Building the case for a vertical brand agent
Generated by Klaviyo AI
Publicis Sapient's Duke Marr argues that widespread AI anxiety is giving brands rare permission to redesign commerce from scratch instead of bolting AI onto an outdated funnel. He introduces the "vertical brand agent," a brand-specific AI that can sell, serve, and act across every surface, and lays out a practical playbook for building the business case for one.
- Your website is becoming a machine-readable database: Product data and policies need to be structured well enough for both people and AI agents to pull accurate answers from.
- Not every category is headed toward full automation: Commodities suit full agentification, while fashion, home, and luxury still want a human-like touch.
- Own the knowledge and orchestration layers, and buy the rest: Brands don't need to build the whole stack, just the parts that carry their proprietary data and decision logic.
- Change management is the real bottleneck: Getting the right cross-functional team aligned matters more than the build itself.
- A vertical brand agent needs its own scorecard: Standard service-chatbot metrics won't capture the value of a tool that combines sales and service.
"We are in the age of AI anxiety," said Duke Marr, vice president of customer experience and innovation consulting at Publicis Sapient at the start of his K:BOS session. "Does the peril outweigh the promise, or does the promise outweigh the peril?"
Marketers in the room were split on the answer.
According to Pew Research Center, 52% of US adults are more concerned than excited about AI in daily life, up from 37% in 2021. Bentley University and Gallup found that 79% of Americans don't trust companies to use AI responsibly. And only 18% of Americans are comfortable with AI tools, according to Scayle.
This AI anxiety, according to Marr, offers brands a rare permission structure to redesign the commerce experience. Tacking AI onto the old funnel won’t work, he said, but brands can start from scratch, focusing on intent, dialogue, memory, and action.
"It's time for the great commerce reset," said Marr. Customers can state outcomes in natural language, agents can remember preferences and context over time, commerce and service become one continuous relationship, and brands show up wherever people express their intent.
"We spent 30 years teaching customers to navigate our systems," said Marr. "The next era will be about teaching systems to understand customers."
Shopping is becoming a dialogue
According to Marr, AI-driven shopping is scaling faster than any prior digital shift, including mobile. Consumer behavior is changing as shopping is becoming a dialogue and people are engaging with third-party surfaces like ChatGPT and Claude.
However, the core metrics brands track tell us the experience is still broken, said Marr. The average ecommerce conversion rate in a recent IRP benchmark snapshot is 2.2%. The average cart abandonment rate is 70.22% according to Baymard.
As consumers travel through the "pretzel" path to purchase (Marr's redefinition of the funnel), they're clicking on websites less and having conversations more. Customers describe an outcome in their own words, an agent asks a clarifying question or two, and it either hands back a recommendation or completes the task. That's happening on surfaces brands don't control, through whatever agents a customer already trusts, and on brands' own sites.
But consumers aren't yet ready for fully autonomous shopping experiences, at least for many shopping categories. According to recent Visa research, 72% of consumers say they're willing to shop with an AI agent, yet only 23% are willing to let one actually transact on their behalf. Consumers want the help, from a resource like an AI customer service agent, but they're not ready to hand over the card.
Your website is becoming a database
Agentic shopping changes what a brand’s website even is. It’s not a digital storefront, it’s a machine-readable database of information that humans and AI agents gather information from.
To prepare for this new age of shopping, Marr says to focus on your brand's website as a repository of information, a digital asset manager (DAM).
"Your commerce website is now a public-facing DAM for your brand and your business, a machine-readable source of truth that agents, including your own, and people read from."
Many brands treat their ecommerce site as their flagship, as the place to do their best storytelling, showcase their most immersive design work, and take the biggest swings on their brand. Storytelling still matters, but your website now has to be readable by something other than a human, and include:
- Product attributes
- Return policies
- Sizing guides
- Warranty terms
All of it has to be structured well enough that an agent, whether it's a shopper's or the brand's own, can pull an accurate answer out of it in real time.
"When an agent shops, your brand becomes a data point," Marr said. "If you're not careful, your brand identity gets reduced to that data point. Own it."
Owning it, in practice, means treating generative engine optimization (GEO) the way brands once treated SEO, structuring product data with the same rigor and specificity, so that when an agent compares your product against a competitor's, it has something accurate and detailed to work with, not a generic listing it has to guess at.
Moreover, Marr said, “everything you do to make your brand show up in third-party agentic surfaces you can also use to power your own vertical brand agent. So why not kill two birds with one stone?”
Vertical agents operate wherever intent shows up
Marr says the industry is moving toward vertical brand agents, brand-specific representatives that can sell, serve, advise, and act, rather than a generic bot bolted onto a help center.
"A vertical brand agent takes your best salesperson, as well as your best customer service person and makes them digital, and universally accessible on any surface," he said. Amazon's Alexa for Shopping and Walmart's Sparky are two consumer-facing versions of the idea already live at scale, and according to Marr, mid-market and enterprise brands can build their own versions, trained on their own catalog, policies, and voice.
"The agent is not a channel," he said. "It's an operating layer that translates brand, customer, and product intelligence into action across channels." That means showing up on the website, inside a store associate's tablet, in a returns flow, and on whatever third-party surface a customer happens to be using.
He was careful not to oversell how much of the purchase agents will actually take over. Category matters more than technology readiness. He described the degree of human involvement in the purchase decision and using AI for customer experience along a spectrum.
- Full agentification (Set it and forget it): 86% of consumers say they'd let an agent handle routine replenishment, where discovery adds little value, according to Wildfire. Pet food, household staples, office supplies, and basic groceries sit in this category.
- Research assist (AI narrows, human decides): 55% of consumers want AI's help most when researching electronics, according to Contentsquare. The same logic extends to automotive, appliances, telecoms, and insurance.
- Assisted discovery (AI helps, human curates): 31% say browsing for ideas is essential, according to FMI 2025. This includes grocery/food, travel, gifting, beauty, and personal care.
- Agent-enhanced (Discovery is the value): If there's anything vaguely emotional to the consumer, they will want to be involved, according to Marr. That includes fashion, home decor, furniture, luxury goods, and specialty food.
How to build the vertical brand agent
When it comes to turning brand data into agent action, Marr offered a vendor-neutral blueprint for a vertical brand agent composed of 5 layers.
- Experience: Chat, voice, and agent surfaces where customers show up
- Orchestration: Reasoning, tools, and guardrails that decide and act
- Knowledge: Governed product, policy, and content
- Data and identity: Customer profile, inventory, orders, and CRM
- Commerce and fulfillment: Cart, checkout, payments, and post-purchase systems
Marr recommended owning the knowledge and orchestration layers, and buying proven components for the rest.
Here’s an example of a real AI customer experience retail tech stack: a Klaviyo Customer Agent trained on product and service data, a product recommendation feed the ecommerce team already manages, and Klaviyo's MCP (Model Context Protocol) Server acting as the orchestration layer that connects a brand's own data to outside agent surfaces.
The playbook: how brands prioritize, launch, and measure an integrated agent
Marr offered a practical breakdown of what this shift entails for brands.
- Map high-friction journeys. Identify where customers struggle across sales and service, then quantify leakage and operational cost.
- Prioritize use cases. Rank by value, frequency, feasibility, risk, and data readiness.
- Build the intelligence layer. Unify product, policy, content, customer, inventory, and service data into governed knowledge.
- Design and train agent capabilities. Define tone, actions, guardrails, escalation logic, and success criteria.
- Launch across touchpoints. Start in one channel, then expand to owned, retail, and agentic environments.
- Measure, learn, and evolve. Track conversion, resolution, effort, revenue, cost, and trust.
Each represents a phase. He estimated a pilot, sequencing the work to show value early and then scale with confidence, could take 3 to 6 months.
He quoted Jason Goldberg, chief commerce strategy officer of Publicis Groupe: "If you show me a company that succeeded with every experiment, I'll show you a company that's very poor at picking their experiments."
Change management is the biggest hurdle to overcome with AI
"The hard part will not be the technology," he told the room. "It'll be the organizational change management. Once you get that aligned, it won't be that hard to launch a pilot and see what happens."
That means deciding early who owns the initiative. Marr suggested a small cross-functional group, with whichever leader among sales, service, and IT is strongest put in charge, rather than waiting for a new role to get created. "Applied AI is a generalist's job," he said, "in all the best ways."
On measurement, Marr's warning was specific: don't let a chief financial officer (CFO) grade this against the metrics built for a service chatbot. Traffic, cost per contact, average order value (AOV), and customer satisfaction score (CSAT) still matter, but a vertical brand agent needs its own scorecard:
- Agent-assisted conversion rate
- Recommended bundle or attach rate
- Cost per resolved intent
- Customer effort and trust
- Intent capture across each surface where the agent shows up
"This is combined service and sales," he said. "Measure it that way."
Where do I even start?
Marr closed by reminding marketers in the room that the great commerce reset is still centered on humans.
"It's not 'Do we have AI?'" he said. "It's 'Can our brand be represented intelligently, usefully, and faithfully wherever customers ask for help?'"
For marketers wondering where to start, Marr's advice is to pick one journey, whether it's sales or service, and build the business case.
"Stop trying to fix the whole funnel at once, and start with the one conversation your customers are already trying to have with you."
Marr built a workbook that helps companies build that business case. Companies plug in traffic, conversion, ticket volume, and cost, and it models the lift, deflection savings, and return on investment a pilot could realistically deliver.
According to Marr, the future will be won by merchants who understand customer intent. The next 30 years will be about systems understanding customer intent at the get-go.
Related content
