AI marketing & sales lessons from Webflow’s CRO

A lot of execs talk the talk when it comes to adopting AI, but few actually walk the walk. Legacy platforms, siloed tools, and the sheer drain on developer time make it feel easier to wait on the sidelines than to dive in.

But Adrian Rosenkranz, Webflow’s CRO, took a different approach. Zapier’s Head of Product Marketing, Angela Ferrante, sat down with him to see what it looks like when leadership gets hands-on.

From building AI-powered agents that think like him to orchestrating workflows across sales and marketing, he shared how every experiment has a real impact when you tie it into business outcomes. And to help you do the same, we’ve included three of his templates (using Zapier Agents) that you can use to get started today.

Skip ahead

1. Provide AI with context and frameworks

Adrian’s first insight was simple but powerful: AI doesn’t know how you think. Left on its own, it will give you generic answers that don’t reflect your strategy, language, or priorities. 

To make it useful, you have to teach it the context you’d normally give a new team member. That might be how you define success, what patterns you look for, and how you like information presented. Adrian did this by feeding in his own notes and insights from books, frameworks, and interpretations, so the AI wasn’t just pulling from the internet—it was drawing from his way of seeing problems. The result? Instead of surface-level summaries, it started producing insights that felt aligned with how he actually runs the business.

One of the clearest examples is how he reimagined deal reviews. For Adrian, the most valuable data after product usage is the conversations reps are having with customers. But reviewing those transcripts call by call just isn’t scalable. To solve that, he leaned heavily on MedPic, a sales framework he already uses to evaluate opportunities. 

By teaching AI to look for the same components he would (metrics, decision-makers, decision process, pain points, and champions), Adrian created a workflow that automatically flags the risks and next steps he’d normally surface in a review. 

That means his thought processes now scale across the whole team, turning messy transcript data into a consistent playbook for improving win rates.

Steal the agent

How the agent works:

• Retrieves new sales call transcripts from Gong

• Analyzes the transcript using the MedPic sales framework to identify risks, metrics, timelines, pain points, and advocates

• Updates the opportunity in Salesforce with that analysis

• Creates a summary of the call analysis

• Sends the summary to the relevant Slack channel

2. Avoid efficiency traps by tying output to KPIs

Adrian cautioned that AI adoption isn’t automatically a win just because it saves time. In his words, it can create an “efficiency trap”—where more is just more and not necessarily better. The real challenge, he argued, is knowing when AI actually enhances work versus when it just adds volume without value.

One way to spot the difference is by tying AI outputs to existing KPIs. If a workflow helps improve win rates, response rates, or customer outcomes, it’s an enhancement. If it just makes you do more of the same without impact, it’s a trap.

Take email as an example. Adrian sends a huge volume of messages, but like most execs, he doesn’t have hours to rewrite every draft until it sounds right. And while AI can help, it often flattens the tone into bland corporate speak, which rarely drives replies. 

To avoid that trap, Adrian designed a reinforcing workflow: one that constantly scans his tone, compares it against his real voice, and learns from how people respond. That way, the metric he’s watching isn’t just “emails sent”—it’s response rate.

To put this into practice, he built a workflow that takes a Slack message (whenever he reacts to it) about emailing someone and produces three versions in his voice: one direct, one collaborative, and one strategic. He still makes the final edits, but instead of rewriting from scratch, he starts with drafts that already sound like him and which are finetuned to get replies.

Steal the agent

How the agent works:

• Watches for an emoji reaction in Slack on any message about emailing someone

• Analyzes the message to figure out the recipient and the key points to include

• Generates three polished versions in Adrian’s voice: Direct, Collaborative, and Strategic

• Sends those options back in Slack for quick review and selection

• Creates a Gmail draft with the approved version

3. Build solutions that weren’t previously possible

Adrian made the point that some of the most valuable AI use cases aren’t necessarily about speeding up existing tasks, but enabling those that were previously impossible. 

Let’s look at an example within the context of marketing. One of the hardest questions for any GTM (go-to-market) leader to answer is: why are we losing deals or renewals? In theory, the answer is buried in customer conversations. Yet in practice, reviewing every transcript is so time-consuming that it rarely happens. And it’s unrealistic to sift through those calls at scale.

AI changed that. By analyzing transcripts automatically whenever a deal is marked as “lost” or “canceled” in the team’s CRM, Adrian turned what was once impossible into a repeatable process. Instead of relying on incomplete CRM fields, AI surfaces consistent themes (like pricing, missing features, or slow processes) that explain why customers walk away.

Steal the agent

How the agent works:

• Watches for Salesforce contacts marked as “Closed Lost” or “Non-Renew”

• Pulls all related call transcripts

• Analyzes transcripts to extract churn/loss reasons and clusters them into themes

• Logs results in a Google Sheet for tracking over time

• Sends a weekly Slack digest with the top reasons and themes

Making AI adoption stick

Adrian’s workflows show how AI can move beyond hype and into the day-to-day of a GTM org—whether that’s structuring deal reviews, reinforcing authentic communication, or surfacing churn patterns. But the bigger lesson is that adoption doesn’t come from top-down mandates. It also comes from building rituals that make AI part of the culture.

At Webflow, that means:

• Running hackathons with cross-functional pods so marketing, sales, and ops teams solve problems together.

• Structuring weekly revenue all-hands with a 3-2-1-1 cadence (3 priorities, 2 customer highlights, 1 team highlight, 1 AI demo).

• Creating an AI GTM council and even a monthly AI award to spotlight creative use cases.

The emphasis is always bottom-up: if a workflow genuinely makes work easier and more effective, people will adopt it. The role of leadership is to model curiosity, showcase wins, and create the space for experimentation. That’s how AI moves from “efficiency traps” to lasting enhancements.