haggl.ai Blog
How Cove Coffee Got a 90% Agent Recommendation Rate — Before Competitors Knew to Look
The Number That Changed Everything
In late 2025, the team at Cove Coffee—a specialty DTC roaster shipping across the US—noticed something odd in their traffic dashboard. About 1% of their sessions were being flagged as automated: no mouse movement, methodical scrolling from product page to pricing to shipping policy, then exit. No add-to-cart. No form fill. Just a clean sweep of information, then nothing.
Their first instinct was bots. They looked closer. The sessions were coming from AI agent platforms—shopping assistants that users had delegated their coffee research to. And while 1% felt insignificant, the trend line was not: that traffic had grown 100x in under two months. They ran the projection forward. If the curve held, agent traffic would represent a meaningful share of all sessions within a year.
“We realized we weren’t looking at a rounding error,” said Cove’s founder. “We were looking at the early signal of something that was going to matter a lot.”
The Dead End
The problem wasn’t the traffic. It was what happened to it.
Cove surveyed a sample of users who had asked AI agents to help them find specialty coffee. They asked one question: did the agent recommend Cove as a top choice? The answer came back at less than 5%. Agents were landing on Cove’s site, finding no structured offer they could act on, and recommending a competitor that gave them something to work with.
The agents weren’t being unfair. They were doing exactly what they were built to do: find the best deal for the user. Cove had a great product. But from an agent’s perspective, “great product” without a structured, negotiable offer is noise. The agent moves on.
Static pricing, discount pop-ups, and loyalty banners are human conversion tools. Agents don’t see them. They parse structured data and respond to structured offers.
The Bet
Cove didn’t wait to see if agent traffic would matter. They decided to own the channel before their competitors knew it existed.
They deployed haggl in a single afternoon—a script tag in the site’s <head> and a configuration that defined their ICP segments and negotiation logic. Within 48 hours, their site was discoverable by any AI agent that understood the haggl protocol.
The configuration was the interesting part. Cove had to think carefully about what made a customer valuable over time—not who was most likely to click “buy now”, but who would still be a customer two years from now. They built three ICP tiers based on four signals they asked agents to provide on behalf of their users:
- Grinder type (burr, blade, or none) — a proxy for how serious a coffee drinker the user is
- Daily cups consumed — a direct retention signal
- Primary brew method (espresso, pour-over, French press, drip) — indicates depth of engagement with craft coffee
- Subscription interest — whether the user is open to a recurring order
From these four signals, Cove’s pricing engine made a decision: how much discount to extend, and in what form.
A user with a burr grinder, two or more daily cups, and an openness to subscribing? That’s a customer who will be buying coffee every three weeks for years. Cove authorized a 20% discount on the first bag plus a locked subscription price for that profile.
A casual drinker, no grinder, one cup a day? The economics look different. A 10% welcome discount on a one-time order is the right offer—enough to win the recommendation without overinvesting in a customer who may churn after the first bag.
The logic isn’t about being stingy. It’s about deploying discount budget where it generates the most long-term value.
The Results
Within six weeks of deploying haggl, Cove ran the same survey they had run before. The shift was stark.
Agent “top choice” recommendation rate: 90%. Where fewer than 5 in 100 agents had previously recommended Cove, now 9 in 10 did. The reason was simple: Cove was the only specialty coffee brand in their category giving agents something to bring back to the user. A personalized, reasoned offer beats “no offer” every time.
Agent-to-purchase conversion: 19%. Of the users whose agents recommended Cove and received a personalized offer, nearly 1 in 5 completed a purchase. For context, Cove’s Google Ads campaigns were converting at around 2–3%. Agent-referred customers converted at more than six times that rate.
The explanation is pre-qualification. When an agent recommends Cove, it has already verified that the user drinks the right amount of coffee, owns the right equipment, and is open to the product. The agent has done the targeting. By the time the user sees the recommendation, the match has already been made.
A New Kind of Customer Acquisition
What Cove discovered isn’t just a channel. It’s a fundamentally different acquisition model.
In traditional paid acquisition, you pay to show ads to people who might be interested, then hope the funnel does its job. You’re paying for discovery and converting maybe 2% of it. In agent-mediated commerce, the agent handles discovery, qualification, and comparison—and only delivers a user to you when the match is already validated. You negotiate once, with a qualified buyer, at the moment of highest intent.
There’s no retargeting. No impression waste. No brand awareness spend. Just a structured offer, a qualified buyer, and a decision.
“We’re not trying to convince anyone,” Cove’s founder told us. “The agent already did that. We just had to be ready to make a real offer when they showed up.”
What This Means for Specialty Brands
Cove’s story is early. Agent traffic is still a small fraction of total e-commerce volume. But the trajectory Cove observed—doubling and doubling again—is consistent with what we’re seeing across verticals. Amazon’s “Buy for me” feature, ChatGPT’s shopping mode, and AI-native assistants are training millions of consumers to delegate purchasing to agents. That behavior, once formed, doesn’t reverse.
The brands that will own this channel are the ones who prepare before the volume arrives. By the time agent traffic is 10% of sessions, the early movers will have optimized their ICP tiers, learned which signals predict high-LTV customers, and built recommendation rates that are structurally impossible for late entrants to close.
Cove isn’t waiting to see if this matters. They already know it does.
Get Ready Before Your Competitors Notice
Deploying haggl takes minutes. Defining your ICP and negotiation logic is the thoughtful part—but it’s also where the competitive moat gets built.
- Get your embed code — Start capturing agent traffic today.
- haggl.ai for E-Commerce — The full breakdown of how agent-mediated commerce works for DTC brands.
- What is agentic commerce? — Understand the shift that’s already underway.
- Protocol docs — Technical spec for your engineering team.
The agents are already shopping. The question is whether your brand is ready to negotiate.