Are We Ready for Agentic Shopping?
It seems everyone is giddy with the possibilities that AI brings to the shopping experience. Me too. Most shoppers already rely on product descriptions, shopper reviews, and expert recommendations to inform their purchase decisions. Now AI can synthesize all that information and more in seconds. Soon, AI promises to create your shopping list, find deals, and even buy the products for you. Heck, maybe AI will one day anticipate your needs and buy that too!
What could go wrong?
Lots, of course, but I believe most shoppers will willingly let AI help them shop once initial fears, ingrained habits, and technology glitches are overcome. I think the bigger question is HOW much control shoppers will give AI and WHAT it will take to earn their full trust. Shopping agents aspire to help at every stage of the purchase lifecycle, from product discovery and research to shopping list creation, through purchase and delivery. E-commerce and open web search have already mainstreamed product discovery. It’s still early days with AI’s ability to credibly research and compare products, but it’s a near certainty. The next hurdle is the biggest by far, learning shoppers’ needs and desires well enough to select the products, baskets, and retailers for them. It’s a bigger leap than most people think. Until shoppers consistently nod their heads at the product baskets AI recommends for them, there is no way they will give an autonomous black box their credit card, despite the promised time and cost savings.
I am not a naysayer. AI is amazing, but its accuracy is still too inconsistent. We joke about AI hallucinations, but the truth is that most errors go unnoticed because most questions asked of AI are on topics the requester does not know well. Most just accept the well-spoken answers as truth. But when you give AI questions on topics the shopper knows well and may be passionate about, like “What should I buy this week and where for my family’s busy week and upcoming son’s birthday party?", I can assure you that any mistakes and omissions will get noticed. Those mistakes will create doubt. Doubt will stop AI at the research stage of shopping. Accuracy is the gating factor of agentic shopping.
Let’s assume that accuracy will get there. What else has to happen?
The next challenge is that shopping agents live in a complicated ecosystem of players with competing objectives, inter-dependencies, and even open competition. As with everything about AI, the determining factor for this ecosystem to succeed is - you guessed it - DATA! And competing parties do not like to share data. Consumers are not super keen to share data either. Unfortunately, mainstream agentic shopping is a non-starter without visibility to a household’s pantry, medicine cabinet, refrigerator, family dynamics, upcoming calendar, budget considerations, favorite recipes, dietary restrictions, health conditions, lifestyle, product ingredients, nutrition claims, and so much more. Just like humans who consciously and subconsciously factor this ‘data’ into shopping decisions, so must AI, either directly via shopper instructions or indirectly through behavioral patterns. The tricky thing is that this data is spread across all parties involved: the shopper, brand, retailer, and the agent itself. Everyone has to play along or the shopper value equation suffers. Unfortunately, existing industry business models will make this difficult. I expect it will take a disruptive move by someone with the means to do it.
Okay, let’s assume the industry finds a way to make this happen. Will shoppers?
Ultimately, I think the pace and depth of agentic shopping will come down to some combination of Confidence, Convenience, and Consequences. The importance of each of these components varies by shopper, but together they will determine how much shoppers will trust agents to do their shopping for them. Growing confidence requires fewer mistakes, of course, but also more transparency in how products and retailers were selected. Improving convenience requires saving time and anticipating needs. Consequences require minimizing costly mistakes and painlessly rectifying them when they inevitably happen. Each impacts the others, and finding the right balance is key.
So are you saying there is a chance?
Yes, I think agentic shopping will become reality in some form. It might take a few tries to get it right, but it has too much potential to ignore. In my opinion, the chances of success improve with the following.
Be overtly shopper-centric - Establish no doubt that the agent represents the shopper, not the retailer or brand advertiser. Do not let ad revenue temptations ruin the shopper experience and dilute personalization, as we are starting to see with retailer media. Be transparent on where you make money from brands and retailers and how it benefits them (e.g. we bring you discounts, not ads).
Support several retailers per shopper - Delighting and advocating for shoppers mandates intriguing product diversity and impartial price/cost comparisons. This requires having multiple retailers or sellers for most product requirements. Anything less, and it's just a retailer app with fancy features.
Shop for the whole house - Real household needs do not fit neatly in your refrigerator, pantry, or medicine cabinet. Even the simplest life occasions or trip missions require shopping multiple 'aisles' and retailers. If the agent’s scope is too narrow, the convenience will not be there either. Eventually, both the shopper and agent will strive to maximize the investment they put into their ‘relationship’ and want to extract as much value as possible.
Establish incentives for retailers - Agents must demonstrate how their cross-retailer shopper value proposition brings them purchase demand that may otherwise go to mega marketplaces (i.e. Amazon). Minimize retail media revenue cannibalization concerns by not supporting things like in-agent ads. Collaborate and share market intelligence data on where they are winning and losing shopper transactions. Consider sharing revenue for non-retailer-specific audience, media, or data solutions.
Avoid replicating existing fulfillment and delivery capabilities - Let others pick the products from the shelves or warehouse bins and get them to your shopper’s door. Ditto for returns and other supply chain tasks. Retailers (and a few third parties) have invested massively in this last mile and are already motivated to keep investing. Let them. But never forget that you effectively outsourced this to them and mentally allocate some of that ‘savings’ to keeping retailers comfortable with your role with THEIR shoppers.
Immediately wire into established brand and retailer processes - AI compute, UX, and data management costs will not be trivial and cannot be compromised without harming the shopper adoption curve. The last thing you need is to be distracted by supporting clunky data operations and the fallout of avoidable mistakes. Tap into existing data synchronization capabilities and be quick to adopt emerging agentic AI standards like MCP. Avoid the risks and inefficiencies of having to scrape and script purchasing on e-commerce sites. If you must, use AI-powered agents (e.g. OpenAI Operator) to navigate sites as they are more resilient to e-commerce site changes.
Leverage proven shopper analytics - Do not focus exclusively on the latest and greatest LLM capabilities to optimize and personalize shopping lists. Integrate the proven shopper analytics and curated data that currently inform your product portfolios, merchandising, trade promotion, etc. This includes injecting choice drivers, price sensitivities, basket affinities, trip missions, contextual moments, and more into your recommendation engines.
Incentivize shoppers to provide feedback - Encourage and reward shoppers for providing you with feedback and context. Educate them on how this helps the agent improve product selection, suggest new products, reduce mistakes, and find them deals. Combine this with tangible rewards like rebates, discounts, samples, prizes, etc. Let them import purchase receipts made outside the agent app. Together, this will help future predictions and personalization, but also enhance your data monetization capabilities.
Proactively monitor and influence what AI is saying - Brands need to constantly monitor, understand, and steer what AI is saying about their products and competitors. This also means monitoring and responding to what consumers are saying via product reviews and social media. Leverage emerging tools to do this, as most brands do not have the resources or expertise to keep up with this fast-changing space. Invest proactively to ensure your content across your product descriptions, brand websites, naive advertising, influencers, etc. is consistent with how you want AI to represent you. Be genuine and honest with your content. Avoid the temptation to sway the truth or unfairly bash your competition, as consumers and AI are getting better at calling you out.
Evolve brand marketing tactics for agentic shopping - Advertisers used to joke that ‘Bots do not buy’. Now what? It's still true that bots do not respond to banner ads. Nor do bots see end-caps. They do, however, see what paid influencers, loyal fans, and naive advertising have to say. Shopping bots also love price promotions, loyalty rebates (e.g. third one free), basket size discounts, and free shipping. Net, brands will need to rethink their marketing mix and ‘align' with retailers on where best to direct their marketing dollars as shopping assistant and agent usage grows.
Prepare shopper agent business models for slow adoption - It could take a few years before shoppers see enough value to go from helpful shopping assistants to trusted buyers. That means shopping agent business models need to generate company-sustaining revenue from the product discovery, research, and shopping list curation stages alone. They cannot depend exclusively on taking a cut of the purchase transaction. Of course, the other alternative is to find someone with massive investment capital and patience to subsidize operations and investment for the big payday (see Amazon).
Establish separate communication channels to humans and machines - It’s already clear that we must establish parallel, but aligned, communication channels to consumers and machines. E-commerce product detail pages are already getting cluttered and diluted by trying to influence humans, search engines, and AI all at the same time. Potentially worse, brands are practically forced to spray the open web with content purely intended to sway AI. It will not take long for the open web to look like a made-for-advertising ’site’ that no one likes or trusts. Doing so will only push more business to walled gardens and further silo industry business models.
Net, who knows where this all goes? It seems crazy to say, but I think the core technology to make this work may already exist. Not so much for current industry business models, financial incentives, and data sharing. I think there are three likely industry scenarios…
Massive consumer-facing AI companies tap into the retailer ecosystem to pull this off (e.g. OpenAI/chatGPT, Google/Gemini, etc.)?
Mega retailers leverage their resources, reach, product breadth, and scaled logistics to pull this off on their own (e.g. Amazon, Walmart, etc.)?
A consortium of consumer-facing shopping, delivery, and promotion solutions put their pieces together with retailer cooperation (e.g. Instacart, DoorDash, Fetch, ibotta, Flipp, Out of Milk, Key Ring, etc.)?
It will be fascinating to see how this plays out. I cannot wait to help. It will take everything I learned over thirty years at the intersection of consumers, brands, and retailers. Let’s go!