Full webinar recording: Mert, Cody & the Agents.
Let's start with the results
Before showing you how any of this works, I want to show you what it produces. Because I know a lot of you have been experimenting with AI for the past year and had mixed results. Fair enough. Here's what 3.5 months of data looks like for us, and keep in mind this was over the summer.
The industry average for connect and reply rates right now sits somewhere between 2 and 5%. We're almost 10x above that on average, across every persona we target. And we're doing this at a scale of about 300 new contacts per week, all with automated research, automated cherry-picking, and automated drafting of cold outreach that Cody then uses to get in front of these people.
What's also important if you want to be truly customer-centric is to not just look at persona performance but also at segment performance. We target three types of companies: B2B software, B2B manufacturing, and B2B services. We're strongest in B2B software for positive replies and meetings booked. Manufacturing has great connect rates but more negatives. We're still iterating on that part of the playbook. But what I like about targeting B2B manufacturing is that unlike software, it's not as noisy in terms of outbound competition.
Why most AI setups create more silos, not fewer
A lot of the AI workflows you see shared on LinkedIn look like this: one GTM engineer is in charge of all the AI automations. They push enrichment into the CRM. The salespeople wait for it. They get tired of waiting. They start using ChatGPT on the side. And now you have three different tools the team is using in three different silos.
Our approach is different. The RevOps person is the architect of the entire agent orchestration. They create agents that live in a shared workspace. The salespeople can access those same agents directly because the agents update the CRM automatically. You have one workspace, one set of digital colleagues, no matter which part of the team you're on.
RevOps builds workflows and schedules them. Salespeople work with the agents one-to-one. But it's the same digital colleagues.
A lot of companies are failing their AI experimentation because they treat AI like a software. If you're onboarding interns, you have to train them. It's exactly the same with agents. That's why you have a training center. That's the approach to a successful pilot.
What the workspace looks like
When you set up a GTM workspace, you start with the training center. If you want your AI agents to be true digital colleagues, they need to know as much as your salespeople. Maybe more. You load in your value proposition, your product offering, and ideally your customer stories. Customer stories are the best training content because they're already contextualized. AI agents are really good at understanding contextualized customer stories.
Then you define your segments. You can work with ICP verticals, key accounts, or both. You assign personas to each segment. Some overlap, some are very niche. And then you bundle the agents into three use cases.
1. Automated lead and signal research
The Account Qualification agents can browse the internet, open websites, download PDFs, and make conclusions about whether a company matches your ICP criteria. For us, we check if a company has a dedicated sales team, if they sell a complex product that requires meetings, if they have clear industry segmentation on their website, if they're B2B, and if they're not a competitor. Every single check collects context that gets used later for personalization.
The Account Research agents go deeper. They look for custom signals: job ads (and read the actual job descriptions), annual reports, YouTube interviews, podcast transcripts, conference appearances. Based on what comes back, you get a custom score of all the positive signals. That score drives prioritization.
The Contact Finder agent is changing how prospecting works. Instead of paying for a license to filter 600 million contacts by job title, the Contact Finder understands your persona and searches for people who match it. It searches in multiple languages. It finds people who aren't on LinkedIn, like clinic directors, sports club managers, or home service owners. And it has access to 20+ APIs for finding email and phone numbers through waterfall enrichment, so you only pay when enrichment succeeds.
The Contact Qualification agent solves a different problem: 8 to 12% of the contacts in your CRM lose accuracy every month because people change jobs. This agent checks if contacts are still at the right company, flags promotions, and if someone moved, it passes the new company back to the qualification agents. You keep your CRM clean and you find warm leads in the process.
2. Personalized customer interactions
The Account Planning agent takes all the research and produces a full account plan in under two minutes. If you target large enterprise, you know how long this takes manually.
The Play Copywriting agent is where the research turns into outreach. We use it for two things: research-based cold call openers and research-based cold emails. Cody books 60 to 70% of his meetings through cold calling, and he has a personalized talk track in his dialer for every single call. People drop their guard because they can tell he's done his homework.
The Digital Twin agent is a different kind of agent. You chat with it. It has access to all the research and all the training center content. If you need to prepare for a meeting quickly, you can get ready in seconds instead of spending 15 to 20 minutes researching across the internet.
Data-driven outreach vs. context-driven outreach
For years, we enriched companies with data from LinkedIn, Crunchbase, BuiltWith, and others. We ran hundreds of micro campaigns with 12 to 40 people each, using static templates with data variables. I was a huge advocate of that playbook. I made a lot of revenue with it.
But the problem is: even when you're granular about it, you still end up with visible data variables. And my inbox is proof. I think my name is Tracy in some database, because I keep getting emails addressed to Tracy. Sometimes Juliana.
If you just put AI on top of data points, it's going to fail. AI cannot personalize on data points. It needs context.
The difference is that when agents do qualification, they don't just return yes or no. They collect the context from the research. The account research agents collect custom signals and produce a score. And then you have two types of plays you can write from this.
Evergreen plays (qualification-based)
These use the rich context from qualification to personalize outreach even when the account research score is low. More than 60% of our meetings come from evergreen plays. They're always runnable, fast to scale, and produce consistent, predictable messaging. The only downside: no strong "why now."
Research-based plays (signal-based)
These are built around a specific, recent signal surfaced by account research. A product launch, a new territory expansion, a conference appearance. They get stronger replies because there's a clear "why now" that ties signal to pain to value. The downside: they're not predictable. You can't control how high the score will be before you research a company. They're more like a bonus.
How Mert runs the RevOps workflows
Mert is our founding solution engineer. He was previously head of RevOps and recently got promoted because, well, what made us successful with customers is exactly what he does internally. He runs four workflows on a weekly and monthly basis that make sure Cody never runs out of leads.
Workflow 1: Newbiz
Mert builds a broad list from the LinkedIn company database. Something like "UK, 50-500 employees, information technology." He pushes it into the newbiz workflow, and the agents take over: qualify against ICP, research signals, find contacts, qualify contacts, write plays, push everything to the CRM and sequencing tool. From 604 accounts in one recent run, 63% qualified as ICP. That ratio tells him the list quality was good.
Workflow 2: Recycling
Dormant accounts are accounts with no open opportunity, no activity for 90 days, and no open tasks. Every first of the month, this workflow runs. It checks if the contact still works at the company, refreshes the research, updates the account plan, and generates new messaging. If the contact left, it finds a new one. Even if Cody contacted them three months ago, they get something relevant because the research is fresh.
Workflow 3: Signal monitoring
This one runs on the second week of every month. Mert marks certain contacts as "nurture" and the workflow monitors them for new signals. Same process as recycling: refresh research, check employment, generate messaging. He also runs about 3,000 key decision makers through a contact qualification check mid-month to keep the CRM data accurate. If the data is wrong, agents correct it automatically.
Workflow 4: Post-event
When the team comes back from an event with a lead list, Mert pushes it into this workflow. It qualifies accounts, does the research, checks if attendees have email and phone (if not, it finds them), and creates event-specific messaging. What might take more than a day manually takes a couple of hours. Cody can follow up on event leads within the same week.
The beauty of this is that I've automated the entire top-of-funnel process. Within a couple of clicks, I can support Cody on a weekly basis with new leads. He won't run out. And I'm not a bottleneck.
How Cody uses the plays to book meetings
Cody's day doesn't start with prospecting. Evergrowth has already done that. His day doesn't start with deep research on accounts. Evergrowth has already done that too. His day starts with filtering prospects by tags, adding people to sequences, and working through his tasks.
He showed four types of plays in the webinar.
Evergreen cold call track
The Play Copywriting agent reads a prompt that weaves together the company's industries, their segmentation, their use cases, and the persona's role. The output is a talk track with a research-based opener, talking points, and calibrated pain questions. Cody doesn't read it on the call. He reads it before the call so he sounds like he's done his homework. And he has. The agent did it for him.
Evergreen cold email
Same idea but for email. The agent identifies the company's business model, the industries they target, and even the CRM they use. Not because we're writing an email about Salesforce, but because knowing they use Salesforce means the agent can paint a picture of exactly how our product would work with their system.
Research-based cold email
When the account research score is high enough, the agent writes an email built around a specific signal. A product launch. A new territory. A conference sponsorship. It doesn't just call out the research. It connects the signal to a specific pain and then explains how we can help. Short, detached CTA at the end.
Research-based cold call track
Same structure as the evergreen track but leading with a permission-based intro that references the specific research. "I did some research and found out you recently raised $56M in Series C, expanded your sales team, and launched a new product." Then it goes into calibrated questions based on the persona's frustrations and the account's research. Every contact gets a different framework because the pains differ by persona.
Cody didn't write the play prompts himself. He used another agent to draft the full prompt, then tweaked the final details. That's how we do the fine-tuning of play agents. You don't need to be a prompt engineer. You need to know what good outreach sounds like.
The takeaway
If you're a good salesperson and you do deep research, you actually disqualify more than you qualify. That's exactly how you want to use agents. Cherry-pick the best companies and the best contacts because you can. I'd rather work with a list where 5% are ICP-fit and I'm 100% confident in them than a list where I have a lot of contacts filtered on random data points.
Mert automates the top of the funnel so he's not a bottleneck. Everything syncs to the CRM and the outreach platform. Cody works from the output every day without ever worrying about where his next leads are coming from. And the agents? They remember 100% of their training. They work in the background. They don't get tired.
That's how a team of 16 people and 87 agents runs a pipeline that most companies would need 3x the headcount to manage.