Jordan Crawford's 2026 GTM stack → Nautilus
Know more about the 1,001st car wash than the owner knows about his own.
Information asymmetry, problem-first targeting, and a Claude Code–centred stack — from the person many call the best GTM engineer working, translated to how Nautilus finds and messages operators.
The idea underneath everything he does
“If you have 50, 100, 1,000 customers, you should know more about the 1,001st customer than that customer knows about their own reality. And this is almost never the case.”
Jordan Crawford is a fractional GTM engineer (Blueprint GTM), a Clay advisor since 2020, and now runs “100% of my work in Claude Code.” His goal: help teams “send messages that they would be proud to receive.” [1:56]
- The default founder mindset is “my money is trapped in your wallet” — which produces “Hi Tim, congratulations on being named Tim… do you have 15 minutes?” [3:29]
- The fix is information asymmetry: know the customer's problem better than they do, then target “who has the worst version of the problem… not based on ZoomInfo filters, based on problems.” [7:20]
- “The message is just a re-description of the targeting.” [12:33]
“If go-to-market is hard for you, you're too wide.”
[18:15] — a vertical is “a knowable universe with knowable problems. Your knowledge compounds, it does not fracture.” Nautilus is already a vertical. The question is whether we act like one.Source: Jordan Crawford (Blueprint GTM) on the FullEnrich podcast, “The best GTM engineer in the world explains his 2026 tech stack,” Aug 2026. Timestamps in brackets.
Where we stand
We already hold the information asymmetry he's describing. We just don't put it in the first email.
His examples
A founder who did a dentist's billing for free for six months before writing code — he knew the cleaning code and had filed 5,652 of them. [4:35] A permit-intelligence founder who talked to a thousand officials until “I knew what they were going to say before they said it” — and has never done outbound. [5:25]
The best message: “here's the output you can expect if we work together, and here's how I know that to be true.” [7:40]
What Nautilus knows that an operator doesn't
- What a membership checkout converts at when it works (New Day: 67%) — and what theirs does today (usually: it doesn't exist).
- What share of members churn through failed payments on each POS, and how much a recovery flow brings back.
- Which POS vendors' e-commerce pages leak customers — from running 43 washes' real data, not a pitch deck.
- What a Google review velocity drop at one site predicts.
The gap: none of this is in our opening message. We lead with “platform” and “AI” — generic. He'd call that the Tim email.
Source: Jordan Crawford (Blueprint GTM) on the FullEnrich podcast, “The best GTM engineer in the world explains his 2026 tech stack,” Aug 2026. Timestamps in brackets. Internal figures: Nautilus production data; approvals needed before quoting non-public customer numbers.
Problem-first targeting
Rank buyers by the consequence of the problem, not by filters. Then write to the top of the list.
His worked example (voice AI) [9:04]
Two prospects would both buy a voice agent: a podiatrist who misses ~10 after-hours calls a month, and a heart-surgery ER where a missed call means a $500,000 re-operation preventable by “three or four questions.” “Both would transact, but one of them has a much greater need.” Let the podiatrist buy inbound. Outbound goes to the ERs — found and ranked by public procedure-volume data.
The Nautilus version
- Who has the worst version? A multi-site express chain with an Unlimited program, a POS we integrate, and no online checkout — every day their site sends members to a kiosk line.
- Rank by consequence: sites × members × the conversion gap. A 6-site chain with 4,000 members loses more per month than a single wash, and we can estimate the number.
- Let single sites buy inbound. Outbound effort goes to the ranked list, with the dollar figure in the message.
Source: Jordan Crawford (Blueprint GTM) on the FullEnrich podcast, “The best GTM engineer in the world explains his 2026 tech stack,” Aug 2026. Timestamps in brackets.
List building
Work backwards from 50 customers you already know. Then build forward from where their data lives.
His method [19:15]
Take ~50 known-good records (email, phone, owner). Have Claude find where each data point actually lives — “Florida has every dentist publicly listed with all of their emails.” Then “go build forward from those sources.” “The models do exceptionally well when they're working backwards from known examples.”
Do it by hand first [20:37]
Research 10–50 leads manually and record your screen narrating: “I trust this PDF because there's a date on it.” Then have Claude abstract what you found to be true. Let an agent find the acquisition path — a state dataset “you can buy directly for $100.”
Nautilus: our 43 as the seed
- For each customer: where did the owner's name, mobile and email actually come from? State business registries, Google Business Profile, the wash's own site, LinkedIn, ICA/regional association lists.
- Record Sam doing ten by hand, narrated. That recording becomes the agent's playbook.
- Then enumerate every wash in TX and CA from the same sources.
Source: Jordan Crawford (Blueprint GTM) on the FullEnrich podcast, “The best GTM engineer in the world explains his 2026 tech stack,” Aug 2026. Timestamps in brackets.
The move that fits us best
“You can download about 80% of a market.” Then let code, not people, read every website.
He enumerates everyone first and filters down [39:56], pulls the whole market's websites locally (Apify for the Cloudflare-blocked 20%) and analyzes them with code [44:06]. “I spend less than $500 on enriching an entire TAM.” [40:48] Cell numbers are the one thing worth paying up for.
Older build of his: scraped new real-estate listings, analyzed the photos for homes needing the most electrical work, sold the list to electricians. [49:32]
Nautilus: the signal nobody else has
- Every car-wash website in our states, downloaded. Code answers, for each: Can you buy a membership online? Which e-commerce vendor (Rinsed / DRB / Sonny's script tags)? How many locations listed? Phone, hours, Unlimited prices.
- Photos answer the tunnel vs. self-serve question we couldn't answer from a CRM last week.
- Output: a ranked TAM with a dollar-sized problem per account, for well under $1,000.
Source: Jordan Crawford (Blueprint GTM) on the FullEnrich podcast, “The best GTM engineer in the world explains his 2026 tech stack,” Aug 2026. Timestamps in brackets.
Two filters he says to drop
Titles lie and headcount bands are absurd. Pull wide, then disqualify on behavior.
What he says
“The titles that we have in our heads are a poor representation of titles in the market.” Since data is cheap: “give me anyone with this wide berth of titles, and then I'm going to go disqualify them.” [34:20] On headcount: “Do you think some CEO hired their 201st person and everything needs to change around here?” Target behavior instead — procurement people “in seat for two years or less… because they're driving innovation.” [38:42]
Nautilus check
- Owner, GM, Director of Ops, Marketing Manager, “Partner” — pull all of them per account and disqualify after. Our best champions have had odd titles.
- Location-count bands are our headcount bands. “5–20 sites” is a filter, not a reason to buy.
- Behavioral criteria that actually predict: a GM or marketing lead in seat under two years; a chain that just opened a site; an owner who posts on LinkedIn about memberships.
Source: Jordan Crawford (Blueprint GTM) on the FullEnrich podcast, “The best GTM engineer in the world explains his 2026 tech stack,” Aug 2026. Timestamps in brackets.
His 2026 stack
“Claude Code plus a bunch of really trusted APIs.” No sequencer, no CRM product, no orchestration tool named.
What he uses
Nautilus
Source: Jordan Crawford (Blueprint GTM) on the FullEnrich podcast, “The best GTM engineer in the world explains his 2026 tech stack,” Aug 2026. Timestamps in brackets. Vendor note: the host is FullEnrich; Jordan discloses he's a FullEnrich investor-advisor and a Clay advisor.
How he spends
“What's the lowest possible dollar amount I can spend for a right answer?”
The ladder [36:27]
Free and programmatic first. Cheap APIs second. Expensive models only to analyze cheap answers, and a full agent “only if I need to escalate.” His analogy: Apple's M1 won on performance per watt.
The guardrails [43:09]
“Please, do not give me unlimited spend. I will for sure accidentally spend this.” He works in the client's own Claude Code account and API keys; clients buy vendors directly. A $67K bill scare turned out to be ~$600 — the API reported cents.
Nautilus
- Spend caps on every provider key, including Shisha's.
- Enumeration and site analysis are free/cheap tiers; reserve paid enrichment for the ranked top of the list.
- Agents use structured APIs for data collection — “Claude will confidently be like, it's done. But it will not do that.” [45:54]
Source: Jordan Crawford (Blueprint GTM) on the FullEnrich podcast, “The best GTM engineer in the world explains his 2026 tech stack,” Aug 2026. Timestamps in brackets.
How he works with the agent
Speak, don't type. Teach it like a blank adult. Protect the context window. Never bypass permissions.
His rules
- “Speak not type, because typing you edit yourself.” Ask Claude to ask you questions. [22:17]
- “The prompts are less important than the tool capabilities.” [23:04]
- Onboard it like “cloning a 30-year-old who knows nothing” — order matters; you can't say “go find an email” without explaining what one is and the eight patterns it follows. [23:11]
- Sub-agents return only the answer: “the models perform so much better in the first 20 to 50% of their context window.” [23:46]
- “I used to use bypass permissions and then I deleted my whole home folder.” Auto mode + hooks that stop on remove/publish. [24:32]
Nautilus
- Sam and Amayr brief Shisha by voice memo; Shisha asks back before building.
- Our playbooks (skills) are his “knowledge blocks” — keep writing them as procedures, not prompts.
- Guards already exist on prod writes, Monaco bulk actions and customer sends. Keep them; add spend caps.
- “Ask for things whose consequence won't burn you” — no “delete what's unused” asks, in code or in the CRM.
Source: Jordan Crawford (Blueprint GTM) on the FullEnrich podcast, “The best GTM engineer in the world explains his 2026 tech stack,” Aug 2026. Timestamps in brackets.
His explicit don'ts
Checked against how we sell today.
Don't
Nautilus today
Source: Jordan Crawford (Blueprint GTM) on the FullEnrich podcast, “The best GTM engineer in the world explains his 2026 tech stack,” Aug 2026. Timestamps in brackets.
Proposed · next 30 days
One build that changes everything downstream: the downloaded, code-read market.
- Backwards from the 43: where each owner's contact actually came from. Sam records ten by hand.
- Enumerate TX + CA washes from Google Maps, state registries and association lists.
- Download and analyze every site: online checkout yes/no, e-commerce vendor, locations, prices, tunnel vs. self-serve from photos.
- Rank by consequence (sites × members × gap) and rewrite the first email to state the number.
Budget target per his numbers: under $1,000 in data. Owners and dates to be set by Sam and Amayr; this is the proposal.
“Here's the output you can expect if we work together, and here's how I know that to be true.”
We have the data to write that sentence to every express wash in Texas and California. The work is reading their websites with code and putting the number in the first line.