Top
The question of how artificial intelligence reshapes competitive advantage has become a dominant theme for equity markets. In recent months we have explored this issue through the lens of financial information services and professional services companies within Industrials. Here, we turn to a different but equally topical area of opportunity and disruption: agentic commerce.
Consumers have a finite amount of money to spend, so realistically agentic commerce is not “free growth”. The more relevant debate is how spend will be redistributed across channels and business models."
Agentic commerce describes e-commerce activity that is initiated or shaped by an AI system (typically a large language model, or an “agent”) after a consumer interacts with it. We think this shift matters for investors because it changes how demand is expressed, how products are discovered, and potentially where economics accrue across the commerce stack (software, payments, logistics, advertising and marketplaces). The near-term reality is likely to be evolutionary rather than overnight disruption, but the direction of travel is clear: the interface through which consumers shop is becoming more conversational, more personalised, and increasingly capable of executing transactions.
From “search and click” to “ask and buy”
In broad terms, agentic commerce occurs when a consumer engages an AI interface to find, evaluate, and/or purchase a product or service. In practice, two distinct models are emerging:
- “Conversational” agentic commerce (human-in-the-loop)
A user interacts with their preferred AI agent to describe what they want, explore options, and narrow choices – but the final ordering and authorisation step is still taken by the user. This differs from traditional search in two important ways:
- Hyper-personal recommendations. Over time, an AI assistant can accumulate context about a user’s preferences, constraints and prior behaviour (e.g. size, style, budget, delivery preferences, brand affinities). That context makes product discovery meaningfully more tailored than a generic query.
- Higher specificity upfront. AI prompts naturally encourage detail. Instead of searching “men’s jeans” and then filtering on a merchant’s site, a user can specify in a single request cut, sizing, colour, price ceiling and use case – often producing better first results with fewer clicks.
- Truly autonomous agentic commerce (agent-in-the-loop)
This is the more futuristic but strategically important model where the consumer delegates the task end-to-end; the user specifying the objective and constraints – and the agent returning with a purchase confirmation. A prompt might read: “Please purchase road running shoes for 5–10km runs, UK size nine, with good cushioning, mild overpronation support, under £150, green/blue, deliver to my home address, charge my Visa.” If this model scales, it compresses the traditional shopping funnel (browse → compare → checkout) into a single instruction and an approval step.
What needs to be in place for agentic commerce to scale?
We see two practical bottlenecks that will determine the pace of adoption: product data access and payment authorisation that preserves trust and reduces disputes/returns.
First, agents need reliable, real-time access to merchant catalogues. Even today, AI-assisted shopping often struggles with stale, incomplete, or inconsistently formatted1-level data. For agentic commerce to work reliably, merchants need to expose accurate information – inventory, pricing, variants, delivery windows and return policies – in a way agents can query programmatically. That points towards a world where:
- product data becomes more standardised and machine-readable (clearer attributes, tagging, taxonomy consistency);
- merchants provide real-time catalogue access via APIs2 / agent-friendly connectors; and
- successful commerce enablement platforms are strategically set up to help merchants organise and serve this data.
Second, in order to get to full autonomy, the payments infrastructure must evolve from “prove a human is buying” to “prove an authorised agent is buying on behalf of a human”. Card networks and issuers have already built much of the underlying toolkit through tokenisation (replacing sensitive card details with non-sensitive tokens) and modern authentication flows. The next step is extending these tools to agent-initiated transactions with clear consumer consent, guardrails (spend limits, merchant/category rules), and robust exception handling. The goal is to enable delegation without increasing fraud or return-related disputes. How much data is captured around these transactions, and by whom (payment networks or LLM3 providers), might have important implications for the economics of processing agentic payments.
How to think about total addressable market (TAM)
Consumers have a finite amount of money to spend, so realistically agentic commerce is not “free growth” for the system. In our view, the more relevant debate is how spend will be redistributed across channels and business models.
One hope for retailers and platforms is that agentic tools increase online penetration by reducing friction, particularly for routine replenishment such as the weekly grocery shop. Even if aggregate e‑commerce spend – currently around 16%4 of overall retail spend in the U.S. – remains unchanged, agentic systems may well increase the number of transactions. An agent is indifferent to splitting a basket across multiple merchants to optimise price, availability or delivery. This could reshape basket composition, fulfilment patterns, and the economics of “one-stop” aggregation.
“Skate to where the puck is going, not where it is now”5
We focus on identifying businesses with intangible assets that underpin strong and sustainable returns on capital. Understanding how technological, regulatory or competitive changes might impact the strength of the franchises we own is a key part of our constant re-underwriting of our portfolio. We see two areas where an advance of agentic commerce might challenge or upend the status quo:
- Impact on the power of brands. Agentic commerce is likely to change how products are discovered (less SEO6-driven, more data-driven), and it could reward merchants that provide rich, reliable SKU data and clear service terms. It may also interact in nuanced ways with brand strength and product differentiation. Where brand differentiation is weak and products are highly interchangeable, agents are more likely to optimise aggressively on price and convenience - which could marginally level the playing field between large and small players. Conversely, products which demonstrable functionality and specific use cases may benefit; an AI agent is better placed than a human to scan hundreds of SKUs to match niche needs. This rewards companies that innovate rather than merely market.
- Impact on e-commerce platforms and aggregators. If AI interfaces become a dominant starting point for shopping, some platforms could face disintermediation risk (reduced ad/upsell leverage), while others may benefit from incremental traffic and improved conversion. Questions around openness, “merchant of record” responsibilities, and bargaining power between platforms and AI interfaces will be central.
Meanwhile, for payments, we believe that agentic commerce appears more evolutionary than revolutionary. As things stand, we believe card networks will probably be net beneficiaries as tokenisation and consent-based authentication become more valuable in a delegated world. Today, about 40% of card transactions are tokenised by Visa and Mastercard for traditional ecommerce use. Merchant acquirers, by contrast, may face pressure as some traditional revenue pools linked to risk and conversion shrink. This is an area we continue to watch as it evolves.
Moats in the agentic commerce world
Even if agentic commerce ultimately represents a modest share of global ecommerce, it is likely to be a part of the equity debate for a while, before we settle into a new equilibrium. We will share our thoughts on these evolving topics with you in our upcoming Global Equity Observers.
1 Stock keeping unit; a code that helps sellers identify products in their inventory
2 Application programming interface: a set of rules or protocols that enables software applications to communicate with each other to exchange data, features and functionality.
3 Large language models
4 Source: Data as at Q3 2025, U.S. Census Bureau https://www.census.gov/retail/ecommerce.html
5 Quote attributed to former ice hockey player Wayne Gretsky
6 Search engine optimisation.
Featured Insights
This image indicates content designed specifically for Financial Advisors / Investment Professionals. This material is not to be used with the public.