Subscribe
Share
Share
Embed
Serious People Podcast: An Operator's Manual to AI is a podcast for business leaders and operators running complex, real-world businesses. The ones selling supplements, managing caregivers, or running service crews. Not software.
Host Noah Levin brings nearly two decades of experience at Amazon, Whole Foods, and healthcare tech to weekly conversations about what AI outcomes work inside a business, and what doesn't.
Equal parts practical and irreverent, but useful above all else.
Christina Johnson (00:00):
Look at your sales data. The number of people who don't actually look at their sales data along with their marketing efficiency data is insane. You can run a really efficient campaign that drives no sales. The example I have for that is that someone told me they ran this super efficient marketing campaign in LA, New York, big cities. Do you know what it did drive any Walmart sales? Because there aren't Walmarts in those cities. So looked great from a marketing metric perspective, but the second they looked at the sales, that's a pretty big whiff. And that's where you need to look at your actual business metrics alongside your marketing metrics.
Noah Levin (00:34):
Welcome to Serious People. I'm Noah Levin. Today my guest is Christina Johnson, founder and CEO of Advisar, an intelligence tool for retail media buyers. If your business uses performance marketing and what business doesn't these days, you're going to love this conversation. We get into why analytics are so hard to get right, what happens when you get it wrong, and how do you use tools like Claude to make better decisions with the data you have. So if you or someone you love is spending gobs of money on ads, this one's for you. We know each other from business school, from Michigan. How did you end up there?
Christina Johnson (01:10):
It's a windy story. So I moved to LA, I was like, "I want to work in film and TV." Managed to get introduced to the producers of Grey's Anatomy right when they sold the idea to Grey's Anatomy to the network. We were developing the pilot. I was the junior person. I was basically photocopying scripts and getting M&Ms, not doing anything fancy, but got to be on this ride where Grey's was nothing and all of a sudden it was nationally famous, which was interesting. But I just realized I wanted to have a life and you just cannot have a life in the film and TV industry in the way that I wanted to be able to exercise and not always be stressed out. I was always on 24/7. So tried a couple other things like nonprofity, wound up at all places in the executive office at Goldman Sachs doing also junior type thing.
(01:57):
Wait,
Noah Levin (01:59):
Wait. Sorry. How do you go from getting coffee on the set of Grey's Anatomy to the Goldman Sachs office?
Christina Johnson (02:07):
Yeah, so it was the 2008, 2009 downturn in New York. And so I'd been in LA and I was like, "Okay, it's time to be back to New York's where my family is." And I was just looking for jobs and it was really hard to find a job. And so submitted my resume to this placement agency who loved the Hollywood background and was like, "Oh, you'll do well on Wall Street because you're kind of unflappable with famous people." And so interviewed in and they were like, "Yeah, I got placed on that floor." And it was right when the CEO of Goldman was being interviewed by the SEC. It was peaked Occupy Wall Street era. But I was like, "I mean, I'll do anything for a story, really, Noah."
Noah Levin (02:48):
So did it turn out to be true that having worked in Hollywood on the set of Grey's Anatomy, you were well-equipped to work on Wall Street?
Christina Johnson (02:56):
I would say yeah, the skills translate and I would say the skills translated to Walmart too. Getting good at observing and not needing to insert myself into a situation and listening has served me really well from all of those situations.
Noah Levin (03:08):
I want to fast-forward and get us to the point where you are at Walmart and finding yourself learning about retail media and how wildly backwards it is.
Christina Johnson (03:20):
So basically at a high level, and Amazon was the first mover on this, retailers started to realize that they were sitting on huge piles of data and that they could use this data to target shoppers across their platforms. So the example I always give is Walmart knows so much about its shoppers that an advertiser, say I have a toothpaste I'm releasing and I want to be able to market to moms with kids age five to 12 with my new toothpaste. I can go to Walmart and say, "Hey, I want to buy the ability to target that mom when she is searching on your site for toothpaste. You know all that about her. I want my ad to come up when she searches toothpaste or when she searches something else kid adjacent." Up to that point, it was really, really hard for marketers to prove out value.
(04:12):
They call it closed loop attribution. It was really hard for marketers to be like, "Well, I ran these ads and then I can close the loop and say definitely drove these sales." And on retailer sites, because the shopping is happening on the site, you can theoretically, and there's a lot of controversy around this, but you can theoretically say, "Hey, you ran that ad and then someone clicked through the ad and they bought the thing, so the ad drove the sale." Now, did the ad actually drive the sale up? I'm not going to go down that rabbit hole, but it was sort of a new way of being able to measure advertising. Retailers realized how valuable this was because when they sell the advertising like that, it's not like selling something that's a product. You get 50 to 90% margin selling advertising, then suddenly you can prop up other parts of your business and do all sorts of fun things with those profits.
(04:57):
I was working for someone whose job it was to grow the advertising business, and I understood, even as a marketer, I understood so little of it that I was like, "I must be an idiot because people are talking about things and I don't know what they're saying, and it sounds crazy to me."
Noah Levin (05:13):
What's an example of something that they were saying that didn't make any sense?
Christina Johnson (05:19):
It was just all these random metrics. They would just talk numbers at you. And I come from this marketing of you build a strategy and who are you targeting? And there are members at the end, but these people were just throwing out the fill rate was this and the click-through rate was this. It felt like gobbledygook. And so that was when people really didn't know what was going on. So I was not stupid. I was just listening to other people's, also trying to figure it out and thinking they knew more than me when really we were just all clueless together.
Noah Levin (05:50):
Tell me what your job was at Walmart when you were working on retail media.
Christina Johnson (05:53):
So I was the lead strategist and then the chief of staff to Janey Whiteside, the chief customer officer. Our team was tasked with things like figuring out who the design target should be for Walmart at that time, who Walmart should be going after, how should we be talking to them, what are the services they want? We were in the front of the board or she was in front of the board all the time because we were building up things like Walmart Connect and Walmart+ issues, Walmart's membership, which were kind of newer additions. So spent a lot of time incubating new businesses for Walmart. That was my job.
Noah Levin (06:25):
So you're inserted into this business that has a mandate, it has traction, it doesn't have scale, but it has this inevitability of we know we can get there, we are Walmart. Were you operating this business? What was your role?
Christina Johnson (06:36):
So there was a time at Walmart, and this happens at almost every company that starts a retail media network that does not come from a digital background, is like you have to get the leadership of the company bought in on doing a media company. Walmart at the time was a 50-plus-year-old brick-and-mortar retailer that had just started doing e-com seriously. And so a lot of the work that my boss did that we were doing was convincing the C-suite that the sales team for Connect did in fact need to be incentivized by sales targets and budgets, which was just new at Walmart at the time. So it was figuring out what the org structure needed to be like, convincing the C-suite that those changes needed to happen. One of the big things was you can't run a successful retail media network without buy-in from the rest of the organization.
(07:29):
So what does that look like in terms of KPIs for the merchants? So I was doing a lot of that higher-level work.
Noah Levin (07:36):
Tell me at what point retail media kind of became your thing and when you first found yourself in a role where you were actually in the shit with retailers trying to explain to them why they should spend their money or why their money was spent well or not spent well.
Christina Johnson (07:56):
It kind of fell in my lap. I worked for the head of Walmart Connect and then consulting firm, and it just so happened that retail media was just exploding at the time. And so we ended up getting a whole bunch of work consulting for other Fortune 500 companies that were building up retail media networks. And it was, how do you build a sales team? How do you get organizational buy-in? What are you offering suppliers? How are you selling into them? It was setting all these fundamental things. It's funny that everybody assumes everybody else has it all together. They'll be like, oh, well Walmart knows what they're doing. Oh, well so-and-so knows what they're doing. Everybody is maybe strategically guessing, but it's like saying they, who is they? None of they have it figured out.
Noah Levin (08:37):
You mentioned a specific meeting when you were consulting where it sounded like no one knew what was going on. You want to tell us about that meeting?
Christina Johnson (08:45):
I was consulting for a retail media network and we were talking with some of the sales team and we were just like, well, how are you communicating to suppliers what they're getting? What do you mean? We're showing them the placement and maybe we talk about impressions. We were like, but sales lift or. And it just became really clear that the sales team themselves didn't know how to look at the data in a way to make an argument to suppliers that their spend had worked, that they should be spending more. And that was so much shocking to me because this was a retail media network that was selling hundreds of millions of dollars in ads. So it meant that there was suppliers on the other side just also taking this word that things are going to work. And so it was just shocking to me to see the people doing the selling, A, not having the data or the information to be able to justify to convince people to sell more, but then that the suppliers on the other end weren't asking those questions.
Noah Levin (09:40):
Why weren't they asking? What is it about this format that just they weren't pushing hard enough?
Christina Johnson (09:44):
I think there's variation and sophistication in understanding marketing because retail media has suppliers who are not historically marketing organizations needing to spend marketing dollars. There can be some organizational pressure or they feel like if they're selling at a retailer, then they need to spend on their retail media network in order to get visibility. So it's a new muscle for a lot of people. So some people I think just didn't know or though they were just servicing the function. And then I think others, like I was saying, I was totally confused when I first started in retail media. I was like, I don't know what any of this means. I think people don't want to look stupid and so they don't ask the question.
Noah Levin (10:18):
If I wanted to be a smart retail marketer, go spend a pile of cash, but spend it efficiently on my toothpaste brand, what are the questions I should be asking? What are the metrics I should be looking at?
Christina Johnson (10:28):
I mean, there's internal questions that you need to be asking, and that's for the people who are not used to marketing. What are the goals of this campaign? Are you actually trying to just drive sales or are you trying to drive awareness? Because if you're trying to drive sales, your sell-through and your unit velocity are the things you're going to look at as KPIs. But if it's awareness, you are going to look at things like reach and frequency, your new-to-buyer percentage, so you're bringing in new people to your category. So getting clear on that internally on your own side is good. And then asking a lot of pointed questions to the retailer, what are the incremental sales that came from this campaign? Not the sales that would've happened anyway that you're claiming credit for, but what sales can you tell me above and beyond were driven by this campaign?
(11:09):
And that's a really hard thing to answer. I don't envy the retail media networks. It's not simple, even though it sounds so.
Noah Levin (11:15):
Tell me when it goes horribly wrong. Is there an example of where someone just blasts the money cannon in the wrong direction and has nothing to show for it?
Christina Johnson (11:26):
It can go horribly wrong a couple of different ways. So more spend doesn't actually necessarily mean better marketing. Marketing is more nuanced than I think people think it is. You need to get surgical about who you're trying to target and then what tactics you're going to target. And this happens all the time where people are like, oh, I'm being told to just spend more. And you're like, but that doesn't actually do anything.
Noah Levin (11:49):
Well yeah, you mentioned there was a supplier that had to use auto bid or was being told to use autobid. Oh
Christina Johnson (11:54):
Yeah. There's a function in retail media where you can just basically set it and forget it. It's not actually really that effective. They were being told by the network to do that. They actually ended up spending $0 and moved zero items because they weren't being picked up by the algorithm. So it just fell flat even though they weren't being told it was the right thing. And so we ended up doing a more manual campaign, which is more work, and then they ended up selling through their whole pallet of items. So it just goes to show there's a lot of incorrect information out there and it's really easy to just misspend or pick the wrong thing. But also it's okay to do that and fix it. Marketing is forgiving.
Noah Levin (12:27):
This episode is sponsored by Valet. Christina and I talked about how much easier it's become to use AI to analyze your data. Once you've built something useful, you need a way to share it. Valet lets you publish the dashboards, analyses, websites, and small tools you build with AI at a stable URL. You can keep them private for your team or make them public. We used it to make a companion page for this episode with three lessons from Christina, prompts you can try with your own campaign data and a worksheet to organize your review. You can find that page in the show notes and to publish something of your own, go to valet.dev in your agent today. Today, you're the founder of Advisar, which is an AI-driven product that helps marketers understand their spend in retail. What got you to the point of realizing that AI was a tool you could deploy against this messiness and worldwide ignorance?
Christina Johnson (13:26):
When I was a hands-on marketer, I used to have to cobble together all sorts of disparate sources. I'd have to pull massive spreadsheets to get to any sort of answer about whether something worked or not.
Noah Levin (13:36):
Where were you pulling data from?
Christina Johnson (13:38):
I mean, everything from Walmart's internal point-of-sale data to reach and frequency information for TV ads, digital metrics, like click-through rate, to things like Nielsen data. It really was just disparate, and it was so dependent on whether you happen to think of the thing at the time. I can't even imagine being a marketer with just Claude right now, even at its base. You used to just have to run around and ask people for information and pray that they gave it to you, and it was very happenstance, but I learned to get pretty smart with cobbling together things that would give me directionally correct information. You could see that the campaign worked because people were adding the product we were advertising to their basket at a higher rate. Is that a direct correlation? No. But is that a piece of directional information that's helpful? Yes.
Noah Levin (14:27):
That's looking at before campaign and after campaign as opposed to a holdout.
Christina Johnson (14:32):
Yeah, no, it was looking at year-over-year data actually because it was seasonal. So with strawberries, if you were looking the month before and the month after, it might give you a misread because strawberries have a season, but looking at the strawberry season the year before and what the basket ad was versus the strawberry season during the campaign is what we looked at. So first when I encountered AI and I saw that retail media was a problem, I was like, "Huh, I bet there's a way to get all these disparate sources together with AI." And then people started showing me their retail media results, and for whatever reason, a lot of them are inflated. I have people who had data and it's gotten better since, but someone showed me something and it was like, "Yeah, I have a 33.7 ROAS."
Noah Levin (15:15):
ROAS is return on advertising spend. Return on ad spends. Yeah.
Christina Johnson (15:18):
Yeah. And so there was just some murkiness in the numbers, but they were getting frustrated because you can't make a good decision on how much to spend if you can't get at least a directional read on impact. And saying that their spend drove 33X sales is just not within the realm of, I mean, shoot, everyone should be spending their ad dollars there. And so then I started to play around with the smart tech dev and be like, "I bet we could get to some directionally better numbers based on their own sales data and the tactics I used to use, and I bet AI could make that go faster." So that was sort of the genesis for all of this. There's a lot of talk about all these different ways to measure things and they're valid, but I was like, "I also feel like you can just get some down-and-dirty, old-school blocking and tackling that AI makes faster for you.
(16:04):
You can also just look at your data in specific ways." I was working with someone who ran an influencer campaign and she was like, "Well, I can't tell if it did anything." And I'm like, "We can also just look at Google search around the time you ran this campaign and see if it searches for your item went up." It's not scientific, but it'll directionally at least say, "Hey, we think this drove some movement." I will die on that hill that you can get a good directional read from the data that you already have. And if you want to get fancy, you can and should, but it's not essential to run good marketing.
Noah Levin (16:37):
Let's imagine for a moment that you are sort of a novice performance marketer. You're running your business, your business is not all about performance marketing, but performance marketing is one of those things you have to do. You're managing a CAC to LTV ratio, trying to get your cost of acquisition down, your lifetime value up, and you aren't yet at the point where you want to buy a product like Advisar. You know you've got Claude, you know you've got a bunch of data sources. Where would you tell that person to start?
Christina Johnson (17:04):
This is so embarrassingly basic. Look at your sales data. The number of people who don't actually look at their sales data along with their marketing efficiency data is insane. And where you can get sideways on that is if you are running search or you're running a campaign and you're just running for marketing metric efficiencies like click-through rates and CAC, you can run a really efficient campaign that drives no sales. And the example I have for that is that someone told me it was a big supplier and they were running their Walmart budgets. They were trying to drive Walmart sales and they were optimizing the campaign for marketing metric efficiencies, so low cost, et cetera. And they ran this super efficient marketing campaign in LA, New York, big cities, and then they didn't drive any Walmart sales. They didn't drive any Walmart sales because there aren't Walmarts in those cities.
(17:57):
So looked great from a marketing metric perspective, but the second they looked at the sales, that's a pretty big whiff. And that's where you need to look at your actual business metrics alongside your marketing metrics and sales, unit velocity, all that stuff.
Noah Levin (18:11):
What's the marketing metric that looks good while sales don't move? How is it possible?
Christina Johnson (18:17):
Return on ad spend.
Noah Levin (18:18):
So what is the R in ROAS if not sales?
Christina Johnson (18:21):
No, it's sales, but it's a fraction, right? So it's like a numerator over denominator and I can inflate that however I want. If I want to put my total sales for the time period divided by ad spend, that's going to give me a fantastic ROAS when really the ads didn't drive that amount of sales. So you can just manipulate it however you want it to look. Where it is helpful is if you're using the same methodology. So if I were looking at a Walmart Connect campaign rate and I wanted to see which tactics were effective, I could look at the ROAS numbers in terms of this ROAS was higher based on this methodology, so likely drove more sales. I can't use it to say this is the exact sales number, but I can use it in its own little set to say this was the highest performer based on this methodology.
Noah Levin (19:09):
When you say that I have to look at my sales, do I have to look at it myself or can I have Claude look at it for me? Yes,
Christina Johnson (19:16):
You can. No, for baseline stuff, Claude is quite good. First of all, I'm a believer in using AI, but I'm also a believer in understanding your stuff before you put it in AI. So if you're just feeding your sales numbers into Claude and not looking at them, that just opens the door for a lot of error. The thing about Claude that I've gotten caught up on when I've been using it, just because I use it all the time just to see what it has, is its training data only updates so frequently. So it's amazing at computation. I've had conversations with other data nerd friends and if you know what you're doing and you want it to do the math and you tell it what to do, it's great. For example, I was using it the other day and its training data ended in, I think it was May 2026 and I didn't know.
(19:58):
And so I was asking it something about Walmart Connect and Walmart Connect had just changed its measurement attribution June 1st. And so everything that Claude was telling me was wrong because its training data stops before it had that information that the attribution had changed. So use with caution is what I would say. I do think you can get pretty far if it's just you and your spreadsheets. Use it. Users do some analysis. It's pretty good at it.
Noah Levin (20:24):
How do you mechanically get the data into Claude when you're using it to do analysis for you?
Christina Johnson (20:30):
CSV files. There's different report centers you can pull from.
Noah Levin (20:33):
Even in the modern era, you're still going into portals, pulling down CSVs.
Christina Johnson (20:38):
Yeah, there are API calls for certain size suppliers, but you have to pay for them. So it just depends on who you work for. But yeah, it's still very down and dirty. It's not a very souped-up system for most retail media networks.
Noah Levin (20:49):
Having an API would be really nice, but logging into a portal and downloading a CSV is also very easy to automate these days.
Christina Johnson (20:58):
Yeah. Walmart will shut you down if they find you doing that.
Noah Levin (21:01):
Will they really?
Christina Johnson (21:01):
Yeah.
Noah Levin (21:02):
I wonder who's going to win that arms race. That's interesting.
Christina Johnson (21:05):
I don't know. I will just tell you historically, because a lot of that and then scraping their site is the other thing, but you just have to be careful. There's a lot of permissions around even using your own sales data with Walmart, and so you just want to be careful. Because trust me, I've had that though too, because I'd be like, I'll make a million dollars if I just have a bot that just automates pulling CSV files, but it can get tricky really fast and you can end up losing access and being put on the naughty list, which you don't want.
Noah Levin (21:33):
So you've got to call it a folder full of CSV files. You go to the farmer's market of data and you stop by each booth and you pull down your CSVs. Are you doing any additional sort of cleanup or merging of those data sets before you let Claude play in that sandbox?
Christina Johnson (21:49):
I'm at the point now, because we've been building tech, that we have purpose built that we don't have to do any cleanup because we don't think people uploading CSVs will be wanting to do cleanup. And so there's been a fair chunk of the build that's been like, take this very, very messy thing and let's put these guardrails around it and make it usable. So that's what our backend has been trained to do.
Noah Levin (22:07):
What kinds of things go wrong if you don't clean up the data?
Christina Johnson (22:11):
A lot of the things that we come up with are if you start to compare across the data, for example, if you want to compare Amazon data to Walmart data, Walmart and Amazon are in different retail calendars, they attribute different ways, and so you have to train your agents to recognize that or else you're going to come up with some really incorrect things. I mean, the other thing is just not every report has the same information in it. So if you're wanting your agents to look at reports, you've got to make sure you tell them where to join and that there's at least like a UPC or something to join. And so that's the kind of backend stuff where we've been like, look at this to join and then look at the data together. And it's pretty good once you do that. My poor data engineer one day was like, "What do you mean Walmart's on a different calendar?
(22:55):
What is this calendar based on?" And I was like, "It's a 60-year-old company. It's a lot of legacy things that need to be taken into account."
Noah Levin (23:02):
If you want to set up marketing for your company from scratch in a way where the data that comes back to you from all these different places is understandable and attributable and chopped in the right formats, and the campaign that you're running on channel A lines up with the campaign you're running on channel B with the same creative, are there any best practices?
Christina Johnson (23:21):
Hire a professional. That gets really messy because the thing is, I mean, you see a lot of AI companies actually spinning up and doing this right now. It's like we can pull in all your different marketing channels and show you a dashboard with all of them and compare some, but it involves understanding the nitty-gritty of each single campaign, each single tactic. So the bite-sized example I'll give you from my world is one of the things with comparing Amazon and Walmart retail media spends is that, I think it's for search, well it's for all of them, Walmart has a 14-day attribution window that it'll attribute sales, whereas Amazon has a seven-day attribution window. So all of a sudden you're not comparing apples to apples even though it's the same tactic, and you've got to figure out some way to normalize that data to be able to compare it.
(24:10):
I think that there's a lot of cool AI tools out there that you can connect all your different sources to that are not going to be that expensive, that protect you from making all sorts of human error if you don't know what you're doing. And I say that as someone who works in this space, but people are like, "Well, why don't you just build this yourself? Or what would prevent somebody from just building what you're building with Claude code?" And I'm like, "Well, theoretically they could and theoretically I could, but the realities of building in Claude or any of the other Perplexity or what have you, as a non-technical founder, A, don't know what I'm doing. It can spin something up that's pretty. Prototyping is not a problem, but understanding all the little details and being able to tell the tech how to get around those details, understanding how to make sure everything is secure.
(25:00):
I've had friends who've lost weeks trying to build something themselves and then it keeps breaking and then they got to fix the broken part. There is still a gap between your layperson and AI technology and building and a technical person that I think a lot of people don't appreciate quite yet.
Noah Levin (25:16):
I'm hearing you say two things that I think are both so true. One is that there's this uncanny valley of software development where you can get something that kind of looks like it should be very close to a production-ready tool, but in fact, there's this giant chasm of things that need to happen to make it maintainable and secure and actually work for multiple users and all the things that SaaS has perfected over all these many years. And then there's the other thing, which is I hear you saying that there's this domain knowledge that's super arcane and doesn't have any value necessarily outside of the world of retail media in your case, but the fact that Amazon and Walmart have different attribution windows or different retail calendars or that Prime Day was an extra day long last year, so we need to account for that, that domain knowledge is, it might be in the training data somewhere, but it's not necessarily going to get shaken out into the code.
(26:16):
If you want to build a bespoke tool set, you need to BYO expertise.
Christina Johnson (26:23):
And that's where for me building what's been so interesting because, so my tech team started in June. They spun out this beautiful MVP in Claude in two days and I was like, oh, this is awesome. And these guys are fantastic, my data engineers, a computational biologist by training, he knows what he's doing. It's taken a couple months to get something where we're like, okay, we think it's not hallucinating on the numbers now. It's been trained with my marketing domain knowledge. I think I even wrote out every single marketing question I could possibly think of and then floated it in front of my wife who's been a marketer for 20 years. So you have to really program it with the human knowledge, which I think a lot of people, including myself, sort of missed. I've found AI tools are very much a reflection of the humans who've created them for better or worse.
(27:13):
And I see it in my own tool where I'm like, oh, that is my fault that I just created in digital form for somebody else, which has been really interesting.
Noah Levin (27:21):
Tell me about the experience of recognizing that there's a problem that could be solved with AI and then actually building an AI product to solve it.
Christina Johnson (27:32):
What I'm finding, at least with what AI is at the moment, is it's an incredibly powerful tool. I'm not taking anything away from that, but the detail work that needs to go into creating something that's actually useful, and then the customer work in terms of understanding what customers want is actually a pretty long and grueling road to get to something that is fully useful. I never developed true software. I can only speak from conjecture, but I would imagine the discovery process you went through with that kind of software still exists today. You can probably iterate faster after it, but people are still at the core of everything. And so understanding what people want takes a long time, and then making sure you're not giving them bad information also gives me agita every night that I'm going to give someone bad data. And so that's been a real focus of ours has been data governance and guardrails.
(28:25):
And so that has taken a while, worth it, but it's not the snap-your-fingers that I think everyone was believing a year ago.
Noah Levin (28:32):
What were the major proof points along the way? I imagine you had to do some of this work solo before going and raising money, going and finding the technical counterparts. What were you able to do on your own to build the proof of concept and validate the idea before you took the leap?
Christina Johnson (28:51):
So I was really lucky and I actually was in an accelerator that had free tech help. And so I had a tech dev who was pretty AI-savvy early on, but the second we were able to get some data and we built something around it and I started showing it to people and they were like, oh my gosh. And it wasn't even that sophisticated, but it was just being able to look at massive amounts of data and have it be distilled for you in a usable way got an immediate reaction from people around here. So that was the first proof point where I was like, oh huh, this is a thing. And then trying to figure out where I personally had something unique to add to it. We haven't touched on it much here, but I'm very aware of the differing incentives of the different parties in retail media and getting people to spend money.
(29:37):
And so you can see that play out in numbers. The only reason I created Advisar is because I was like, I feel like suppliers just need a third party that can tell them these numbers are correct or these aren't actually action on these numbers and here's how to make a good competent spend in retail media and/or adjust your campaign better without garbage of I'm trying to get you to spend more money. I'm trying to get you to. Because both agencies and Connect A lot of them are trying to do that because agencies are charged by amount of spend you spend with them. They charge a percentage and Connect is just wanting suppliers to spend more money. And so when I started articulating that thesis and it resonated with people as well and I saw that the tech could actually do it, that was my other thing is I can say this.
(30:17):
Tech actually going to be able to get to a more rationalized number? That was another proof point where when I started showing it, I just have been showing suppliers. I mean, I live in Northwest Arkansas, so two doors up the street is a supplier. So I just show as many people as I can to get as much feedback as I can. And that's been the thing that's validating all along because nothing else has really been easy. Raising money is challenging. Building an AI is challenging, but hearing people continue to articulate the same problem was the thing that kept me going.
Noah Levin (30:45):
What's the biggest thing about the product and the shape that it's taking now that would surprise you six months ago?
Christina Johnson (30:51):
I don't want to say it's so easy to make bespoke, but we can program business context in cross data, and then we just built a function where you can program your own questions that you want to have the agents react to every week. And it's like how bespoke these things can be made for not that much money is exciting to me. People are like, oh, what about tokens? What about how long it takes to develop? And building AI native is really nimble in a way that I don't think I've ever seen. And then I wish I as a marketer had the ability to build something that bespoke and then be able to ask it questions and get real answers to my own problems instead of having, like I said, trying to run around the marketing floor at Walmart and to beg people for data.
Noah Levin (31:35):
I also am having the experience of I can build anything. Isn't that a superpower? I feel like a wizard when I sit down and try and build software for myself. I also think there is a downside to losing the constraint of work is hard, resources are scarce, need to be prioritized, which is that it's so easy to build everything that you don't have to do the upfront editing to know that you're building the most important thing. And software is less opinionated as a result. There's a right way to use software that only has a few features, but if every B2B customer gets to ask for its own version of the same thing, you end up with this sort of feature sprawl. And you see this in some of the AI tools, by the way. We talked about this on an earlier episode of the pod, but there's Claude Cowork and Claude Code that do very overlapping things.
(32:29):
And it's kind of because it's possible for these platforms to build so fast that they were able to build what turned out to be two very much the same things at the same time for different audiences. The product manager of these tools was asked, "Is there a right way to use these? How do you think about thing A and thing B?" And she was very transparent. She said, "I think that's the consequence of moving fast is we kind of built things without a coherent vision." So do you find yourself doing much upfront editing?
Christina Johnson (32:55):
So for me, I had to do a lot of editing upfront because first of all, my fundraising took way longer than expected and I'm not independently wealthy. And so I wasn't resource constrained in terms of the tech, but I was very resource constrained in terms of money. And so that made me get very focused because I was paying people by the hour. So I was like, "Well, I needed to do these things."
Noah Levin (33:18):
Why was fundraising hard for you?
Christina Johnson (33:20):
First-time founder, pre-seed, which means I didn't have clients at the time I do now. I live in Arkansas. I'm a woman. It's a tech field. And fundraising is just hard. The investors I have now are great and the right investors and it was a long and painful journey, but it wasn't a slam dunk. In this whole fundraising time, I started out with someone who agreed to be a founding CTO, but didn't want to stay on very long. So he built me at my initial MVP, and I think what I hadn't realized was how good he was with data. And so he created this great, very credible MVP. He left and I ended up using a tech dev who was also very good, but zero data background. And so she ended up creating a next-gen version of one of the MVPs and I was just like, "Yeah, these numbers look good," like anybody, "These numbers look good.
(34:14):
I think it's great. Let's show it to the client." And he looks at it and he's like, "This is not correct." And then starts asking the chat function questions and he's like, "And this is not correct." And he's like, "I can't even give feedback on this because the numbers are so wrong." And it just goes to show you, that's the other caveat for people like me who are operators who are not technical people is you don't know what you don't know. And so there's so many different skill sets in the tech world as well. And both my tech devs were good, but in very different ways, and I hadn't realized it, and then I hadn't done the checks on the backend.
Noah Levin (34:54):
When you look ahead and you think about what this product that you're building is going to become and what's possible with the technology you're using to build the product, what do you think is going to be different about being a marketer in a year or three years?
Christina Johnson (35:10):
There's a lot of talk, and I think you've mentioned too, you've seen this in your project work, a lot of people are going to agentic buying for digital media. I don't know what I think about that yet. I think it definitely has a place and I think there are ways that it can help, but I think that, again, it's the garbage in, garbage out if you don't program it to take certain nuances into account. If you don't take it, it can make mistakes. So I think there are going to be things about marketing that are faster like that, being able to analyze data, but I do think the core skills are going to be there. I still think you need to understand who your customer is, and that is you. The AI is not going to understand who your customer is. I think you're still going to need to ask the right questions of what is this campaign meant to do?
(35:53):
Why am I doing it this way? And I think when you start to automate everything, you lose some of the nuance of the human understanding aspect of it. I've talked a lot about how I reflect a lot on how AI reflects my own flaws back to me, strengths and flaws. I think that's going to be the same thing with marketing. You're still going to need to have those marketing skills. You'll just have more data to act on them with. You'll just be able to do things faster and react faster. But that core competency of understanding what the marketing is meant to do and how to reach people, I think stays with humans.
Noah Levin (36:25):
I think I agree with you and disagree. I'll try an angle on for you.
Christina Johnson (36:28):
Okay.
Noah Levin (36:28):
Okay. Here's where I disagree with you. I think that life is just a series of prompts. The AI is always going to be the middle mile with context sitting on one side and some sort of an output that you do something with the action on the other side. AI is the computer in the middle. It's doing some computation on whatever you tell it. And so whatever is the thing that the AI can't understand, maybe it's who is your customer or what is the strategy for reaching that customer? Maybe it just needs the context a step earlier that you're using to make that judgment and it can do that part too. And whatever the action is, like go make the ad buy, maybe it just needs to understand that context and it can go a little bit farther down the chain and you can kind of grow it from whatever it's doing now outward or both directions.
(37:14):
So I think that if you look really closely at whatever it is that you're doing before and after the AI gets involved, it kind of looks like context in as a prompt, action out, even if you're the one who's doing the computation. But then I'll steelman your point of view, which is if everything reduces just to AI doing the work, there is something missing at the end of that process. There is a role for the human element in taking points of view and taking accountability and being credible. And this podcast as an example, there's a reason that you and I as humans are showing up on this podcast to present ourselves as the avatars of the work we're doing. And so there is a role in at least some businesses for the person to come in and inject their subjective point of view and kick it off kilter a little bit.
(38:03):
Mechanically, I think the AI can probably do all of it.
Christina Johnson (38:06):
So I guess where I got stuck is AI doesn't learn from experience in the same way that we do. So the AI is not in the store observing how people are shopping. There's a piece of the human experience that it just can't see. At least now I fully acknowledge that I don't know what's going on in Stanford and some AI lab and who knows, but I guess what it reminds me of in a certain way is when everyone was like, "Online learning is going to take the place of schools. It's just so much easier, so much faster." And what happened is it really became a hybrid and certain people were missing that full human experience because there's an intangible there that the online doesn't capture. And this isn't a fully formed though for me yet, but there is an intangible of the human experience that the AI can't quite capture that you and I experience being out in the world that we can then feed into the AI.
(38:58):
But I don't know that I'm bought into the AI is going to be able to replicate that without us feeding it.
Noah Levin (39:03):
All right. Well, we'll do a follow-up episode in 10 years.
Christina Johnson (39:05):
In 10
Noah Levin (39:06):
Years. It'll
Christina Johnson (39:07):
Just be our avatars that
Noah Levin (39:08):
We program to do this. I'm just saying it out loud now and I'm sure my agent will call you in nine and a half years and schedule it. So bringing it back down to earth, just to close us out, if you're a marketer listening to this podcast and you want to do one concrete thing to use AI to make better decisions around your spend tomorrow, what is that thing?
Christina Johnson (39:31):
Learn how to get all your data, pick a campaign, pull your sales data, pull your marketing data, learn how to develop a really good prompt to ask what happened in the campaign and just use Claude to analyze it. It doesn't have to get super fancy. That's my other thing is AI can do a lot in just as basic form. So yes, I'm building an AI capability for more advanced stuff, but you can get really far with Claude and your CSV files and you shouldn't be afraid to use it. That's my very basic advice.
Noah Levin (40:00):
Good advice. Who should be reaching out to you about Advisar?
Christina Johnson (40:04):
Brands or suppliers who are selling into retailers and they're spending in retail media and they'd want another opinion. They want help organizing their data. They want to pick my brain. We're in beta now. I'm looking for another beta partner, but really any questions around retail media, we're a good source. If I don't know the answer, I'll send you to someone else who does. That's my other, I'm not going to claim to do stuff I can't. So if you really want to have your incremental holdout test, I will send you to people who do that.
Noah Levin (40:31):
The moment I have any reason to calculate ROAS for my life, I will be calling you immediately. Christina Johnson, thank you for coming on Serious People.
Christina Johnson (40:40):
Thanks, Noah. Appreciate it.
Noah Levin (40:41):
Thank you for joining us for the Serious People Podcast. If you like what you heard, subscribe on YouTube, leave us a review on iTunes, or sign up for our newsletter at seriouspeople.ai/podcast. When we're not podcasting, Serious People helps businesses put AI to work in their daily operations. Visit us at seriouspeople.ai to learn more. The Serious People Podcast is sponsored by valet.dev. Go try building your own agent today at valet.dev or click the link in the show notes. See you next week.