Agentic Commerce: Retailers, Take A Methodical Approach To Investment

The 2026 National Retail Federation (NRF) rang in the new year (post-CES) with an overwhelming number of product announcements around “agentic commerce.” The hyperscalers introduced new protocols, frameworks, and platforms to facilitate agentic commerce while payments and infrastructure brands (e.g., Amex, MC, Paypal, Stripe, Visa, etc.) announced support through endorsements or integrations. Finally, a handful of the US’s largest retailers announced the initial assistants or services built thereon. 

My take: Agentic commerce is an “and” - not an “or.” While agentic commerce (technologies) enable a combination of experience enhancements and innovations, the ability to scale moments including personalization, and operational efficiencies, most of the impact will be internal-facing in the near term. The impact on the end consumer’s experience is not as near. Philosophically, new technologies initially enhance experiences before they deliver net new experiences. And, digital touchpoints seldom if ever disappear. Yellow Pages and analog phone calls still linger even in this modern digital age. 

What does all this mean (WIM) to business-to-consumer (B2C) companies?

1. Agentic commerce will complement conversational commerce - not replace it. Conversational commerce most often refers to the ability to engage with a human or chatbot via a conversational interface at some stage of the shopping process (e.g., discovery, research, or purchase). To date, these scenarios seldom include payments due to the difficulties with integration, privacy, and data on the backend - unless you are in Brazil or India using WhatsApp. In theory once it is fully functional, agentic commerce will “take care of everything” once you give one of these virtual assistants a command such as “feed my kids tonight.”
   

2. Agentic commerce mostly impacts businesses today - not consumers. Agentic AI is real in that these models can reason, decide, execute, and iterate to power applications or automate workflows, primarily below the surface of consumer experiences. They handle messy integrations to multiple applications or data sources. Remember Andy Jassy’s quote on saving 4,500 developer-years on code migration?  Some have the authority to handle exceptions to free up humans (e.g., service, supply chain). Consumers will mostly appreciate this fancy automation through speed or relevancy - not the convenience of using a conversational interface to make purchases, at least today.
   

3. Early pilot sweet spots include youth brands …. Based on my experience with researching consumer technology behavior over my 25 year analyst career, the typical early technology adopter skews young (25 to 45 years), male (55% to 60%), affluent, and educated. Younger adults and teens also adopt early, but at lower levels given that they have less autonomy and spending power. According to a 2025 PEW Research Center study, 30% of US teens use chatbots daily. In comparison, 20% use TikTok or YouTube almost constantly. PacSun combines both in their PS Community Hub.
   

4. … or convenience purchases. Consumers will lean into more invisible experiences or agentic commerce (i.e., using voice or text commands to instruct a service with the necessary autonomy, context, and credentials to make purchases) when they seek convenience. Remember, consumers don’t always seek convenience. Depending on shopping, consumers use shopping to socialize or learn. For many it is a source of entertainment. Consumers will lean into more immersive experiences when they have those needs. It is therefore not surprising that early demos of the hyperscalers’ and LLM’ agentic commerce capabilities focus on more task-oriented shopping (e.g., Walmart, Target, Kroger, Chewy, Instacart, etc.) for groceries or simple, repeat purchases such as pizza with Papa Johns.
   

5. Automated end-to-end shopping is the aspiration, but AI powered features are the near-term reality. AI-first, end-to-end shopping experiences are still nascent. Amazon (Rufus), Google, OpenAI, and Perplexity (Buy with Pro) all have offerings, but shopping is still a relatively nascent use case. In a September 2025 study published by OpenAI, just over 2% of conversations pertained to purchasable products. Niche products like Faircado is a browser plug-in that looks for refurbished rather than new products. Tech solutions are rapidly adding features to their platforms to create more personalized content, replace search, make content more explorable by agents, and automate workflows beneath the service. Over time, consumers should expect most aspects of shopping to be a bit better i.e., faster and relevant.

Retailers: Use the HOST methodology to optimize choices for your customers and business. 

FOMO and FUD are real for brands as they face pressure to adopt AI to create more effective and cost-effective experiences for consumers. According to a July 2025 MIT study, however, most AI pilots fail to scale and deliver true returns. Brands need to prioritize their efforts. Pressure is on to add chatbots to shopping and service experiences, build virtual assistants, and create more personalized content for customers. I recommend the HOST methodology developed through collaboration with Julie Ask Advisory and Analysis.Tech (see image). 

1. Use our  HOST methodology to make smart use case choices that align with your users’ abilities and business objectives, risk, and abilities. Just because 800M people use OpenAI’s ChatGPT each week doesn’t mean they are shopping or researching products there. In fact, few do - just 2% according to OpenAI’s September 2025 study. WhatsApp has 3B monthly active users and facilitates the transfer of 100B messages each day. Outside a few geographies like India and Brazil, WhatsApp is not a commerce platform. Pay attention to your consumers’ trust, ability to write prompts, and comfort with sharing data. Just because they research products on these platforms will mean that they trust the platforms with their payment credentials.
   

2. Focus first on upleveling owned properties. GenAi’s ability to build memories or hold context is upleveling every brand’s need to know their customer; therefore, you want to observe as much of your customers’ activity as you can. Most of agentic’s near-term opportunities is in automating messy or complicated workflows that are behind the scenes, including order management, supply chain, and merchandising strategies. Adding conversational interfaces to assist with research, decisions, and service will increase conversion rates. Personalized content is another opportunity.
   

3. Start with upper funnel support of third-party experiences (e.g., GEO). While brands are unlikely to depend on OpenAI or Perplexity for sales volume (at least yet), they should prepare for consumers to seek information or do research within those applications. Anthropic, Google, and others are developing protocols (e.g., MCP, UCP, A2A) to facilitate the collection of information via agent browsing on behalf of consumers. Too long for this blurb, but brands should dive into generative engine optimization and consider the implications of agentic browsing. This will not be a one-size-fits-all strategy. Perplexity is suing Amazon for blocking its agents.