract coming due, a migration finishing, a mandate from a new leader, a build that slipped. That is when an account enters the window. That split is the work.
About 5 percent of a category is in-market in a given quarter; the other 95 is not. Heuristic, not a law, from John Dawes at the Ehrenberg-Bass Institute, written for the LinkedIn B2B Institute in 2021 (EB-1). The original paper.
What the brush-off actually is
Boundary, not evaluation.
Prospecting is not evaluation
You cannot negotiate interest on a thirty-second call.
The capacity model already plans 16 percent meeting-to-qualified. If your own discovery-to-qualified sits near 10, you are filling the calendar with the 95.
| Trigger | What to log | Re-entry |
|---|---|---|
| Incumbent renewal | Month of the window, and which contract | Quote the month they named |
| Migration or cutover | The date, and what is moving | "You said the migration wrapped in August." |
| Fiscal unlock | When budget opens, whose budget | Come back on that date |
| Internal build | Sprint versus a slide. A date if they have one. | Ask whether the build shipped |
Log the fact. Exit. Come back when it is due, quoting their words. Then sections 11-21 still run.
Do not manufacture the meeting. A dated trigger in the CRM is the win on a 95 account. A Stage 0 that cannot buy is not pipeline.
After the meeting
The meeting is not the job.
Pipeline gen gets you the conversation. What the AE does next is project mode. The AE who hit 200% two years running as we went $1M to $10M ran every live deal this way. The full run is playbook 02.7.
This page manufactures the meeting. 02.7 is how the AE runs it. Hire for the two buckets on the AE profile.
11 · The Test
A plan you execute, or a checklist you resent.
Every rep I have coached has built a pipeline generation plan at some point. Most of them built it once, for a manager, and never opened it again. The document is not the problem. The difference between the two versions below is whether anything written on Monday reaches a buyer by Thursday.
Plans that get worked
Short, specific, and falsifiable.
Checklists that rot
Long, generic, and safely unmeasurable.
Friday afternoon, read Monday's plan. If nothing on it reached a buyer, it was homework. Reps do not abandon planning because they are lazy. They abandon it because the last six plans changed nothing they did.
12 · The Program
Who owns which accounts, and what each seat is held to.
Context: the portfolio these quotas draw from is Part One.
A pipeline gen quota only works if the ownership underneath it is unambiguous. Any decently funded scale-up runs a mix of quota-carrying reps and BDRs, paired. The pair works a fixed set of accounts each month, and the split is deliberate: the rep takes the accounts that need a peer-level conversation, the BDR takes the volume. Neither of them works inbound. Inbound is handled by an agent workflow, described in module 03 of the Sales Playbook, so BDR time stays on outbound where the quota is.
The rep owns
25
must-break-into accounts
The named accounts where a first meeting has to come from a peer conversation rather than a sequence. The rep personally owns all outbound prospecting into them.
The BDR owns
75
accounts, rotating monthly
The rest of the pair's monthly hundred, refreshed every month so nothing sits half-worked into a second cycle.
Each month the pair identifies 100 new target ICP accounts between them. Campaigns into those accounts are prioritized on six personas.
5
Monthly rep goal: new first ICP meetings, stage 0
8
Monthly BDR goal: new first ICP meetings, stage 0
Thirteen a month across the pair is 156 a year, against the 147 a ramped seat is planned to in the capacity model. The spread is cushion, not an error. Quota is set above plan on purpose, so one soft month does not put the year behind.
Not all of those meetings come from the pair. Plan the mix, then hold the pair to their share of it.
The mix above draws from a bigger menu. Part One maps all nine engines.
An ARR quota tells a rep what to deliver. A pipeline gen quota tells them what to put in, which is the part of the year they still control in month one. Track both, review both weekly, and let the CEO see both.
13 · The Plan
Five fields, filled backward from the meeting.
Most pipeline plans start with a list of accounts and hope a meeting falls out of the far end. Start at the other end instead. Name the meeting you intend to book, then work backward through what has to be true for it to happen. The same five fields every week, in the same order, because the order is what makes a miss legible.
| Input | Where the number comes from | Worked example |
|---|---|---|
| Meetings you self-source each week | The capacity model. One per ramped seat, with the BDR carrying the other two. | 1 |
| Weeks of work before an account is ready to be asked | Your gate count in section 14. Three gates, roughly a week each. | 3 |
| Your account-to-meeting rate | Your own last twenty worked accounts. Not a benchmark, not mine, yours. | 1 in 3 |
| Accounts in flight | Weeks of work, times meetings needed, divided by your rate. | 9 |
| New accounts entering each week | Accounts in flight divided by weeks of work. | 3 |
Run your own rate through it before you accept the nine. A rep converting one in six needs eighteen accounts in flight and will drown trying to work them properly, which is itself the finding: at that rate the targeting is wrong and adding accounts makes it worse.
Five fields, filled the same way every week, is a system. Five fields filled differently every week is a diary. The value sits entirely in the repetition, because repetition is the only thing that makes a miss legible.
14 · The Gates
You have not earned the ask until you can say three things out loud.
Here is the rule I install first, because it changes behavior faster than any script. An account stays on the plan until you can say all three of the statements below from memory, in your own words, without opening a tab. Each one is earned on a call, not read off a website. Most reps skip straight to the ask holding none of them, get silence, and diagnose a targeting problem.
Three statements, each earned from a human, each checkable on a Friday. Most reps run twelve accounts for one week apiece, hold none of the three on any of them, and conclude that outbound is saturated. It is not saturated. It is ungated.
15 · Build It
Fifteen minutes, most of it not typing.
Fill in five things. The panel builds a research prompt for one account, sized to your motion and carrying last week forward. Paste it into an assistant with web search on, then spend the remaining twelve minutes reading the output like a skeptic and deleting the parts it made up. Nothing here is stored or sent anywhere.
Run it with web search enabled. Everything it returns is a hypothesis until a human confirms it on a call.
An assistant is fast at gathering and confident when wrong, which is a bad combination on a first dial. Make it flag every inference, then cut anything you would not say to the person's face.
16 · The Touches
Nobody replies to a well-written email. They reply to a reason.
The plan tells you who to work. This is what you send them, and in what order. Most outbound training optimizes the words, which is the wrong end of the problem: the words are being judged by someone who decided whether to engage before reading a full sentence. What you actually control is the shape of the message, how specific it is, and whether each touch does a different job from the last one.
Swipe to see all columns
Two different failures, two different fixes. Rewriting your value proposition does nothing about the first screen, which is where most of your sends die.
Swipe to see all columns
Five touches, five mechanisms, two channels alternating. This is the touch pattern that carries you through the three gates in section 14.
The first below-the-line call is a timing and fit split, not a pitch: whether a window exists, not whether they like the deck.
| Touch | Channel | Mechanism | What it must carry that the last one did not |
|---|---|---|---|
| Day 1 | Call, below the line | Human contact | One question about what they run. No pitch, no ask, no voicemail longer than fifteen seconds. |
| Day 2 | Email, below the line | Open loop | Something you learned or failed to learn on the call. The email exists because the call happened. |
| Day 5 | LinkedIn, above the line | Specific proof | One named comparable. LinkedIn is where peer claims carry most, so spend the proof point here. |
| Day 9 | Call, above the line | Direct ask | What they run plus what the team is measured on. This is the call that stops sounding cold. |
| Day 14 | Email, above the line | Real loss | The competitive clock, and an honest close-out. Your attention is finite and saying so is not a threat. |
The rule underneath the table: no mechanism repeats, no channel runs twice in a row, and every touch carries something new. A sequence where touch four is touch one with "just circling back" on top is not persistence. It is a rep confirming the buyer's original read.
| What you see | What is broken | What to change |
|---|---|---|
| Low opens, low replies | Screen one, before the message is even open. Sender, subject, preview text. | Change the shape of the subject line, not its cleverness. Try statements and fragments instead of questions and hooks. |
| Good opens, low replies | Screen two. You got read and gave them nothing to do about it. | Rebuild around one mechanism instead of around a value proposition. |
| Replies, but hostile ones | The opener wrote a check the body did not cash. | Make the body deliver exactly what the hook implied. An open loop that resolves into a generic pitch is worse than no loop. |
| Replies, no meetings | The ask is too heavy for a first exchange. | "Worth ten minutes?" rather than "book a 30-minute call." Replying should cost less than ignoring. |
Before you send
Five checks. Any one failing sends it back.
What I would not do
Advice that circulates and should not.
Five touches that each do a different job is a sequence. Five touches that each say "just following up" is a rep teaching a buyer to ignore them, one message at a time. The second is far more common, and it is the actual reason most people think outbound stopped working.
17 · The Order
Copy is the fourth thing that matters.
Copy is the part of outbound everyone wants to argue about, and it is the least likely thing to be broken. When a campaign underperforms there are four candidates, and they are not equally likely, equally expensive, or equally fixable. Rank them by how much each one moves a reply rate, then fix them in that order. Most teams work the list upside down and spend a quarter rewriting subject lines while the fault sits two layers underneath.
| Input | Where the number comes from | Worked example |
|---|---|---|
| New logos needed this quarter | Your quarterly number divided by ACV | 5 |
| Closed-won from meetings held | Your last forty stage 0 meetings. Yours, not a benchmark. | 20% |
| Meetings required | Logos divided by close rate | 25 |
| Positive reply to meeting held | Booking rate times show rate, both from your own history | 35% |
| Positive replies required | Meetings divided by that rate | 72 |
| Positive share of all replies | One month of your own inbox, counted honestly | 45% |
| Replies required | Positive replies divided by that share | 160 |
| Reply rate | Your last complete campaign | 4% |
| Sends required | Replies divided by reply rate | 4,000 |
This is the same exercise as the account table in section 13, run from the other end. That one derives accounts in flight for the rep. This one derives send volume for the BDR. Chain them and the quarterly number stops being a target and becomes a Monday activity level. Run your own rates through it before you accept the four thousand. A team converting one in four hundred needs ten times the sends, which is not a capacity finding. It is a targeting finding, and buying more sends makes it worse.
| What you see | Range | Where the fault sits |
|---|---|---|
| Open rate | 40-60% | Below 30% is the floor, not the copy. Treat opens as an instrument for deliverability and never as a result. |
| Reply rate | 2-8% | Below 1% with healthy opens is the offer or the list, in that order. It is almost never the subject line. |
| Positive share of replies | 40-50% | Low here is offer framing. You reached the right people and handed them the wrong reason. |
| Meeting booked from a positive reply | 30-50% | Low here is the weight of the ask, or how long a reply sat before a human answered it. |
| Show rate | 70-80% | Low here is confirmation, not outreach. The meeting got booked and then left alone for nine days. |
| Closed-won from meetings held | 20-30% | Not an outbound number. If every layer above cleared, outbound did its job and the problem is downstream. |
One variable at a time, per section 19. Fixing the offer and the list in the same week gives you a better month and no idea which change earned it.
Where volume belongs
The claim that outbound is a volume game first and a skill game second is half right, and the half it gets right is the BDR lane. The 75 rotating accounts are worked with sequences, and a sequence needs enough sends for a rate to mean anything. Four percent on eighty sends is three replies and no information. The top 25 are the opposite: worked by a rep, one at a time, through the three gates in section 14, where adding volume spends the account faster without producing a meeting. Teams fail by running one lane's logic in the other, and both directions are common.
Offer, list, floor, copy. A campaign that fails at the offer cannot be recovered at the copy, but a team that only knows how to rewrite copy will try anyway for a quarter and then conclude that outbound stopped working. The order is what makes a bad month diagnosable instead of demoralizing.
18 · The Week
Where the fifteen minutes sit.
The plan is worth building only if the week is built around it. Five moves, in order, on the same operating cadence as the rest of the Sales Playbook.
Fifteen minutes on Monday, one self-sourced meeting a week, 49 weeks. That is the third of the three meetings a ramped seat needs, and it is the only one nobody else can book for you.
19 · The Rerun Problem
Forty-nine weeks of evidence, or one week run forty-nine times.
Two reps do the same volume for a year. One is measurably better in December than they were in March and can tell you exactly why. The other is running week one on repeat with fresh logos in it. The difference is not effort and it is not talent. It is three pieces of unglamorous administration, and every rep I have coached who got sharper had all three.
| Failure | On a rep's calendar | The fix |
|---|---|---|
| Two tests, one channel | Two subject lines hitting the same inbox pool in the same week. Both numbers are noise. | One experiment owns one audience slice at a time. |
| Fake wins | Opens, clicks, and "a lot of good conversations." Nobody has ever forecast on a good conversation. | Grade on meetings and pipeline, every time. |
| Effort instead of experiment | Reply rate flat for three weeks, so the answer is more of the same, sent harder. | Change what is being tested, not how loudly. |
| No paper trail | Nothing written down, so the only memory of the account is how the rep feels about it. | Log the change, the reason, and the result. |
An assistant can propose a hundred variants a week. It cannot tell you which one deserves to survive, because that call is a business number and the number is yours. Four quarters of keeping the winners and cutting the rest does not give a rep better instincts. It gives them a longer log, which beats instincts.
20 · What the research costs
Cost per token is the wrong unit. The arithmetic going around is also wrong.
Section 19 says a week only compounds if you can afford to run the research behind it. This is that number. The idea circulating right now is that the most expensive model is secretly the cheapest, because it finishes analytical work in one pass while a cheaper model needs four rounds of correction. The idea is right. The multiples attached to it, in the version I keep getting sent, are not, and they are wrong in the direction that sells you something.
What is being claimed, and what is true
The claim: the top model costs five times what Opus costs and twenty-five times what Sonnet costs, so it had better be worth it.
Actual published rates on 27 July 2026: Fable 5 is 2x Opus 5 and 5x Sonnet 5 at its introductory rate. Not 5x and 25x. Whoever is quoting those multiples is comparing against a rate card that does not exist, and the inflated version conveniently makes a $20 agency line item look like a bargain.
The correction matters because it changes the answer. At 2x, the frontier tier is a reasonable default for hard analysis. At the claimed 5x, it would be a luxury you ration. Anthropic's own guidance is to start with Opus and step up only when you need the ceiling.
| Model | Input | Output | Where it belongs in a GTM stack |
|---|---|---|---|
| Haiku 4.5 | $1 | $5 | Classification and enrichment at volume. Tagging, deduping, routing, anything you would have written a regex for. |
| Sonnet 5 | $2 | $10 | Copy. Drafting, variants, rewrites. Introductory rate through 31 August 2026, then $3 and $15. |
| Opus 5 | $5 | $25 | The default for analysis you will act on: account research, campaign diagnostics, competitive reads. |
| Fable 5 | $10 | $50 | The ceiling. Long context in one pass, no human checking the middle of it. |
Rates change. Check the published pricing page before you build a budget on this table, and treat the date above as the expiry.
Swipe to see all columns
One question decides it, and it is not "which model is best." It is what a wrong answer costs you.
Route by what a wrong answer costs. A redraft, use the cheap tier. A call with a buyer you get one shot at, pay for the pass that does not need checking. Most teams have this exactly inverted: premium models writing emails a human rewrites anyway, and the cheapest available model deciding who is worth calling.
21 · The Stack
The tools that run it, and the one that runs them.
Main article: module 03 of the Sales Playbook.
None of the above requires a large team. It requires a small one with the right stack and a model sitting in the middle of it. What follows is what I install: two lanes, one that finds and enriches the accounts, one that touches them, with Claude orchestrating both through their APIs rather than a human clicking through seven interfaces.
Find and enrich
Attio
The CRM, and the system of record for the hundred accounts. Everything else writes into it, so the account list and the meeting count live in one place.
Apollo
Contact discovery against the six personas. This is where the named humans come from once the account list is set.
Clay
Enrichment and research at list scale. Job posts, funding, tech signals, headcount moves, waterfalled across providers so a thin record gets filled rather than dropped.
Touch and dial
Origami or Instantly
Automated email sequencing. Either one works. What matters is that the copy is built from the research above rather than from a template with a merge field in it.
HeyReach
Outbound LinkedIn at team scale. The second channel on every account, run in step with the email rather than as a separate campaign nobody sequences.
Nooks
Auto-dialing for cold calls, and the most important tool on this page. Dials are still what produce first meetings on the top 25. Nooks is what makes the volume survivable.
Claude sits in the middle and drives all six through their APIs. Build the account list, enrich it, draft the sequence from what the research actually returned, load the campaign, queue the dial list, and write the result back to the CRM. The rep reviews and approves. They do not do the assembly.
The rule that makes this safe is the one that has not changed: a human approves every external send. The model gathers fast and is confident when it is wrong, which is a bad combination on a first touch. It drafts, a person signs.
The full engine, including the inbound agent and the workflow catalog, is module 03 of the Sales Playbook.
Six tools and a model is a two-person outbound function. None of it books a meeting on its own. What it removes is the assembly, so the rep spends their hours on the twenty-five accounts that need a human and not on moving records between tabs.
Social Content
The founder's profile is the channelHow the best run it
Personal profiles pull roughly three times the engagement of company pages, and buyers arrive pre-sold from weeks of reading. The cadence that works is small and relentless: three posts a week in one lane, plus daily comments where the ICP already argues. Pipeline forms in the signals, not the impressions: repeat commenters and profile viewers from target accounts get a warm note referencing what they engaged with, and site-visit resolution connects the post to the account that showed up two days later. Every strong post becomes retargeting fuel and a validated topic for the SEO engine.
Consistent founder-led programs see inbound conversations roughly triple inside 60 days. Inbound closes an order of magnitude better than cold.
Every number in this engine carries a claim ID. Read the social content card in the open source repo.