How AI Agents Are Powering the Next Generation of Outbound

71% of B2B decision-makers ignore cold emails because they lack relevance, 43% say the emails feel impersonal, and 36% don't trust the sender.* Reply rates keep falling with them. Around 6.8% in 2023, closer to 5% in 2025, and tracking lower again in 2026.* Outbound email volume has never been higher, but the volume of conversations it produces has never been lower.

The first wave of AI didn't fix this. It scaled the noise.

The next wave is different. A Series B fintech platform we work with produced 49 outbound MQLs (meaning conversations actually started with a buyer, not form fills) in their first month using a new approach to finding in-market accounts. That number isn't possible with the old playbook. It's only possible when AI handles the work humans were never good at, and humans handle the conversations that actually close deals.

Why outbound stopped working

Buyers see right through generic outreach now. They get 15 cold emails a day, and most of them sound the same: "Saw your team is hiring." "Noticed you're scaling." "Quick question on your [department] growth."

Three vendors hit the same target account on the same Tuesday with the same intent signal. Three subject lines so close they could've been A/B tests of each other. The buyer deletes all three in one swipe and gets on with their day.

This is a structural problem, not a copywriting one.

The shared intent data trap

Nobody fixes this problem by sending better email. The intent data your competitors are buying is the same intent data you're buying. Bombora and ZoomInfo (the two biggest intent-data providers in B2B) aren't running secret feeds for you. They sell overlapping signals to every vendor in your category, often at near-identical price points.

Think of it like the Taylor Swift Eras Tour ticket queue. Everyone's in the line. Nobody's moving.

That's not a signal. It's a shared queue.

If every competitor can see the same intent surge at the same time, the surge isn't worth seeing. We dug into this in how top B2B teams are using buyer intelligence. The teams winning today have stopped relying on data anyone with a credit card can buy.

Signals built for you, not sold to everyone

The fix is signals nobody else has. We call these unique-to-you (or U2U) signals.

A U2U signal is a buying trigger built for one business, based on behaviour competitors can't see.

Generic intent data tells you a company's people are reading about your category. So is everyone else's. A U2U signal tells you something more specific: a target account just posted two senior FP&A (financial planning & analysis) roles in a quarter, and their tech stack suggests they're consolidating ERP (enterprise resource planning) systems. That combination only matters if you sell to finance teams in transition. So it only fires for you, not for every vendor in the category.

Get to the right account, with the right context, before anyone else has noticed. Bring a human in only when there's a real conversation to have.

The hard part is that no human team can monitor enough signals across enough accounts to make this work at scale. That's where AI agents come in.

What AI agents actually are

If you've used ChatGPT, you've used an AI model. An AI agent is a different animal.

A model waits for you to prompt it. You type a question, it gives you an answer, you move on.

An agent doesn't wait. It runs on its own schedule. It watches for triggers, makes a call when something is worth acting on, takes the action, and pays attention to what happened after so it can do better next time. One waits to be used. The other is already working while you're in your morning standup.

Most "AI for sales" tools on the market right now still need a human to push the button. Agents don't.

What an agent can do that a person can't

A human SDR can research one prospect deeply, or send a hundred emails shallowly. Pick one. Agents do both at the same time, on every account, all the time. They surface what matters and drop it on a human's desk when it's ready for a conversation.

Why Finding the Value First Changes Everything

Most AI-powered outbound fails for the same reason old-school cold email did. The message still lands as an interruption. "Hey, we noticed X. Book a call?" Doesn't matter how good the trigger was. The prospect sees another vendor asking for their time before earning the right to ask for it.

The signal was good, the execution was fine, but the problem is what was at the centre of it.

So before a single message goes out across any channel, we stop and ask: what can we actually give this person?

A fintech client we worked with offered a 30-minute session showing prospects exactly how their sales velocity compared to similar-stage companies in their category. No pitch. Just a number the prospect didn't have and suddenly wanted to talk about. That was the thing of value, the meeting came after.

Another client ran a proof-of-concept engagement at a reduced rate for the first two accounts to commit before a set date. Real scarcity, because they only had capacity for two.

And sometimes it's just knowledge. A short briefing built from what the signal told us about their situation. "We track companies going through X transition. Here's what we've seen work. Happy to share the full version if it's useful."

Three completely different offers with the same logic underneath,  give something worth having before you ask for anything.

The meeting is a natural consequence of someone getting value and wanting more. 

That's the whole model.

Instead of five touchpoints all built around "book a call," you have five touchpoints all built around the same piece of value. The postcard references it, the LinkedIn message deepens it, the microsite unpacks it, and the ad follows the buyer wherever they go. By the time the rep picks up the phone, the prospect isn't being cold-called, they're being followed up on something they already found interesting.

Agentic outbound only works if the thing at the centre of it is worth something. Otherwise you're just automating noise.

How Punch! does it: Agentic GTM

Now, along with genuine value to offer, you have the two ingredients: signals competitors can't see, and agents that can act on them around the clock. Agentic GTM is what happens when you put it all together.

At Punch!, we run five specialist agents, coordinated by one orchestrator. One agent can run a channel on its own. The full team can run your whole engagement layer.

Meet the agents:

• ORBIT, the orchestrator. Decides which channel fires, when, and why.
• SPARK, signal-triggered email outreach.
• LINK, LinkedIn engagement triggered when the buyer's actually paying attention.
• PEN, handwritten postcards. Drafted by AI, penned by a robot holding a real pen, physically delivered.
• REACH, person-level ads that target a named buyer, not a company.
• PAGE, bespoke 1:1 microsites built around the specific signal that fired for that prospect.

Five specialists and an orchestrator, all running off one intelligence layer, so the messages never contradict each other across channels.

Does this replace the tools you already have?

Short answer, no.

Your CRM stays. Your outreach platform stays. The agents sit on top, watching for the signals your existing stack can't see, and firing the right channel when the moment comes. The signal layer is what's new. The execution mostly runs through tools you already own.

That matters because nobody has the appetite for another six-month rip-and-replace project. This isn't one.

What that looks like in practice

A U2U signal fires on an account our client has been watching for a quarter. ORBIT reads the context (the account, the trigger, the role, the prior engagement) and routes it. SPARK drafts the email. PEN queues the postcard for later in the week. REACH starts serving 1:1 ads to the three named buyers inside the account.

By the time the human SDR picks up the phone, the account has had four touchpoints, all carrying the same context, none of them generic. The rep isn't cold-calling. They're following up.

Every touchpoint leads with value: a benchmark against peers, a short read on their current setup, or a recommendation they can act on. Something useful, given before anything is asked. The CTA is soft. "Happy to share what we're seeing across similar companies." Not "Book a 15-minute meeting."

Why AI alone doesn't close

Most of the "fully autonomous AI SDR" tools flooding LinkedIn right now (Artisan, 11x, the rest of the breed) are basically Mailchimp with a personality. The signal layer is the moat. Everything else is commodity.

Here's the simple split:

The work a good SDR does well (live qualification, navigating four stakeholders inside an account, the judgment to read when a call is going sideways) is the work AI does worst.

The work AI does well is the work the SDR hated anyway. List building. Signal monitoring. First-touch outreach. The follow-up admin that eats their afternoons.

Agents surface and engage. Humans convert. Every call lands with full context, so reps stop wasting the first ten minutes on discovery questions and start the conversations that actually progress. As we wrote in why AI isn't coming for your sales job, the reps leaning into this don't lose value. They gain it.

Does this actually work? The results so far

Here are real numbers from clients running a U2U signal model with agentic outreach plus human SDRs (anonymised because the client logos aren't ours to share):

• A Series B fintech platform: 49 outbound MQLs in month one. 314 in the first six months. Conversations started with companies like Boeing, DuPont, and BBVA.
• An interactive content SaaS company: 736 outbound MQLs over 16 months. A sustained average of 46 per month. Conversations with the likes of IONOS, Stanford, and TikTok.
• A digital adoption platform: 313 outbound MQLs in 12 months, with 30% coming from enterprise expansion campaigns. Conversations with Shell, Roche, Google, and the NHS.
• A content marketing platform: 216 outbound MQLs in six months. 36 per month sustained. Conversations with Etsy, Intuit, Glassdoor, and Audi.

The word "MQL" gets thrown around loosely, so worth a flag: in this context, an outbound MQL is a positive response to an outbound campaign where the contact confirmed they want something sent and a conversation has started. Not "they downloaded a whitepaper." Not "they opened the email twice." A real conversation, started.

How to think about adopting this

Most teams adopt agentic outbound wrong. They buy a tool, upload a list, and call it transformation. It isn't.

Start with the signal, not the channel

(I know, I know. Every vendor demo starts with the channel. Look at our beautiful email composer! Look at our AI-generated subject lines! Cute. Now show me your signal layer.)

The question that matters is this: what specific behaviours indicate someone is in-market for what we sell? Not "they downloaded a whitepaper." Not "they're in the ICP firmographic." What are they doing that competitors can't see?

Once you have the signal layer right, the channels are easy. The agents know what to say because the signal told them why they're reaching out.

Treat the signal layer like infrastructure

The teams getting compounding returns from this treat signals the way a product team treats data pipelines, or a finance team treats forecasting models. Someone owns the work. Budget sits behind it. New signals get tested, refined, and rolled out every quarter.

It's infrastructure, not a feature.

Channels and tools sit downstream of that. Teams that bolt AI onto a campaign-based model get a brief sugar high in activity, then a return to flat results. Teams that rebuild around continuous, signal-led pipeline see the kind of step-change we explored in using predictive analytics to close B2B deals twice as fast.

What to measure 

Most teams are measuring the wrong things. We've written about how most teams are measuring outbound completely wrong; short version, dials per day is exhibit A.

The right metrics measure whether the system is producing meetings that progress and convert, not whether your reps hit the activity number their plan said they should.

The five numbers worth watching:

• Cost-per-qualified-meeting. Should fall quarter over quarter as the system learns.
• Signal-to-meeting conversion rate. Proves the signal layer is working.
• Pipeline coverage from signal-triggered accounts. Isolates what the agents are actually contributing.
• Future pipeline indicator. Accounts engaged across "not right now" cycles that re-enter when triggers fire.
• AE meeting acceptance rate. Proves the meetings reaching reps are worth their time.

Cut open rates, isolated reply rates, and sequences enrolled from the executive view. They flatter teams running broken systems.

What this means for your team

Three things change when you move to this model:

1. The senior SDR role gets more valuable, the entry-level SDR role goes away. Every conversation now starts with real context, so the people who can hold those conversations matter more than the people who used to build lists by hand. (We mapped this in detail in how to become an SDR in 2025.)
2. The marketing-sales fight over MQL definitions stops mattering. Marketing owns signal definition. Sales owns conversion. Both operate from the same intelligence layer.
3. The CRM stops being the centre of your GTM stack. It becomes a system of record sitting underneath whatever's actually running the signals.

This is uncomfortable for teams whose org chart and budget were built around the old model. Founders and revenue leaders who delay the shift will spend 2026 funding higher headcount to produce flatter pipeline. Their leaner competitors will be compounding an advantage they can't catch up to without rebuilding from scratch.

Should you invest in this now?

What is your current outbound system producing per pound spent over 12 months, and what would the same investment produce inside a system built around unique signals, agentic execution, and human conversion?

Another year funding flat pipeline buys you a slightly bigger version of the same problem. A year building a system that compounds buys you a category lead your competitors can't shop their way out of, and the runway to widen it before the rest of the market catches on.

Agentic by Nature. Human by Choice.

Punch! is The Human-Agentic GTM Partner. We use unique-to-you buying signals to identify in-market accounts and generate pipeline. Activated by AI. Converted by humans. Book a call to see Agentic GTM in action.