5/19/26: Most Personas Are BS. Here's the Data-Driven AI Fix
Greetings everyone! It is time for the GTM AI Podcast with today our guest the CEO of Wrench.ai, Mr. Dan Baird.
As usual, this podcast and the goodies along with it are always free and we would love your feedback of any topic you want covered from the vast network of amazing people we know.
Please subscribe and share with your peeps
Let’s get into it.
You can go to Youtube, Apple, Spotify as well as a whole other host of locations to hear the podcast or see the video interview.
Your “personalized” outreach isn’t personal. It’s just polite.
Every cold email in your inbox right now mentions where you went to college, what company you joined, or the post you wrote three weeks ago.
That’s not personalization. That’s a name swap on a template.
Dan Baird, the founder and CEO of Wrench.ai, has spent nine years building what he calls a “RoboCMO.” On this week’s GTM AI Podcast, he made one distinction that should reorganize how every revenue leader thinks about AI outreach in 2026:
“Personalization and relevance are not the same. Most teams are personalizing to YOU. Almost none of them are adapting the message to your audience’s preferences.” — Dan Baird
That gap is why your reply rates are flat even though your tool stack tripled. Here’s what to do about it.
1. Stop letting LLMs do your decisioning.
LLMs are autocomplete on steroids. Machine learning is statistics on steroids. They are not the same tool, and they don’t solve the same problem.
When you plug Clay (or any LLM-only stack) into your outbound, the LLM is doing both the message AND the targeting decision. That’s where the wheels come off. LLMs are great at writing the sentence. They are bad at telling you which sentence will convert this specific buyer.
The fix: separate the layers.
Use ML to build the buyer profile (personality, preference, adoption stage, jargon density).
Use the LLM only to translate that profile into the message.
Audit the decisioning. If you can’t explain why the AI chose this angle, you can’t optimize it.
Dan’s framing: “It’s a much more predictable, much more likely-to-actually-work outcome.”
2. Most of your personas are BS. Burn them.
Here’s the Coach K test: open the deck where your personas live. How dusty is it?
Dan went further: “Most of the time, the person with the coolest-looking turtleneck gets to decide what our personas are. They aren’t driven by data.”
The replacement is not 12 prettier personas. It’s 4 to 8 data-driven ones split across the adoption curve:
Early adopters want jargon, complexity, and the cutting edge. They tolerate friction. They pay full price.
Late adopters want simplicity, social proof, and discounts. They want to see it work before they buy.
The same product needs two different messages for these two groups. Not because of demographics. Because of risk tolerance.
The tactical move: pull your last 100 closed-won. Tag them as early or late adopters by behavior, not title. Rewrite the top of your funnel for both. Test the split.
3. People mirror their salespeople. Use it.
Dan ran a study against a 27-million-person database for a direct sales client. The question: what predicts a top performer before training?
Answer: top reps got their first commission check fastest, and the customers they closed mirrored their own personality type. Extroverts closed extroverts. Introverts closed introverts. It wasn’t training. It was psychological matching.
The implication for your team:
Stop assigning leads round-robin.
Score leads on personality signal (LinkedIn writing samples, post cadence, jargon density).
Route to the rep whose pattern matches.
Watch cycle time compress.
“You can remove a ton of psychological friction by literally just mirroring someone’s personality.” — Dan Baird
4. The new content moat is novelty + relevance, not volume.
The incremental cost of new content is effectively zero. Everyone knows this. The next move is not more content. It’s content that lands in the upper-right quadrant.
Dan demoed a 2-axis scatter plot on the episode:
Vertical = how often you (the sender) talk about a topic
Horizontal = how often the topic correlates with conversion
The trap most teams fall into: top-left quadrant. You talk about it a lot. Nobody converts on it. (Talking about yourself is the most common offender.)
The gold: bottom-right. Topics you’re not saying that would convert. Those are your novelty plays.
This week’s homework: list 5 topics your buyers actually convert on. Now list 5 topics you’re putting in cold emails. How much overlap? That delta is your conversion ceiling.
5. Stop being first. Be a fast second.
Dan has been building in AI for 9+ years. His take on Open Claw and other tip-of-spear releases:
“I’ll let other people make the mistakes at scale. AI is fantastic, but it allows you to make mistakes at scale, too.”
He’s not anti-innovation. He’s anti-paying-the-tuition-on-someone-else’s-bugs. The pattern: track new tech aggressively, integrate it 60–90 days after the early adopter horror stories surface. You get 90% of the upside with 10% of the risk.
If your team is currently first on every new model release, ask one question: what’s our scar-tissue budget? If you don’t have a number, you’re paying it whether you know it or not.
The bigger pattern
AI is splitting the GTM world into two camps. Camp One thinks AI is a productivity tool. Camp Two thinks AI is a decisioning layer. Camp One is buying more tools. Camp Two is building a moat.
The companies winning in 2026 aren’t sending more emails. They’re sending fewer, better ones, to people who actually want to hear the specific message they’re sending. That’s it. That’s the entire game.
Your AI stack is only as smart as the data it’s reasoning over. Fix that, and the rest gets easy.
The Personalization-vs-Relevance Audit
12 questions that tell you if your AI outreach is built to convert. Or just to send.
Why this exists
Every cold email in your inbox right now mentions where you went to college, what company you joined, or a post you wrote three weeks ago.
Dan Baird put it cleanly on the podcast:
“Personalization and relevance are not the same. Most teams personalize TO the recipient. Almost none of them adapt the message FOR the recipient.”
This 6-minute audit tells you which side of that line your outreach is on. Score yourself honestly. The score determines what you do next.
I also put together a Claude artifact so you can do the AI version instead of old school text version below. Access the Claude artifact here
How to score
For each question, give yourself a score of 0–3 based on how true the statement is for your current outbound, ABM, or AI-assisted outreach motion.
Score Meaning 0 We don’t do this at all 1 We do this sporadically or inconsistently 2 We do this for some segments but not as a system 3 This is operationalized across the team and measured
Maximum possible: 36 points. Add yours up at the end.
SECTION 1: Personalization basics (table stakes)
These four items are the bare minimum every modern outbound team should already be doing. If you score under 8 here, your outreach is being filtered out at the inbox layer before relevance even matters.
1. Every outbound email contains at least one signal specific to the recipient (recent post, role change, content they engaged with, event they attended). Score: _____ / 3
2. We A/B test subject lines on every campaign with at least two variants and statistical significance, not gut feel. Score: _____ / 3
3. Our outbound is segmented by buyer persona, not just by company size or industry. Score: _____ / 3
4. We have a documented brand voice that every AI-generated message has to pass through before sending. Score: _____ / 3
Section 1 subtotal: _____ / 12
SECTION 2: Relevance (the layer most teams skip)
This is where the real differentiation lives. Most teams stop after Section 1 and wonder why their reply rates are flat. The questions below are the ones that compound.
5. We have mapped what topics our highest-converting buyers talk about. Not just demographic data, but psychographic and behavioral signals. Score: _____ / 3
6. We can articulate the top 3 things our buyers want to hear that we are NOT currently saying in our outreach (Dan’s “bottom-right quadrant”). Score: _____ / 3
7. We adapt the style of our message based on the recipient’s communication preferences (more data-driven for analytical buyers, more social proof for relationship-driven buyers). Score: _____ / 3
8. We segment our messaging by adoption-curve stage (early adopters get jargon and complexity; late adopters get simplicity and social proof). Not just by demographics. Score: _____ / 3
Section 2 subtotal: _____ / 12
SECTION 3: Decisioning architecture (the moat)
This is the layer Dan spent half the podcast on. If your AI is an LLM with a prompt strapped to it, you are vulnerable to anyone who builds a real decisioning stack underneath it.
9. Our AI outreach tool is NOT making targeting and messaging decisions from an LLM alone. We have a machine learning, scoring, or statistical layer beneath it. Score: _____ / 3
10. We can audit and explain why our AI chose a specific angle, topic, or recipient. The decisioning is not a black box. Score: _____ / 3
11. We route leads to reps based on fit signals (personality match, communication style, vertical expertise). Not just round-robin or geo. Score: _____ / 3
12. We measure not just reply rate, but the downstream impact: meetings booked, opportunities created, closed-won, and average deal size by AI message variant. Score: _____ / 3
Section 3 subtotal: _____ / 12
YOUR TOTAL: _____ / 36
What your score means
30–36: You’re in the top 5%.
You already get it. Your team is treating AI as a decisioning layer, not a sending layer. Your work now is compounding the moat. Keep tightening the feedback loops between message, conversion, and ICP definition. Most companies will never reach your level. Use that.
22–29: You’re personalizing well, but relevance is your gap.
You’ve nailed the basics. You’re segmenting, testing, and signaling. But you’re still mostly adapting messages to make YOU look thoughtful, not adapting them to what the recipient actually wants to hear. The next 90 days should be focused on Sections 2 and 3. Build a relevance map for your top 3 ICPs. Pull your last 100 closed-won and tag by adoption curve. Stop guessing.
14–21: You’re sending. You’re not converting.
Most teams sit here. The work is real, the tools are bought, the emails are going out. But the math isn’t moving. The fix is almost never “more volume” or “another tool.” It’s pulling decisioning out of your LLM and into a real scoring or ML layer. Start with question 9 and 10. If you can’t audit it, you can’t optimize it.
0–13: Your outbound is on autopilot, and the plane is descending.
This is fixable, but it requires a different kind of investment. Not a tool. A system. Don’t buy another AI SDR until you’ve rebuilt your persona definitions from data and separated your decisioning layer from your generation layer. The good news: most of your competitors are also scoring here. The team that moves first wins the category.
The three moves to make this week, regardless of score
1. Pull your last 20 outbound emails. Count how many were “about the recipient” vs. how many were “adapted FOR the recipient.” If the ratio is worse than 50/50, you’ve found your priority for Q3.
2. Run the bottom-right quadrant exercise. List 5 topics your closed-won customers actually talk about. List 5 topics your team is putting in cold emails. Find the delta. That delta is your conversion ceiling.
3. Audit your decisioning layer. Ask the team running your AI outreach this one question: “When the AI picked this message angle, what was the reasoning?” If the answer is “the LLM decided,” you are reasoning from a black box. Add a scoring or ML layer underneath.
Next steps
If you scored under 22 and want help building the relevance and decisioning layers, that’s exactly what we work on inside GTM AI Academy.
Reply to the email this came in on with your score, and I’ll send you the specific frameworks we use for your range.
— Coach K GTM AI Academy