Latest / Retail Media Breakfast Club / Walmart's Sparky AI Shopping Assistant: My Hot Take
Transcript
- Walmart Sparky === [00:00:00] I finally had a chance to play around with Walmart's new Sparky AI shopping assistant, and this is a fairly late mover in the AI shopping. Assistant space. We've got Amazon, who is early. Instacart was actually the first retailer to come out with a AI shopping assistant. And now that there's a whole bunch of AI shopping assistants available on retailer dot coms, ~um, Sparky, ~Sparky kind of seems to be playing catch up here, but. [00:00:27] ~But ~I think that this gives Walmart ~a bit of a late mover it, but here's why. But here's why. Sparky has ~a bit of a late mover advantage in the AI ecosystem [00:00:34] and may signal that Walmart actually has. [00:00:36] More strategic priorities outside an onsite shopping assistant. I'm gonna share some of my initial observations as well as some observations from other industry [00:00:46] participants and commentators ~where I think Walmart's really gonna go with this.~ [00:00:47] Where I think Walmart is gonna take this in the future and what brands should be doing right now to prepare for any of those scenarios. Let's go. [00:00:56] Good morning and welcome to the Retail Media Breakfast Club. [00:01:00] I'm your host, Kerry Masters, and every weekday at 6:00 AM Eastern, I bring you candid and real discussions in the world of retail, media and marketplaces in just 10 minutes. So bring your own cup of coffee and we'll get smarter about retail media together. [00:01:15] let's talk about where Sparky is right now. Sparky really feels like it's still finding its place. You can find it right now on the app only within search results and on product detail pages. Although Walmart says that soon, you'll be able to access Sparky on the bottom navigation bar just like we have with Amazon's. [00:01:33] Rufus today. So a few observations from my playing around a couple of hours with it today. First issue is that it hallucinates. I asked it about Seven Second Water and Sparky fabricated two different explanations. First suggesting that seven second water meant. Drinking water really quickly and then, um, suggesting that it denoted a, seven second face rinse routine, neither of [00:02:00] which has any grounds whatsoever. [00:02:03] When I asked Rufuss the same question, it corrected me and it said. Seven second water is not a thing. I think you're talking about eight second water. And so, you know, rather than hallucinating and telling me, just making something up, in order to answer my question. So that is very common with AI has been addressed with Rufuss, but not yet with Sparky, FYI eight. [00:02:25] Second Water is a type of hair care product in case you're wondering. So, uh, some other. Accuracy and usability issues that I noticed. Sparky calls out seller ratings rather than product ratings by default when I ask for ratings of a product. Still some issues with pricing. I noticed some of the commentary online was saying it's. [00:02:43] That Sparky is not pulling any pricing info at all. I didn't find that to be the case, but I did find it very inconsistent providing irrelevant results. For example, I searched for toys priced between $40 and $50, and ~it's, ~it shared only products priced under $20 with me. And it [00:03:00] is. Inconsistent with how it pulls fulfillment information. [00:03:04] It could tell me which products were available for pickup in store, which was helpful, but um, was inconsistent with sharing fulfillment information on third party items. It could pull, delivery times for some items, but not for others. So, all in conclusion here, not really yet ready for prime time in my opinion. [00:03:23] And that's probably why we're not really seeing it. predominantly displayed in the app experience. Let's move on to what's working really well and a couple of key differentiators here. One is the conversational promise. Sparky does do a really good job of understanding natural language queries rather than the keyword. [00:03:44] Driven searches that we're used to in the search bar. It's actually quite conversational. I asked it about. The benefits of low toxin coffee. And after explaining to me what it is and why it's important, it asked what type [00:04:00] of coffee I wanted. Did I want beans? Did I want ground, did I want instant capsule coffee, et cetera, so that it could help me to find the right type of low toxin coffee for me. [00:04:11] And, um, someone else, uh, in the. ~Uh, ~LinkedIn Commerce, ~Sophia, um, ~sphere that I follow, um, and appreciate his perspective on all things AI very much is. Carter Jensen, senior manager of enterprise marketing capabilities at General Mills, and he's done a big, deep dive into this as well. He observed also Sparky seems to really remember. [00:04:31] Context very well, maintaining conversation threads even after other interactions. And another good takeaway ~from, ~from another, ~uh, ~expert in the space, Joe Murphy, founder of Shelf Site, ~um, ~did some technical analysis that reveals that Sparky can potentially access browsing history. However, Sparky doesn't appear to actually understand or contextualize ~any per ~any prior purchase history, which really limits. Its ability to answer personalized questions and be truly [00:05:00] helpful. It would be so much more helpful if we could ~a ~query it about our past purchases. [00:05:05] Hey, can you suggest something that you think I might like? Or ~What was that? ~What was that brand of shampoo that I bought a couple of months ago? I really liked it, ~so,~ [00:05:13] so that's something that isn't. Yet in the makeup of Sparky. But the thing I found really interesting and different to what we're seeing with other assistants is that Sparky cites its sources in the Sparky interface ~face in, ~in the Sparky conversation, you'll often see ~a, uh. ~A link to the sources that Sparky has used for that answer when it's talking about products specifically ~that it's usually.~ [00:05:39] ~So it, ~it usually says, this response uses information within walmart.com and customer reviews. So it's pulling product information from its catalog, from the information provided by manufacturers and sellers and customer reviews. That seems kind of obvious, but in the more conversational advice type responses, it shares. [00:05:57] What external sources it's using for ~that, ~[00:06:00] that information. So for example, in my low toxin coffee query, it actually shared six or seven different sources that it pulled information from. And I think that this transparency around sources could be a really big trust builder. Now, it's obviously not fully there yet because it did totally hallucinate those responses to the questions. I had about seven second water where I will say no source was provided. So throughout this whole experience, sources were provided sometimes. [00:06:30] And that those were the contexts where it was making stuff up, and then in the situations where sources were provided, ~it did have it, ~it was accurate. [00:06:38] All right, ~let's move. ~Let's move on. So ~what is the, ~what does the future look like for Sparky? Walmart, I think ~could, ~has an opportunity to really lean into, its unique. Assets, which is its store network, its consumer reach, the reams and reams of customer preference data and shopping history, in store buying habits, combining that with [00:07:00] online browsing behavior. [00:07:01] Walmart really sits on an unparalleled data goldmine that could be used by Sparky to be ~a, ~a truly helpful, smart assistant, possibly even more so than what Amazon could possibly do. Now, I think there could be something quite helpful Here again, Carter Jensen pointed out this scenario. [00:07:19] Imagine Sparky creating your shopping cart on a Sunday night with all the things that you need and inspiration for. Some new purchases as well. Could order some things for you ~that ~from ~third party, um, ~third party marketplace, as well as create a list of things that you wanna pick up in the store in your next visit to a Supercenter. [00:07:34] Another is integration with Walmart's value added services. All of these things that make shoppers more sticky and loyal to Walmart franchise. Things like Auto Services, assembly Services, the Vision Center, ~subscription to ~subscriptions, et cetera. So these are. Assets that Walmart has, they could really integrate with Sparky, and I hope that they do that. [00:07:55] I do remain convinced, however, that ~an ~a retailer.com [00:08:00] onsite shopping assistant isn't the most exciting use of AI out there. I think that consumers are increasingly going to turn to their own ai. ~Uh, ~agents and assistants for help interacting with retailers rather than going to a retailer.com and engaging with that assistant, which, ~um, ~is always gonna be limited by the context of their own walled garden. [00:08:23] I shared last week Walmart is preparing for this open ecosystem where third party agents can interact with Walmart systems. [00:08:30] I personally see, ~um, ~the personal shopping agent, like the chat, GBT being our shopping sidekick and integrating in an agent to agent model with retailers as the more expansive opportunity. [00:08:42] but Walmart is clearly pursuing. More than one future scenario here. [00:08:47] Now bringing this back to land for the brands listening out there, we are moving towards a world where maybe only three to five products get recommended for any given query. We're not going to have the endless aisle pages and [00:09:00] pages of results that people might scan through. [00:09:02] It's like having a personal shopper who only shows you their top picks, ~which means that brands don't make, ~which means that brands who don't make it into the egg AI~ egg ~AI consideration set. Become essentially invisible. AI shoppers don't scroll. So if you're not optimized for ai, you will be invisible. And the onus here in brands is to develop enough content and context across many different channels that we know the AI assistance and LLMs are crawling and prioritizing. [00:09:29] There's lots of best practices and information out there. I'm not gonna repeat it all, but things like pr. Reviews and discussions on Reddit, customer reviews on any retailer site, including your own D two C site. All of these high fidelity sources for AI are [00:09:44] Opportunities for you to ensure that your product and brand is being picked up by the AI systems, whether they are onsite assistance or the LLMs that we're interacting with every day. [00:09:55] [00:09:57] we're left with is a [00:10:00] trust gap with AI assistance. [00:10:02] Sparky's approach to citing sources and showing consumers exactly where our information comes from. [00:10:07] Helps to address this skepticism amongst consumers, ~uh, ~that they feel towards AI shopping tools. I wrote recently about some new research showing that 55% of consumers are skeptical of recommendations from AI shopping assistance. So if Walmart can fix. These accuracy issues while maintaining that transparency, I think they could really differentiate Sparky in a crowded field, but the real winners will be brands that prepare for this AI driven future. [00:10:36] Whether shoppers are using Sparky Chat, GPT or Future Agents, the same principle applies. If you're not optimized for AI discovery, you're invisible. Complete product information. Genuine customer reviews and clear value propositions aren't just good marketing. They're survival essentials. In a world where AI curates our choices.