Topline · 19 Apr 2026 · From the week of 13 April
The #1 GTM Engineer In The World | Jordan Crawford, Founder @ Blueprint GTM
These are notes on the conversation, checked against its transcript. The episode itself has the full discussion.
In brief
Jordan Crawford, founder of Blueprint GTM and an advisor to Clay, joins hosts Sam Jacobs, Asad Zaman and AJ Bruno on Topline to discuss how go-to-market teams can use AI. The episode covers his 'Great Inversion' thesis that tools should inform systems and strategy rather than the reverse, the changing role of RevOps toward demand generation, and practical examples such as extracting competitor user data and identifying pain-qualified segments. It also covers churn and account prioritisation, contract review, and the outlook for AE compensation. The central argument is that leaders need hands-on familiarity with AI tools so they can design systems and strategy from what the tools can actually do, rather than issuing top-down mandates.
For founders
- Jordan argues that top-down AI mandates from CEOs fail when leaders cannot say what AI means or how to implement it.
- Jordan says teams should start from the outputs AI tools can produce, work back to the systems layer, and set strategy on top of that.
- Jordan says AI works best on tasks with a well-defined output and inputs it can process, so he asks customers to define the exact output first.
- Jordan says customer words and product actions are closer to the truth than CRM fields, so he builds nightly customer dossiers from them.
- Jordan says a campaign is a useful test unit because a small group can run one and the list can be rebuilt within days from the results.
For revenue leaders
- Sam Jacobs says many RevOps people still see their job as reporting and comp design, but he argues the job is now to generate demand by running many campaign tests.
- Jordan says a vague 'use more AI' RevOps mandate should be replaced by a named workflow, such as pulling a task out of a specific SDR team's work to save them time.
- AJ Bruno says he is bearish that top 10% AE on-target earnings will rise 50% in two years, expecting AI to compress the middle and leave fewer but not richer AEs.
- Jordan says a pain-qualified segment describes the buyer situation that gets the most value from a product and replaces a traditional ICP.
- Jordan says that for churn analysis, the team first had to establish which Salesforce fields were trusted, which took about four weeks and excluded the MRR field.
What was said 22, most useful first
A competitor's password reset page that confirms whether an email has an account can be used to find its real users by submitting a list of emails. Listen
Jordan describes a competitor whose reset page said it could not find a user with the entered email, rather than giving a generic message. He says he submitted 110,000 password reset emails drawn from the total addressable market and identified the competitor's actual users this way.
“What this one competitor did is they said, uh, we can't find a user with this email. So what I did is I took all of the emails in their tam and submitted 110,000 password reset emails.”
Listen to the episode Pipeline & demand generation Link to this Report a problem
Customer words and product actions are closer to the truth than CRM fields, which Jordan calls the furthest thing from the truth. Listen
Jordan says he runs nightly jobs that compile complete customer dossiers from customer words, product telemetry and CRM data. He ranks the CRM as furthest from the truth, customer words as closer, and customer actions as closest.
“The CRM is weirdly the furthest thing away from truth. The customer's words are closer to truth and their actions are the closest to truth.”
Listen to the episode Retention & customer success Link to this Report a problem
AI analysis of customer words and actions can flag accounts worth about ten times more than others, so teams can ignore the rest. Listen
Jordan says his nightly runs identify the accounts flagging the most issues, and that with an understanding of why customers stay or churn, some accounts are worth 10 times more than others. He says the other 90% are ones intervention will not save, while accounts that say a priority will come in six to 12 months are worth engaging.
“And so ignore this 90%. These 90%. These are the ones where actually intervention can help.”
Listen to the episode Retention & customer success Link to this Report a problem
Campaigns are a practical unit of GTM testing because a small group can run one, and the list can be rebuilt within days from the results. Listen
Jordan describes a client who cold called hypothesis-based lists, with AI reading the call transcripts to show which segments to focus on and which to drop. He says a campaign shipped on Monday led to corrected titles by Monday afternoon, a new list with 100 new people called by Tuesday, and a list based on the new insight by the end of Tuesday.
“Campaigns are kind of a canonical unit because you can have a very small number of people go and execute them in a test.”
Listen to the episode Pipeline & demand generation Link to this Report a problem
Companies should ship a campaign and fix it, rather than first settling territory and rep assignments in the sequencing tool. Listen
Jordan says companies often ask whether they can push a campaign into Outreach, then worry that Outreach has 468 SDRs assigned across different accounts and territories that must be right first. He says they do not need to do this. Instead they should ship the campaign, see if it works, and when it fails, fix it and ship it again, because even a failure makes huge progress.
“Just ship the damn thing, see if it works, and when it fails, it will fail because it makes huge amounts of progress.”
Listen to the episode Sales process & deals Link to this Report a problem
AJ Bruno expects AI to compress AE pay in the middle of the distribution rather than raise the top 10% of reps' on-target earnings by 50%. Listen
AJ Bruno says he is bearish on a 50% rise in on-target earnings for the top 10% of AEs in two years. He estimates top reps make about $300,000 to $400,000 today and says he expects fewer AEs but not necessarily richer ones. He says he would be bullish on the top 1%.
“I think there'll be less AEs but not necessarily richer AEs”
Listen to the episode Sales team, hiring & comp Link to this Report a problem
Teams should start from what AI tools can produce at the output layer, work back to the systems, and place strategy on top. Listen
Jordan contrasts this with the older approach of setting strategy first, then building systems, then choosing tools. He says leaders need to play with the tools to learn their strengths and jagged weaknesses, and that this is not the same as vibe coding.
“because you can basically start from the sort of outputs layer, work backwards the system layer, and the strategy layer can sit on top of that.”
Listen to the episode Strategy & market Link to this Report a problem
Replace a vague AI mandate for RevOps with a named workflow that can be pulled out of a specific team's work. Listen
Jordan says telling RevOps to make things more AI is too vague to act on. He suggests naming the team and the task, such as an SDR team's work, and removing that task to save them time. He adds that RevOps overlaps closely with what AI can do but needs a strategy and a CRO who knows the tools.
“you need to say, well, okay, our SDR team is doing this, we need to like pull that out of their work and save them time.”
AI coding agents can identify every user of a large software vendor, including each user's tier, from a single overnight run. Listen
Jordan describes a client that sits on top of a large vendor and wanted to know all of its users. He says Claude Code determined about 103,000 users and whether each was on enterprise, mid-market or free tiers. He stresses the output is deterministic, so it is easy to check, and that a team would only think to ask this if it knew what the tools could do.
“Claude was basically able to not only determine all 103,000 of their users”
Listen to the episode Pipeline & demand generation Link to this Report a problem
Attendee names from a phone-only conference app can be captured by screen capture, extracted with AI, enriched, and sorted by who to meet. Listen
Jordan says the conference app was iPhone-only, so he connected the phone to his computer and had an agent scroll the app and take screenshots overnight. AI extracted names and titles, enrichment added photos and LinkedIn URLs, and the attendee list was sorted by who to talk to.
“And so overnight, I had it scroll my iPhone and take screenshots.”
Listen to the episode Pipeline & demand generation Link to this Report a problem
A pain-qualified segment is a description of the buyer situation that most needs the product, and it replaces a traditional ICP. Listen
Jordan starts by asking what segment would need the product most if he could draw a perfect boundary around the market, then finds data that identifies people in that situation. The segment is described back to the buyer, for example 10 of 12 top ex-president's club SDRs leaving after a comp plan change.
“A pain qualified segment is really a replacement of ICP.”
Listen to the episode Positioning & marketing Link to this Report a problem
A permissionless value prop is a message that is independently valuable to the recipient, even before they buy. Listen
Jordan contrasts this with a pain-qualified segment message. His example is competitor understanding, such as what top competitors do differently on comp plans that has improved rep retention, which he frames as about 2x average rep retention based on his reading of LinkedIn.
“And a PVP message is a message that is independently valuable to them.”
Listen to the episode Positioning & marketing Link to this Report a problem
Public rep sentiment and interviews with former reps could be used to produce a brief on what is wrong with a prospect's comp plan. Listen
Jordan describes an imagined autonomous system that gathers public information on how people discuss comp plans, then presents a company with the problems it predicts from its own data. He uses a hypothetical in which 12 former SDRs were interviewed, and says the same approach could target a competitor's long-tenured reps. AJ Bruno said QuotaPath already runs a version of this play.
“you can kind of imagine if you could suck up sort of all of the public information about how people are doing comp plans and how they're talking about them.”
Listen to the episode Sales team, hiring & comp Link to this Report a problem
In a churn analysis, the first step was determining which Salesforce fields are trusted, which took about four weeks and excluded the MRR field. Listen
Jordan describes a customer churn project in which the first step was establishing trusted Salesforce fields, a process he says took four weeks and found the MRR field not trustworthy. Next the team built customer journey files, which he calls dossiers, and agreed on the schema before analysing why customers churn.
“which fields in Salesforce are trusted. So that's like a four week process, but not this MRR field.”
Listen to the episode Metrics & finance Link to this Report a problem
The RevOps role is shifting from reporting and comp design toward running demand-generating tests, and tool capabilities should set priorities. Listen
Sam Jacobs says many RevOps people grew up believing their job is to design comp plans and deliver dashboards. He argues that time spent building a perfect dashboard could instead go to running tests that generate demand, even if the dashboard is less polished.
“well, actually your job now is to generate demand. And like I need 50 campaigns tested by tomorrow, at least via email.”
Listen to the episode Pipeline & demand generation Link to this Report a problem
A team that agreed to ship campaigns in meetings for six weeks had shipped nothing, and that it needs to become a doing culture. Listen
Jordan describes a client who said in six meetings over about a month and a half that work would be done next week, with nothing shipped. He built a dashboard with 40 copy-paste prompts for Claude Cowork that exports lists from a segmented team list, and says the rise in calls came from someone deciding to hold people accountable, not from AI.
“We probably need to become less of a meeting culture, more of a doing culture.”
Listen to the episode Leadership & culture Link to this Report a problem
If shipping a project faces heavy organizational friction, teams should pick a different project they can ship quickly. Listen
Jordan says the way to choose AI projects is to design them to ship at a rapid rate. He gives the example that getting everyone to change how they use Salesforce is unlikely to happen, and suggests asking what the work would look like at 100x.
“And if there's going to be a lot of organizational friction to shipping something, maybe we should pick a different project.”
Listen to the episode Strategy & market Link to this Report a problem
Every job needs to be unbundled into its tasks, and job descriptions should be rewritten around what an AI-enabled version of the role looks like. Listen
Jordan says leaders must identify the series of tasks each role contains, which tasks should not exist, and which should get more time. He cites Jack Dorsey's public account of doing this at Block and describes redesigning an organization from screen recordings as a massive, risky undertaking.
“every job now needs to be unbundled, which is you need to kind of figure out what are the series of tasks that everyone has”
Listen to the episode Hiring & team building Link to this Report a problem
AI works best when the output is well defined and the inputs are available, so the exact output should be defined before choosing an approach. Listen
Jordan says that when he works with customers he asks them to define the output first and show the exact thing they want, then identify where the inputs are. Problems are ill-defined for AI when the output is something like a closed-won deal. He suggests asking what you would have engineers build if you were a product manager.
“So when I work with my customers, I say, define the output first. Show me the exact thing.”
Reviewing historical contracts can show which terms large buyers are likely to request, so those reviews can be prepared in advance. Listen
Jordan says a team can take all its contracts, identify the deal terms large companies will not accept in contract review, and list the terms they are likely to ask for. He adds that the team can identify which legal reviews are acceptable from historical contracts and define the fastest-closing complex deal and what happens in its final stage.
“If you find that there are certain deal terms that large companies won't accept in contract review, you can take all of your contracts and say, here are the things that they're probably going to ask for.”
Listen to the episode Sales process & deals Link to this Report a problem
Jordan Crawford is bullish that Clay will reach a $50 billion valuation within five years, mainly because production workflows tend to run there. Listen
Jordan discloses that he is an advisor and that his financial situation would change if the prediction holds. He says the main risk is that tools like Claude Code could absorb everything, but that when workflows need to be operationalized, Clay is usually the system, and its enterprise proofs of concept are sticky.
“clay is usually that system, not because you can't go faster in cloud code, you can”
Listen to the episode Strategy & market Link to this Report a problem
Top-down AI mandates from CEOs fail when the leader cannot explain what AI means or how to implement it. Listen
Jordan says many CEO letters demand an AI-first organization without saying what that means in practice. He argues that leaders who have not used the required systems cannot set a workable direction, which is why such mandates stall.
“You read these letters from all these CEOs and they're like, we need to be an AI -first organization and I can't tell you what that means or how to implement it, but goddamn, you need to be able to do it.”
Listen to the episode Leadership & culture Link to this Report a problem