Information Quality Is What Wins In The AI Era

0:12Tim Frank: Welcome to Ad Exchanger Talks, the podcast devoted to examining the issues and trends in advertising and marketing technology that matter most to you. Today's episode is sponsored by Basis, the leading intelligent operating system for autonomous advertising. Its enterprise AI solution transforms campaign briefs into strategies and media plans that integrate directly into omnichannel activation. You can learn more at basis.com.

0:50I'm Alison Schiff, editor in chief of Ad Exchanger, and you're listening to Ad Exchanger talks. And for that, I thank you. My guest this week is Tim Frank, Corporate VP of Monetization, Commerce, and a little thing called the AI Economy at Microsoft. In Microsoft's view, the web has bifurcated into three webs, and all three now exist simultaneously, which means it's an interesting time to be alive, especially if you're a business. We've got the human web, which is basically the way things have worked for the past 25 years, where people search for things and drive traffic to websites. We've got the LLM web where AI synthesizes different options, but it's still the human taking actions.

1:31And then the true agentic web, which is like the do it for me web where agents find, evaluate, and transact on a person's behalf. We'll dive into all of it, plus lots of other good and nerdy stuff, including the Universal Commerce Protocol, GEO, and measurement in the AI era. But first, there's still time to snag your ticket to Programmatic I.O. New York taking place on September 28th and 29th at the New York Marriott Marquis. It's kind of the place to be, so just saying. Podcast listeners get 10 % off with the promo code POD10, so what you waiting for? Get your ticket and I'll see you there. Hey Tim, welcome to the podcast.

2:09Hi Allison, happy to be here. So usually I start out by asking people to share something about themselves that not a lot of other people already know. But I want to start by talking about your love of space, because I've seen you describe yourself as a space nerd on LinkedIn. And I also read a post that you wrote about how you stopped an internal meeting a bunch of months ago because you wanted to take over the screen to share the Artemis 2 launch. So before we talk about the AI economy and monetization and all of that jazz, and we're going to nerd out on that extensively. What got you so interested in space exploration?

2:52Well, when I was a kid, I read a lot about what happens at the edge of a black hole and what's the physics like for that. And it led me to my first job and first eight years working in space, everything from the science of it to how we might explore and learn new information about how the universe formed. And I've always been tracking like, what is the upcoming launch? And Artemis was not just a normal launch. It was a really big deal. NASA kind of recreating this huge program. And it wasn't like the most critical work meeting, but it was one where a bunch of people that I was friends with and colleagues with, I was like, look, this is the moment.

3:32This is like the 10 second countdown and then the key launch. And so it was a fun way to just bring that into the office culture a little bit. And some of that like fun questions that come afterwards, like, oh, what would have happened if it didn't go well? Or like, when do you know it's safe to turn off the live stream? So that was kind of the impetus. When do you know that it's safe to turn off the live stream? On the bottom, usually there's like a timeline of launch and like what's happening. And so there'll be a couple of points, like when it reaches the maximum like physics load on the system once it gets past that usually called max q then you know that everything is downhill from there and it's very likely that physics will not destroy the the spacecraft there's a couple of key events main engine cutoff miko that's when the booster has done its job and it's probably on its way to orbit And, you know, if it's your money on the payload, then you care that it gets all the way to the location it needs to do its job.

4:37But from a pure physics perspective, it's probably going to go well by that point. But Max Q, not to be confused with performance max. Totally unrelated. Maximum physics torque. And then maximum performance delivery. Yeah. Yeah. By the way, Tim, I have a feeling I might have robbed you of sharing a fun fact about yourself because I do usually ask people to share one thing about themselves that not a lot of other people already know. And then I just redirected you elsewhere. So do you have one? I do. I had prepared and I had heard of this legendary question. Legendary. Well, you know, I think that fits.

5:20Well, one of the things that I've been really into and it doesn't make its way to LinkedIn as much is my child was in robotics the last three years. And it's been really fun to get into it with her and with the team. And if you're not familiar with it, there's a central organization that posts a design challenge every spring. So like as the world championship ends, they post next year's challenge. And it's a fascinating opportunity for young people to go through and decompose a problem, build strategy, do some game theory about how it might work, and then actually go about building the robot and the software and the system.

6:03And that's been something that's been great to get into. The adults in our program have a rule that the adults don't touch the robot or the code. Not all teams follow that rule, and you can see which parents kind of get way too involved. And I think it's just like a really excellent way in this age of AI, people wonder, what should young people study? And I don't know the answer. But I do think learning a lot about different disciplines and then how they come together is going to be valuable. It's going to be a really good foundation for the future. And robotics is just one way that that's super approachable, even for middle school kids, to have that learning themselves.

6:44Thank you for bringing it back. And I also want to draw a connection between space and what we need to get to space and like what you now do on a day to day basis. So creating the AI economy or supporting the AI economy and what you're building at Microsoft, because to go to space, you need more than a rocket to build and support the AI economy. you need more than a model, you need infrastructure. So expand on that thought a little bit for me. I've given you a platform to lift off. Okay, well, it's perfect. You absolutely have. I think that the web that we know at a very macro level, the one that we got good at clicking on and finding stuff and knowing how to do searches to get exactly what we want back, and I don't know how many tabs you have open, but people have gotten really good.

7:37Large quantity. Large quantity. This internet's amazing. And while that internet is here today, there's a new parallel internet that is growing. And we sort of think of this as three different internet eras that all exist at the same time. So let's call that first one the human internet. And that's the one where, you know, humans kind of driving everything. We have the LLM internet. This is the one where maybe you're asking for the internet to help me choose. It'll do a ton of the work, but the human still makes the choice at the end. And then the third internet, the third era of the internet is the agentic web, which is do it for me.

8:17And this is I fully trust an agent to carry out the task completely for me. All three of these exist at the same time. And you can think about any task you've done in the last week or so. You might be in different states for different activities. Things that bring you joy where the friction is still fun, you want to be more a part of that. But maybe things where you know exactly what you want done and it's not as interesting to be a part of it, you'd love to delegate that off. And what we're seeing for a lot of businesses is to exist and be successful in the human internet of the past, bots were thought of as bad.

8:56I got to keep them out. I got to verify that there's a human. But what's happened is every consumer interaction has sort of become an enterprise sale. You had the human, but they've now hired an agent to help them. And what businesses are in the middle of is adapting to that world where you have to consider, well, what's the human experience? And this agent that they've sent out there, how do we treat that differently? And so a lot of what we've been doing is trying to build forward compatible solutions that'll work for all three eras. So you don't have to guess at what rate the uptick is for each of those new eras that are coming online.

9:38Let me pause there because that was kind of - Yeah, well, I have a bunch of questions. I'm like, which one should I ask first, but I want to talk about the user experience and what people gain in this world versus what they give up. And you talked about places where friction might still give people some enjoyment. And our editorial director, Sarah Sluice, she wrote a column for us over the summer with a great headline, I used an AI agent to do my back to school shopping. It was easy. so why do I feel sad? And the piece was about how she used an AI agent to handle the more tedious parts of back to school shopping.

10:20But the experience still felt kind of lacking because it removed the browsing and the comparison shopping and the impulse discovery and a sort of like a nostalgic ritualistic aspect that comes along with the beginning of the year. And, you know, people are lighting a few fires in their backyard. It's like very sensory. you know? And so efficiency is great, but convenience does remove some of the joy, perhaps. But is that just me being a nostalgic weirdo who's still pretty used to the human web? They're the three eras, and I'm still in the first era. I don't think that that's the case. I think there's a lot of people that have fond memories of back to school.

11:04And it might differ for each person what's fun. You know, is buying the supplies the cool thing? Is getting the new syllabus and seeing what you're going to learn the cool thing? Is it the closed side of back to school? I think for each person, it differs. Maybe just to share in my family, we also have back to school. I have kids that are just newly arriving, one in a brand new school. And the way that we chose to use technology was we actually looked for some of the pencils and pens that we love the most in the house. And we gave that as input and said, well, what do I like about this? What are the properties that are great?

11:44And help me find specialty retailers that provide this type of an experience. We were in New York over the summer. And so there was a few of these like Japanese textile shops and paper shops and really cool places. I love that stuff, by the way. So we had a whole map built out, assisted by AI. But what it helped us do is sort of pre-qualify like, oh, what are we really interested in? Where are these places around the city? What's a smart loop to go to the mall? And in the end, we physically visited the stores, bought the items, probably spent more than we expected or would have if we had purely outsourced that.

12:26And for us, it was a much more joyous experience because we didn't spend the time kind of having to compare and find everything ourselves. Now, that's just a different version of back to school in our world. But what it helps, I think, emphasize is that you can actually jump in between these different eras. And I don't think that the human experience is going to go away. And maybe that's a relief for most people. I think it's a low risk that will just completely outsource all joy and all fun. But I do think it can make the human enjoyment even better. What does it mean, though, this experience that you're describing for a business that's trying to figure out where to put its energy and how to be discoverable as you build out your map for your visit to New York so that they're one of the shops that you visit or they're one of the brands of Japanese textile-based pens and stationery and whatnot that you are aware of so that when you see it, you buy it?

13:26This is so important because a lot of businesses right now are intentionally blocking these types of agents and systems. And it's important to know that this is a very fast growing channel. And if you're not available, then you're never going to be found. So we kind of have three different things that we recommend. First is to get found. Second is to get recommended. And then third is to be chosen. So if we kind of talk about each of those on getting found, like you should not block the agents and you should structure the information for the products or services that your business has. Like why did you build the pen or the piece of paper?

14:09Like what thought went into it. Provide all those specifications. Maybe the humans never scrolled to the bottom of, and maybe you never even put it on your site, but now the agent will read that. It will tirelessly get to the end and consider all that information. If you do have products, you want to have it structured with like a merchant center feed, you want to have pricing and availability. How many shops were there in New York that I just never found? Because when my agent went looking, they weren't quickly returning results. They weren't compatible. They weren't in the index that was checked.

14:46This is the stuff that's super important because this is not only true within Microsoft's ecosystem, but every AI is building in a heuristic into the harness that says, when I get this type of question, here's how I go and search for it. And I'm not going to make the user wait for me to go crawl everything on the internet. So that's the get found piece. And we can kind of walk into these other ones if you want to go deeper on recommended and getting chosen. I would love to, although as a total side point that has nothing to do with anything, I've often thought that heuristics instead of heuristics would be a great band name.

15:22I don't know if you agree. Ever since I was in college and I first was exposed to the term. Certainly if it was like a mood, you know, like more musical, more like visceral. Yeah, it would need the right vibe. I do want to talk about recommendations, but first, I don't want to get away from the question of whether to block or not block. Microsoft, you guys have a clear point of view that you really shouldn't be blocking by default, but there are some publishers that do. And they feel like it's the only way to protect themselves, not that they don't want to be part of the AI economy, but they want to do it on their own terms.

16:05And so they would block by default and then let others in depending based on whether they get paid, whether there's a respectful relationship there, because I think otherwise they feel like they're giving away their crown jewels just on a platter. So, I mean, this is such a big question. I don't know that you have an answer because there is no perfect answer yet. But how does a publisher or a commerce site or really anyone whose livelihood is tied to the Internet, how do they protect themselves so that they're not getting scraped to like oblivion without recompense and also not cut themselves off from like the fastest growing source of traffic?

16:49There's a stat. I think it's human that put it out there, which is so wild. Last year, agentic browsing traffic was up like 8 ,000%. Yes. Almost like not worth tracking at that point. Like it might as well be a bazillion percent. Well, the question is probably will it be bazillion or trillion or whatever X. Going into this calendar year, there was a lot of people guessing when bot traffic would surpass human traffic. It happened, right? And it happened in June. I checked recently, it's around 60 % now bot. Once an exponential line, though, is close to a roughly static line, that's the interesting point at crossover because that exponential is going to dwarf the human internet pretty quickly here.

17:38So I think that the ignore the new channel because it might not be a big deal, I think that's a very dangerous position to take. And if you kind of analogize this to e-commerce maybe 20 years ago, people were uncomfortable at first to put a credit card online. Like, what's going to happen? Am I safe? And so we've actually been looking through and talking about all the different components that need to mature and be ready for people to feel comfortable in this new AI model. We call it Prism. There's five components to it. Can kind of come back to that in a minute. I want to go to your big question, though, which is all the businesses of the world today have a business model that they know and they understand.

18:25And it's really scary to think about what's going to happen in these new eras that are coming online and you see them. It's a big tidal wave of agents. And when we talk about the AI economy, what we're really talking about is how do we help businesses succeed in those future eras? And that might mean different business models, but it certainly means looking after all stakeholders that are a part of that ecosystem and ensuring that they get a fair shake, that they have positive feedback loops and incentive systems. And so you're bringing in publishers, which is IP. Obviously, merchants have things to sell.

19:03We're thinking of it for the stores to make sure that they can be discovered and their hours are there, maybe local services, you know, how do I book an appointment, travel, try booking a hotel right now through an agent. It's quite challenging. I want a restaurant that has outdoor seating and will let me bring my dog and is inexpensive and has a good wine list, like go find me one and just make the reservation. It should be within this radius. That's right. All of these things that we do today. And so what I would offer to businesses, especially ones that kind of have what's just a human compatible interfaces, let's say that restaurant.

19:41In the future, if there are trillions of agents, you almost could imagine that their phone line, if that's the only way to book a reservation, is almost going to be getting denial of service attacks just because of all the volume. And so companies need to build systems to be compatible, to talk to these agents and to allow them to get information and take action. And really, that's like the ideal way this should work. It should probably be on the same reservation system that you're using for humans because ultimately there's a finite number of tables and servers and capacity of your restaurant.

20:16And it should just be another channel that's available to manage. What businesses don't really have though is a company or a set of partners saying, we're going to help you transform your business that way. And so that's one of the initiatives that we've built out. Right now, we call them brand agents, but it's a way to take your business as built for the human era and make it compatible for the AI era. And that's not only to AI systems, but that could be to other agents. That could be to a vast variety of future models that come online and give your business the ability to talk and work with them.

20:56At one point, just a couple of minutes ago, you said something about future eras, but it is worth pointing out that it's kind of just happening today, right? It's not futuristic to say that agents are going to do these things. They're already starting to do these things. They're kind of the current era. Yeah. Especially when you look at the latest models that are very good at computer use, they're kind of saying like, we're just going to go over the top and use it the way that you've designed it for humans to use it, that's fine, except that we really should have more of a native implementation so that those systems can be efficient.

21:36And so like, you know, in my paper example in New York, maybe there were three other businesses that we just completely missed out on because I was only going to wait as a consumer for 30 seconds for that query loop to finish. and whatever wasn't there didn't get surfaced. And so I do think that that's this really important part for businesses is if you're wondering how does your business look in the AI era, you're gonna want a software that helps you. And so there's many solutions for this. One that Microsoft offers is called Clarity and it provides human internet analytics. And what we're doing is we're adapting it to show, well, what about this new agent that got involved in so many of the different consumer journeys.

22:25And it'll give businesses an understanding of where and how do they show up? What does that look like? Not only on Microsoft's products, but across the internet. And then what can they do to improve either their data quality or structure or brand story? What are people saying about them? Because really for brands, it's not just their own narrative. It's what do the reviews say? What are people saying in different forums? Agents are going to go triangulate this. And it's not just the brand's message anymore. It's the triangulated message of what is everyone in the ecosystem saying. And I think that's very powerful.

23:04And I think it's going to drive the right behavior over time. It's a kind of very expansive social listening. You know, just it's listening to all the things. Yeah. Yeah. And you can just think about what in your purchase history would you have ever spent the time to be that comprehensive and exhaustive for? Probably not all the things that you buy. And the agent doesn't have the same time management issues or the same fatigue or the same comprehension challenges. And so it's going to basically raise many purchases to be at the same elevated level of a considered purchase. So we're going to take a quick break, but when we're back, I want to stick with talking about recommendations because I have a few more questions about that.

23:51And I also want to talk a little bit about measurement because measurement is changing drastically. So stick with us.

24:08i'm sarah sluice editorial director of ad exchanger and i'm here with mark mckechran vp of product at basis which is supporting our podcast this month hello mark hello how are you doing i'm good especially because we're going to talk about three-letter acronyms which i think is one of ad tech's favorite things, although this goes beyond just ad tech. So help it explain this to our listeners. What is an MCP server and how are advertisers using it? I do love a good TLA. MCP stands for Model Context Protocol. Anthropic built it, open sourced it in late 2024, and then handed it to the Linux Foundation.

24:51It's plumbing that connects the systems to agents. It lets an AI model reach into a piece of software, see what's available, and pull the right lever instead of hand-developing of the spoke integration for every tool that the model touches. And over the past year, Google, Meta, Amazon, and others have rolled out their own ad platform MCP servers, as have DSPs and SSPs. A buyer can now say, pull my campaign report or build this audience in plain language and have that request land correctly inside whichever platform is listening without building a custom connector for every combination. And eventually those instruction sets, they begin to look like standard adapters.

25:39Okay. So MCPs are taking off like wildfire, I would say, and with good reason because they're very useful. So tell us a little bit more about what advertisers are able to control when platforms have an MCP server. So control depends on who built the server. An agent can pull reporting, diagnose a stall campaign, or explain a CPC spike. But in most cases, a human still has to make the change. Some folks like Meta have enabled agents to create audience, set budget, and launch a campaign all by itself. But where it's furthest along, advertisers can do all that. You can let the agent read performance data, build and activate audiences, adjust the budget, launch campaigns, and even generate creative variations all through natural language.

26:27Letting one agent shift budget across platforms on its own is where the industry says it's heading. But that's not the standard practice yet. The industry direction is obvious, though. Fewer read-only endpoints, more agents with their hands actually on the wheel. and at that point governance and guardrails become the real conversation not the pipes themselves so you're talking about a lot of new ways that people can connect and companies can connect to each other so does that mean we're going to be solving for media fragmentation marketers run dozens of platforms to get to all that media dsps walled gardens with the clean rooms, measurements, they're all bolted together with their own quirks and logins.

27:13Each of them historically needed their own bespoke API integration, its own authorization, and some team member who knew how it all fit together. MCP doesn't unify the stack, but it has the potential to unify the interface to the stack. One agent could sit on top of all of them. As an example at Basis, we have Compass, our proprietary strategy and planning agent. It has direct integrations with buying platforms, including our own DSP. Advertisers use it to upload briefs along with their research, performance history, which is then synthesized into a proposed media mix, audience personas, and creative strategy.

27:55Initiating an omnichannel planning conversation through one interface. Okay. So if I'm thinking about MCP servers as a shorthand, it might help unify my interfaces if I'm a buyer, but the platforms themselves remain fragmented. Indeed. There's a lot of room for innovation under the hood. MCP is the unifying layer that is the command and control on top of it. Well, thank you, Mark, for helping explain MCP servers. I know I understand them a bit better after these few minutes. And thanks to Basis for supporting our podcast. Thank you.

28:35All right, we're back. And I want to quote you from LinkedIn again. You really are quite prolific on LinkedIn, which is very helpful to me when I was coming up with my questions. But you posted recently about how the biggest budget doesn't win anymore. The clearest answer does. I want to talk about what the clearest answer really means. And for those who are listening, I just did air quotes with my fingers around clearest answer, because a lot of brands, like they don't just compete on pure information. They're also competing on identity and emotional associations that they've built over the years with people.

29:17So it feels like there might be a risk over time with a gentic discovery that might eventually disadvantage brands that have spent a lot of time and money investing in this sort of intangible brand equity, like vibes, essentially, like in favor of brands that just have the most complete product data. I'm curious what you think about that. Well, it's a big topic. And in Cam this year, I'd say this is probably the most common discussion point was just like, what is the value of brand and how does that evolve? One little fun saying we have is that information is the new bid and really like information quality is something that matters a ton.

29:58But what I want to do is broaden the term information. It's not just like what is on the website because we were talking before that agents will triangulate. And when they go to triangulate and they listen to people say, well, what are the best, you know, car brands or the safest cars or the best pens and paper, you know, back to that example, other humans have made recommendations with rationale. That's all going to get ingested too. And so I think that if you're a brand and you're, thinking like, what is the consideration set? The human status quo is the starting point. And then obviously things will evolve from there.

30:40If you're just thinking, I'm going to go create a brand new brand and document to the nth degree, everything that we stand for, that's great. But when the agent goes to corroborate that, what you stand for has no bearing and no reference point to triangulate from. And so the agent is going to be unemotional about reasoning over that and saying, well, wait a minute, this is a brand new brand. Or this brand just got acquired by a conglomerate. And when they did, quality dropped. Or they made a claim about how they source their materials ethically or organically. Those claims can now be evaluated and either refuted or reaffirmed.

31:21And so I actually think what this will help is really separate the brands that are true and what they say shows up in the community, shows up in the product quality, and that those brands are going to be just fine. One thing that's cool about this world is you actually, the same way you could do UX research on humans, you can do UX research on the agents. And the same properties apply, which is they don't get tired. They explain why they do the things that they do. And if you are a brand today, I would encourage you to go to each of the different, you know, major labs models and interview them.

32:02Ask them what they think of your brand. Ask them why they think that. What sources are they going to? This is the first thing that you can do once you're sort of available and not blocking. How do you go from that to getting recommended. The same way you would talk to an analyst and you'd say, well, what do you like about our strategy or our product or our services? What do you dislike? And it's an opportunity to improve. I don't think of it though, as like gaming a system. I think that that element of the human internet, I hope that element goes away. And this creates more enduring positive feedback loops so that the things that you say, you have to actually back up for it to survive agent scrutiny.

32:45Well, that's a perfect segue to talking a little bit about GEO because there's this whole cottage industry that's popped up around generative engine optimization or like AEO, answer engine optimization, like other acronyms and consultants, tools, frameworks. They're all promising that they can get you cited by AI. And some of it feels useful and some of it feels like SEO rebranded and some of it feels like straight up snake oil and no one really knows what they're talking about, but they pretend that they do. And to be fair, to be fair, to be fair, I don't, to be fair, I can speak. A lot of these rules, they're not really rules because they're changing really quickly.

33:33So the question is like, what can marketers do to separate what's real from the noise because it's so noisy out there. And I know that just from the pitches that I get from GEO companies that want us to write about them. I mean, we get pitched every single day. I think that there's no question people are looking for an advantage and these companies offer short-term advantages. I would separate out two types of assistance that you could get. Type one is everything we've been talking about, which is be true to the products and services that you offer, explain why they're good, and make sure that everything else that's on the internet is consistent and reaffirms that message.

34:18And if you have people that return your product a lot and you distribute through a retail channel, and that retail channel says that there's a lot of returns for this item or it has a low star rating, that's actually the thing to address. is why are your customers not keeping your product? And no amount of GEO is going to change that fundamental truth. The second part of what these companies offer is often a service where they will go and they will kind of manually ask a bunch of hypothetical queries to each of the different AIs. And they'll report back on how do you show up, how's that changing over time, and what might you do to trick that version of the model.

34:59I think that's a game that every business needs to decide how much they want to play in, because it is going to change over time. And you better believe that the team behind the model is looking at that and saying, how do we avoid getting tricked? How do we only reference sort of the information as the quality item that we care about in service of the user? And I think that that'll be a transient game. So we do not play in that space. and we advise really on that first element, which is do the things that you say. If there are issues in your products or in your ecosystem, they will be found out, but that's the place to invest.

35:39And things like clarity will give you a sense of where that's happening, what are people saying, what should be the highest priority items. What if your competitors are saying bad things about you? Just like shit talking, How do you avoid getting a dinged for that? Because somebody asking for a recommendation for the best XYZ, like they might go to a model and the model will say, well, avoid this thing because clearly it's problematic. You should go with this other thing. And I'm sure the models are trying to fight against that. But it feels like one way you could game the system. Well, one thing that's so fascinating about AI is because they've modeled a lot of the things like how our brains work and how humans exist in social structures, it equips us for a challenge like you've just raised.

36:36Well, what would you do if a competitor was putting out bad PR against you and it was sticking? Well, we know what we would do. There's many different tactics that we might choose. Like we could refute it. We could call it out as bad behavior. We could provide facts and evidence to refute the claim. All of those techniques that should have worked in the human world should also work because the agents are going to consider that information and be extremely logical. Now, depends on each company's situation, what would be on brand and feel authentic for them to refute it. But I think those same things are going to work well.

37:11And you can go and interview the model. And you can say, why are you continuing to perpetuate this message? Maybe you will actually learn some treats about what is being claimed that then you can also go and productively address. I mean, that's the thing that's crazy here is like if there's one piece to move from being visible to getting recommended, go ask each of the models why they're doing what they're doing. It's free research and they will just tell you the answer. It's really good advice. I'm actually I'm going to co-pilot now because I want to ask it about me. So I'm going to ask you a question and you can answer it.

37:50And while you do, I'm just going to see what Copilot thinks about me and what it thinks about you. I want to do some research about us. Okay. Let's see what comes back. But while I do that, let's talk a little bit about measurement because this moment we're living in, there's a complete measurement paradigm shift happening. Like for years, publishers, brands, everybody, they cared about certain things. They cared about traffic. They cared about people taking specific actions. Maybe there were some slightly mushier things. There are a lot of vanity metrics out there. But people come to a website.

38:23People click on thing. People hopefully buy thing, et cetera. But AI is answering the question that you have in a chat interface, and a person isn't leaving that chat. So if content gets cited, I mean, maybe people click through, but often no one clicks through. So in this world, what are the new metrics? What should businesses actually be tracking beyond sales and beyond the stuff that they've used as a directional measure, all that vanity stuff? And it's not a bad thing that vanity metrics go away. Well, let's start from the bottom of the funnel because that is where the ultimate conversion or value creation occurs.

39:07There's a whole set of standards that have been proposed and starting to get wide adoption around structuring the things that you might want to sell online and making them available with like an offer. And if you've heard of the Universal Commerce Protocol, UCP, this is a popular way to define and structure what you have to sell and do it in a way that agents can consume and reason over. So this is a popular standard and Microsoft supports all popular open standards on the web. That's kind of the approach that we've taken. But if you pair that with some of the agentic payments that are coming online, you really are able to say, I do want to be compatible with this AI channel.

39:54I do want to make my stuff available. I'll give like two adaptations that we've added on top of this protocol that I think are really interesting. The first is that you can actually add your loyalty association with UCP. And so we launched something with Target recently, where if you're a member of Target's loyalty program, you can link that and see member pricing and shipping and all your rewards and offers in one place. And so it's an interesting thing where if you've already chosen a retailer or a brand that you love, this is now a system that allows you to authentically link and show that expression.

40:34And then the AIs have the opportunities to carry forward that relationship into whatever AI experience exists. Another one that's relevant in the commerce space, I'm sure that you've made a considered purchase and wondered, is there a coupon available? And the process of going - I always wonder where there's a coupon available, whether it's a considered purchase or not. That's right. And maybe, you know, you spend more time for larger purchases to look for one. UCP actually allows you to test the coupons ahead of time in an AI loop so that you don't have to go and copy and paste over. And it can say, oh, I already checked all the coupons and I've applied the best one for you.

41:15These types of consumer benefits are going to be awesome. Microsoft is building these experiences into all of our shopping surfaces in Edge and in Bing and in Copilot and things like that. But this is the first rendition of what's possible. And so when we think about all the things that help remove friction at the time of conversion, which is sort of that ultimate measure at the bottom, I think that there's going to be new efficiencies that were never present in the human era because of the benefits in the shape of AI. Now, if we move up the funnel and we consider awareness or consideration or other types of mid-funnel touch points, I think that they're going to match a lot of what we know today, which is, did my brand get considered more often?

42:05If it did, then how did that show up in recommendations to its user? And I think that the issue really is that we're saying we have the human who hired an agent. What are the agent metrics? And I think a lot of the agent metrics are going to be very similar. They're just proxy metrics because the agent is screening out a lot from the user. And the last thing I'll say here is that it's very rarely that just the brand or the retailer tells the whole story. Many, many things are triangulated and you look for third-party data. It's going to authenticate and verify the truth of what's the claim. So let's say I'm looking for safe SUVs.

42:46If I go to every carmaker, they're going to tell me how and why they are so safe. But I might want a third party that's tested them all. And I might want to look at third-party crash test data to be an overlay on top of that. and my agent can then take into account, well, what types of trips am I taking? How many people are in my family? And when I go and look at the report that comes out, the citations to the places that were referenced is going to be a major part of whether I feel comfortable to proceed with a car or a purchase. And so I do think a lot of the stuff, if you squint, is going to look very familiar.

43:25It just hasn't all been standardized yet. I've started asking this question of people because I find the answers very interesting. What is one thing that you would never, like ever outsource to an agent to do for you? And what's something that you already have an agent do for you? And it's been a good decision. Let's start with the latter because I'm constantly looking for use cases like that. I do a bunch of stuff at work where I create like a morning briefing for the day. And it really highlights, like within the Microsoft ecosystem, Copilot's quite good at going through like my IMs, plus my email, plus calendar, plus docs that I've been working on.

44:10And kind of gives me a like, well, what has changed since you last looked viewpoint? And then I've instructed at the types of things and people that are really important. So it's actually quite opinionated about like, what should I do and what should I look at? Is it perfect? No. but then I've also created loops so that it learns from what I ultimately go and spend my time on. And what's amazing about it is it's all passive. I don't have to tell it every time. It is recursively improving as it goes. I think this is going to happen more and more, but it doesn't change my workday in that it just focuses my attention on what's going to be most important.

44:49So I think that's huge. There's an example I like to give with visuals and everything, but it's around. So this is on the what do I think I'm never going to outsource. I was planning a party. Both my parents were turning 70. So it was like a big milestone. And I was trying to get everyone together and recreate some of the summer lake vacations that we took as kids. That was my one vacation every single year was to go to Wisconsin where my family was. And in doing that, like, it's so fun to both have some of that nostalgia, like we were talking about with Back to School, of what those vacations meant.

45:28And to talk with my family about, like, what was awesome about it. I mean, getting bit a bunch by mosquitoes is a byproduct, but that's also part of the authenticity of it. And so when we were trying to recreate that, it was really cool, like looking through different options, kind of figuring out what experience would feel authentic to it. and so while I might even use AI to screen some of that just so I can cover more ground when it really comes time to like well what place do I want to take my family for a long weekend or a week that judgment is just as much a reflection of like me as a gift giver or a party host as it is just getting to a decision or an outcome and I think those types of things that like really wrap up our humanity and emotion together I don't see that stuff going away for most people for a long time, if ever.

46:23For me, never. That aligns with what I've been hearing from people. If you enjoy it, if it gives you pleasure, even if you could do it more efficiently, whatever that means, then don't use the tool. You don't have to. But if you can use something to automate a tedious task or surface information to yourself in a way that's actually useful to you, then yeah, go for it. We are still humans with agency. And that friction is in many ways like the joy of life. Like it's good friction. It's fun friction. Right. I mean, if you have to make an effort to get something done, then you're proud of the result at the end.

47:03So do you want to feel really good about yourself? Yes. I just I asked a co-pilot, what do you think of Tim Frank of Microsoft? Tim Frank is one of the most strategically important leaders inside Microsoft's consumer AI organization. He's effectively one of the people shaping how Microsoft turns AI innovation into sustainable business models across Copilot, Bing, Edge, MSN, shopping, payments, and even in partnership with Xbox and LinkedIn. So there you go. Sounds like a fun job. I mean, I'll just say the lived experience is equally awesome. working with all these teams and trying to bring everything into the AI era.

47:43It's an incredible set of assets and properties that Microsoft has and a lot of talented people working on this together. So to bring us home, let's say we were to have this same conversation in, I don't even know what timeframe to use because things are changing so quickly, but I'll just say two years, just fully acknowledging that in two years, is everything I think I know about this will probably be embarrassingly wrong. What's one thing that you think is true and will remain true? And what's one thing that you think will age like milk? Wow. It's interesting to reference two years in the future because it has not even been three years since the big LLM moment.

48:28Yeah, that's actually a great point. Right. Right. Just as an aside, I think it's funny when I see like job requirement listings that ask for like 10 years of experience in this stuff. It's like you you cannot have 10 years of experience in most of these fields. Look, I think that Jevin's paradox and the growth of AI and the the the explosion of performance and demand because costs fall is going to be massive. And I think that that is not going to change. We don't yet have a new Moore's law, but it really in two years should dwarf the human internet in terms of traffic and is in terms of like impressiveness of these systems.

49:14I would say this calendar year to date, and we're only in the ninth month, the way that we build products has already meaningfully changed at least one or two times within the year. And the way I would say that most simply is that as a product person, it used to be that you directly built the product. And now a product team builds a system that builds the product. And that separation just is wildly different from what has been happening for years. And I think that it'll continue to change and adapt massively. And the disciplines and the job titles that we knew, say from 2020, might be unrecognizable in two years in terms of what people do.

50:02So I would look to that. And I don't think you'll even have to wait two years to see this in the most forward frontier companies. What I think is going to change is I love the natural conversational interface, but I don't think that all successful AI products are going to be a chat interface. And I think that we've been so enamored with what's possible because these models are so good in words. I think that we're going to have a lot of other interfaces, and I think that the same technology is going to be adapted to many more places. I'll just give you two quick examples. One is in shopping, which we've talked a lot about.

50:43If you go to shopping sort of pre-LLMs, it's a lot of grids with pictures. And I think that there is a lot of room to change what that looks like for the human interface. That'll be much more engaging and enjoyable. This is one. The other thing is LLMs are good at next word prediction. Like that's kind of how they started. And they've obviously grown in capability. If you though separate out like next word from other things like behaviors, emotions, like there are so many other sequences that are around us in our life. And I think that the core technologies are going to start getting applied to many other domains.

51:24I think in medicine, this is going to be very pronounced. I also work very closely with health team here. There's a lot of cool stuff. We don't know all the things that are going to work, but I think that today's world of I'm just going to like tell the words to a conversational agent, I don't think is going to be the only interface in the future. I totally agree with that. I think back to how publishers used to put their stuff online. They just made it look exactly like a magazine because, well, one, it was hard to imagine what else to do. And two, it was familiar to people. So it was almost like a stepping stone to the next phase.

52:03But what it's going to look like is hard to imagine because we're just going to have to wait and see where it goes. I mean, the limits of our world are the limits of our imagination, actually. There's no reason why things have to look or feel or be interacted with in the same way they always have. So eventually we'll break out from that. We're totally in that phase where you take the old medium and you jam the format into the new medium. And we haven't yet done sort of the native re-articulation of what should this stuff look like. I think that's a really simple way of saying it. And I'm excited for it.

52:40It's going to be different. It's going to be powerful. So when we do this podcast again in two years, I'll just have my agent reach out to your agent and set something up. I won't be blocking. I'll have good interfaces for it. You know, there'll probably be some good identity verification that'll come online by then. And we could see how much of this was way too ambitious or way too naive about all the fun. I will hold you to everything you said. Oh, no.

53:29strategies, and media plans that integrate directly into omni-channel activation. Learn more at basis.com.