When JB Daguené started his sales career in 2007, signed contracts arrived by fax. Salespeople would sprint across the room every time the machine rattled, hoping it was their closed deal. Most of the time, it was a cold fax. Marketing spam. Delivered by a machine that shook.
In a recent podcast episode, Toni Hohlbein sat down with JB to trace the line from that fax machine to today, where some sellers brag about needing 730 emails to generate a single positive reply. The conversation became a sharp, honest autopsy of how an entire profession gradually buried itself — and what the path forward actually looks like.
Where it started: printed research and peer-to-peer calls
JB's first real sales role was at Trustpilot, where he was the first sales hire. The company went from 50 to 5,000 customers in two years. JB personally closed 600 of them — roughly one per day.
The numbers are striking, but JB was quick to add context. He was working 10-to-12-hour days, driven by eight years of personal debt from a startup that had gone bankrupt. An uncapped commission model meant he could pay it off faster. But the real reason behind the numbers, he told Toni, wasn't just work ethic. It was curiosity.
Every morning, JB would print stacks of lead research — social media profiles, online reputation, conversion funnels. The Salesforce admin hated him for it. But when he picked up the phone, the first 20 seconds of every call were grounded in what he'd actually learned about the person and their business. Within two minutes, JB said, the dynamic shifted from seller-buyer to peer-to-peer.
Toni pushed back on this. In his experience running sales teams, research was often an excuse to delay picking up the phone. Reps would spend five hours "preparing" and then finally make a call when the day was almost over. How much research do you really need if you're selling one product to one persona?
JB agreed there's a balance. But he made a distinction: the research wasn't about thoroughness for its own sake. It was about confidence. When you know something real about the person you're calling, you sound different. And that's what ultimately led Evergrowth's first business model — outsourcing research teams so reps wouldn't have to choose between preparation and productivity.
The Predictable Revenue era: brilliant, then dangerous
Both Toni and JB pointed to Aaron Ross's Predictable Revenue as the inflection point that shaped modern B2B sales. JB called Ross "the godfather of the SDR" for establishing the playbook: specialized roles, cold email at scale, one KPI per function.
Toni shared his own memory of the book's impact. His CEO read it overnight and walked into the office the next morning saying, "That's what we're doing now." What started as writing a few emails quickly evolved into layering cold calls on top — a dual-channel approach that worked extremely well for years.
Two things made email effective in that era, JB explained. First, data was still hard to get. If you had ways to gather contact information your competitors didn't, you had an edge. Second, people actually read their emails. Getting a cold email felt novel. You'd open it and think someone had put thought into reaching out to you specifically.
Then data became cheap. And when that happened, shortcuts followed.
If you actually read the book, your ICP is far more granular than industry, company size, and region. But those were the only three filters available in most databases, so that's what people used.
JB made a point that stuck: if you actually read the book, your ICP is far more granular than industry, company size, and region. But those were the only three filters available in most databases, so that's what people used. When Evergrowth's consulting team cleaned lists manually, only about 30% of contacts actually fit the ICP. The other 70% were effectively spam targets. The industry just didn't call it that yet.
Button clickers and the COVID acceleration
COVID was the turning point where things got noticeably worse, JB argued. People started working from home. Direct phone numbers weren't widely available from data vendors yet. So teams doubled down on email — not strategically, but desperately.
At the same time, money was cheap. Companies were still buying. Hiring continued. You didn't have to work particularly hard and results still came in. It was, as JB described it, a very strange period.
What happened in parallel was more damaging. A generation of salespeople became what JB bluntly called "button clickers." Ask them how they sell and the answer is a sequence of tools: "I use Outreach, then Apollo, then I copy a template and press buttons every day."
Toni recalled a similar moment. When his team rolled out SalesLoft, he asked his sales ops lead a simple question: what's stopping a rep from hitting "select all" on their list and clicking send? Each email counted as one activity. At that point, he said, he wasn't sure you could call it sales anymore.
JB noted that the teams behind tools like Outreach weren't promoting this behavior. They positioned sequencing as a productivity and pipeline management tool, not a mass email blaster. But the market used it differently. Most teams weren't running micro-campaigns of 20 to 40 people with carefully segmented messaging. They were automating at volume, and the quality collapsed.
AI: more damage than good (so far)
When usable LLMs hit the market in late 2023, JB said, the sales profession got worse, not better.
The technology enabled people to generate email copy faster, spin up domains, warm up inboxes, and send at a scale that was previously impossible. AI-written "personalization" became a party trick — poems referencing where a prospect went to high school, or icebreakers about the local football club. JB called it ridiculous. Buyers can see exactly what's happening.
Toni raised a question about why the AI-powered spam approach seemed to saturate faster than earlier playbooks. The basic SDR email play worked for 10 to 15 years. AI-generated outreach seems to have hit diminishing returns in a fraction of that time.
AI didn't create a new playbook. It just accelerated the decline of an old one. The industry was already scraping the bottom of the original approach. AI turned the scraping speed up to maximum.
The numbers tell the story. When Evergrowth ran manually researched micro-campaigns, connection rates exceeded 20%. Now the market celebrates 3% reply rates — and tools market "20% positive reply rate," which is 20% of that 3%. JB posed the obvious question: why not study the tiny fraction that actually responds and build similar lists, instead of repeating the same approach to 97% who don't care?
Google steps in, and the inbox war is lost
Toni brought up Google's recent crackdown on warmed-up inboxes — the infrastructure that enabled millions of cold emails to be sent from hundreds of disposable Gmail accounts. Microsoft had already been enforcing strict firewalls. Google was catching up.
JB framed it simply: Google and Microsoft have to protect their users. If they don't, people will abandon email as a product entirely. The operators spinning up disposable domains and mailboxes to fight against the email providers are fighting a war they can't win.
The deeper problem, JB said, is the mindset behind it. The people building these systems think sales is about sending emails. They think buyers are just an email address attached to three data points — first name, country, company size. But buyers don't introduce themselves that way. They talk about what they're working on, what challenges they face, what they care about. You can't reach them with scraped fields and rotated domains.
The phone is back
Both Toni and JB agreed that the phone is experiencing a resurgence. But there's a generational gap to close.
JB's youngest brother, 11 years his junior, doesn't even have the call button visible on his phone. Why would he call anyone? And that's the generation now entering sales roles. They don't lack motivation — they simply never learned how to use the phone as a communication channel.
Training is the answer, JB said, and it's 100% doable. But the phone also raises the stakes on preparation. A research-based opener in the first 20 seconds changes the entire call dynamic. JB described recordings from his own team where prospects visibly drop their guard because the caller clearly knows something real about their business.
At Evergrowth, agents generate the research and turn it into contextualized call openers calibrated around the prospect's specific pain. It's the same approach JB used at Trustpilot with printed research stacks — except now 25 digital colleagues handle the preparation instead of one person and a printer.
What Clay does well, and where it falls short
Toni asked about Clay's role in the current landscape. JB was respectful but direct.
Clay is an excellent enrichment tool that consolidates what used to require three or four separate products — LinkedIn data, company data, Crunchbase data — into one interface. The community-led growth strategy behind it is genuinely impressive.
But most Clay usage, JB observed, still powers the same spam cannon playbook. Agencies use it to build massive outreach lists faster and cheaper. The fundamental approach hasn't changed.
And there's a structural issue. GTM teams already suffer from too many silos — a legacy of the Predictable Revenue model's specialization. Adding a tool that only one person on the team can operate creates another bottleneck. That person becomes the gatekeeper to AI. Everyone else defaults to ChatGPT. Now you have two disconnected AI experiences inside the same organization.
What "agentic" actually means (without the hype)
JB acknowledged the hype. Gartner's latest cycle puts "agentic" and "agents" at the very peak. The term has become meaningless — people call a custom GPT an agent.
Evergrowth's definition is deliberately simple: digital colleagues.
Remove the word "digital." A colleague is someone your whole team can access, work with, and collaborate alongside. That colleague shares the same institutional knowledge as every other colleague. They don't operate in a silo.
JB broke it down into three components: an AI layer, a set of tools the agent can use (triggering other agents, writing to a CRM, drafting messages), and a degree of autonomy in how it uses those tools. Not a rigid if-then workflow. More like giving a competent person tools and context, and trusting them to make decisions.
The AI has capabilities, it has tools available, and it has the flexibility to decide when to use them — not a rigid script, but judgment. Like a real coworker with a toolbox on the table.
At Evergrowth, the internal sales team consists of two people and 42 agents. On any given day, they might work with 17 or 25 of them, depending on the workflow.
Minimum Trustable Product: why MVP doesn't work for AI
JB shared a hard-won lesson from building an AI product after years in the SaaS consulting world. The traditional MVP approach — ship something minimal, iterate based on feedback — breaks down with AI.
The reason: AI can be 99% smarter than you. But the 1% where it says something wrong becomes the only thing users remember. Trust evaporates instantly. Especially with salespeople, who JB described as the most opinionated users imaginable — and the ones whose opinion matters most because they bring in the revenue.
The solution is what JB calls the Minimum Trustable Product, and it starts with the training center. This is where teams teach agents their value proposition, ICPs, and personas. Once trained, agents retain 100% of that knowledge, every time. Humans, by comparison, retain about 20% of what they're trained on.
Critically, all agents need to share the same training data. If one agent researches based on one understanding of the ICP and another writes copy based on a different understanding, the output is incoherent. Shared context is the foundation.
Hallucination is a prompting problem
Toni asked about hallucinations — a topic that comes up constantly in AI conversations. JB's take was blunt: hallucination is bad prompting.
LLMs are designed to please you. If you ask a question and the answer doesn't exist in any available data, the model will still produce something — because you didn't give it a fallback instruction.
The fix, JB said, is embarrassingly simple. Every prompt should include: "If you don't find the information, reply with 'I don't know.'" Every agent at Evergrowth has a fallback built into its prompt by design. Hallucination isn't an unsolved mystery. It's an engineering discipline.
Toni added a complementary angle from his own experience building in the AI space: much of what people call hallucination is actually the model guessing at internal definitions — a metric, a term, a workflow. It guesses correctly nine out of ten times, so you trust it. Then the tenth time it guesses the other way and you call it hallucination. The real fix is being deliberate about what you define and feed the model, which maps directly to JB's point about prompting with fallbacks.
Why 95% of AI initiatives fail (and what the other 5% do differently)
JB referenced a recent MIT finding that 95% of AI initiatives fail. Evergrowth, running GTM pilots since November 2023, now sees the opposite — a 95% success rate. The difference comes down to two hard disqualification criteria.
First: you need a process to digitalize. If you want digital colleagues, you need documented workflows for them to follow. It's the same principle as hiring human employees — drop them into an organization with no process documentation and they'll flounder. AI is no different.
Second: AI is not software. Too many teams treat AI implementation the way they'd treat a HubSpot rollout. Configure it, build a dashboard, let it run. That's not how digital colleagues work. You train them, build workflows with them, and teach your team how to collaborate alongside them.
The companies making the shift from "flip a switch" to "build a team" are the ones seeing returns. There's dedication required: training, coaching, teaching.
Context is the new competitive edge
JB closed with what he sees as the fundamental shift happening right now.
Three years ago, data was gold. The biggest database won. That era is over. Data is a commodity — every vendor has the same company information, the same contact fields, the same funding rounds.
What replaces data is context. And context is architectural, not transactional.
One agent researches a company and builds an understanding of their situation. It passes that context to a second agent that qualifies the account. That agent passes richer context to a third that finds the right contacts. Then a fourth agent writes outreach specific to that person's role, seniority, and communication style — grounded in everything the previous agents discovered.
Each agent builds on what came before. The context window grows. The output gets sharper. That's the architecture that replaces the spreadsheet — not more data columns, but more understanding.
The next few years aren't about who has the most data. They're about who builds the best context.
This article is based on a podcast conversation between Toni Hohlbein and JB Daguené. To learn more about Evergrowth's Agentic GTM Workspace, visit evergrowth.com.
