← AdTechGod Pod
AdTechGod Pod
Ep. 133 How Fluency Is Automating AdOps Without Replacing Human Creativity with Eric Mayhew
00:30:18
Loading summary
Transcript32 lines
- [00:00]
A
This is adtech God and I command you to listen to this house ad. So if you're listening to this show, just know that you've really stumbled upon a giant network of content across advertising, marketing, media, publishing, and of course, the people that work in this great advertising industry. So go to market, check out all of our brands. We have multitude of shows from the Brand Forum, the Advertising Forum, the Monopoly Report, the Ad Tech God Pod, the Market, and more. We are bringing more podcasts to our network. We are consistently and constantly bringing on new shows. So check it out. Market or search for any of those brands in the app that you use to listen to this podcast. Enjoy the show and see you all soon.
- [00:49]
B
Welcome to the AdTech Godpod, your window into the world of advertising technology and the people behind it. I'm your host, Ad Tech God.
- [01:03]
A
Welcome to the ADTech Godpod, where we speak to the founders of our industry. Today's guest is Eric Mayhew, the Chief Innovation Officer and co founder of Fluency. Eric is the founder and with that comes a perspective many of us are interested in hearing. At least I know I am. His background was heavily focused in the auto space. He had worked@dealer.com, goodrich Corp, and more prior to starting to Fluency. So on this podcast I really want to understand how that pivot happened, what his background is and what made them start Fluency. Overall, just a quick disclosure, this is a sponsored episode, but as an entrepreneur, I love to hear from other entrepreneurs. So I'm really excited to have him here with me today. Eric, thanks for joining me on the pod.
- [01:43]
C
Hi atg, it's super nice to be here today. Thanks for. Thanks for having me. This is going to be fun.
- [01:47]
A
Thank you. I know we've been working so closely with Fluency over the last few years. You have an amazing marketing team team that's really been tied in close with ours over the last few years. I would love to learn more about you though, Eric, what your background is. I saw that you worked@dealer.com, i'm very familiar with that company, just FYI, having worked in the automotive space myself many, many years ago. So I'd love to hear about what brought you to your point today of founding this company and how maybe that background@dealer.com and others helped get you here.
- [02:16]
C
Yeah. Awesome. And thank you for going out with the marketing team. They are incredible. We're blessed to have them. So I'm super excited to get to talk.
- [02:24]
A
I hear everything, Eric.
- [02:25]
C
Yeah, I know. I'm super excited to get to talk about this today, you know, this has been a labor of love for a long time for many of us. I mean There are about 140 team members now in Fluency and we are all sharing this, this common mission together and it's, it's super exciting. So the kind of the origin story, I and three others started fluency in late 2017 and we did have a pretty clear mission in mind. The spark for the company came from actually experiencing the problem we wanted to fix firsthand. I, I, I could tell you the number. Well, I'll just say I've been, I've been in the space for about 20 years. So about 20 years ago I, I joined a rapid growth website company. So they wanted to add advertising services into their offering. You just mentioned who they were. We all know them well and they are an incredibly cool company. That was, it was a great time to be there. But you can probably imagine that, I mean they were a website company first and website tech scales pretty well and has great margins for businesses. That existing business was hyper scalable, hyper profitable while providing excellent outcomes and a high level of service for their clients like that. All those things happened all at the same time in that website business. So they brought me in and in comes this advertising business that I was kind of in charge of bringing into the business. Now I'm a technologist first, but I kind of led that charge from a technology side. We worked hard to make sure we could offer everything people might want to do in advertising. There were so many things, so many long tail pieces that they wanted to get to. And I'm guessing most people listening have probably used an ad platform sometime in their career. And you all know it's got tons of settings, tons of features, and what feel like a million different ways that you can set your campaigns up. So back then we built that system and started down that path. A lot of buttons to set up a lot of campaigns, any way you could imagine. But then you have to solve this practicality problem. Giving teams of people tools with a lot of flexibility is important, but there really just isn't enough time in the day nor margin in any business to spend all that time clicking buttons. We wanted to scale that business, but the more volume we had, we kept feeling the pain more and more that human button clicks don't get more efficient with growth. In fact, it kind of seems to go the opposite direction more as we added more clients and there were more customizations and more deviations and more, more requests, people would get tired. Their efficiency would actually go down rather than up. And it was, it was pretty challenging. There were far too many days when the last light on in the in the office was mine and a few people in adopts. And I knew there were far too many weekends that were interrupted for our ad ops teams because clients needed specials taken down on a Sunday because it was the first of the month or, you know, you can ins, or any one of those types of reasons why we have to come in on the weekends to go do that work. As more channels emerged and more sophisticated features became available in the, in the space, no person could really keep up. So the four of us put our heads down, our heads together and wanted to solve this problem for both ourselves and for others. We wanted to make the economics of the industry work. When advertisers are just trying to do the right thing, experiment with new strategies and get to the high value long tail of advertising, today we can feel that those margins get really compressed. You don't get to that long tail. And even though we know we want to do the compelling work, the great feature work that Google Meta, the open web all supply us, there's just not enough hours in the day to get there with button clicks.
- [06:00]
A
So let me ask you, Eric, working@dealer.com, now at Fluency, that automation piece is super important and the repetitive task piece I think is really important. So having worked in adoption many years ago and just for a brief period of time, I just remember a lot of it was clicking a lot of buttons, flipping back and forth from various platforms and running, to be quite frank, like really menial tasks that just required time. It was like exporting Excel docs, pivoting those Excel docs, making decisions on those Excel docs, and then implementing whatever your allow list, disallow list, your optimizations that you need to do. Is that like the core feature of Fluency? Is it more of like the automation piece and removing those often menial tasks that waste time to focus on strategy or, or what is the overall, I guess, elevator pitch of fluency that you would go to market with to say this is why you would use our platform?
- [06:53]
C
I think you're pulling it all the right chords there. So whenever we looked at what we were trying to solve here, we were just looking for those timed vampires that were just pulling people's, just extracting the, the soul out of people all day. You mentioned pulling from spreadsheets and uploading sheets from, you know, point A over into point B. Those are all great things that computers can automate and they are extremely time consuming for people to do it. You've got, you introduce room for error, you introduce problems all along the way. It's kind of a pointless and unfulfilling day for your ad operations people when they have more strategic work that they'd like to do. So what we did is we looked at the ad ops workflow that we knew about and since then, about 150 large agencies that we've gotten to work with to see those commonalities, those patterns that people do that really are not the high value parts of the job. So we took all of those pieces and we said, let's build an infrastructure, let's build an operating system that allows those things to happen automatically. Get them out of our way, get us back into the point as advertisers that brings us the joy. We as advertisers got into this business because it's a creative space, it's fun. You, you have a creative expression digital kind of, kind of squeezed a little bit that out of us and became a lot of button clicks. So if we can get those out of the way and we can get back to the creative side, the enjoyable side, the strategy side for, for our clients, then we know that we're in the real deep, meaningful work.
- [08:24]
A
I talk to a lot of people that work in ad ops just because I've always found them to be the foundation of any and I've said this for 100 and something episodes. I've always found out operations to be like the foundation of the business, to keep the car rolling and it's important right without the wheels falling off. What do you think are some of the biggest challenges when you talk to people about automation? Especially when you're talking to an out operations person that may feel a little hesitant to automate aspects of their job because they feel that there's that feeling of oh, this is important to me that I keep doing this because that's job security. Yet there's other ways that they can utilize their time to improve the business than, you know, CSV files and pivoting for sure.
- [09:04]
C
I mean, I think the most important thing is automate what you want to get rid of. There's a lot of tedium in our days. The copy and paste function is rarely someone's value, your value system. They really say I'm the best at copy and paste. So if we can kind of make that part, the pieces that go away. So there's a, there's a great joke about AI and I think it relates here to automation or AI and we'll get into AI a little bit as we go is that AI can write poetry and make music and do art, but it doesn't do the dishes. And I think what we're really trying to focus on here is doing the dishes. That's not where the value is. The value is unlocking that human creativity, that inspiration and that strategy that we want to bring to the table. And I'd like to give that time back to people as much as possible. So the general rule when people are anxious about that is you can automate the parts that you want to automate and you can retain the parts that you don't want to. This platform is built in a way, you know, the solution here is built in a way that you can express it your, your way. Automation oftentimes rule systems, whenever I see this signal, I would want to do this action. Great rule systems for us to think through. And the important part of that is it's the same rule system that we would do as, as a person. So if I were to ask, we'll use this, a hypothetical name, Seth. It's a real name. It's actually one of my, you know, the person I learned most of the ad ops function from. If Seth were to say, every time I see this pricing change, I would do this in my ad copy. Let's make a rule about that. Let's make it so that I am doing Seth's work, Seth's way and retaining Seth's value prop. But I'm not making Seth click all those buttons. And that's to me, that's one of the hyper values of this rule system that allows you to define what you want to automate, retain what you want to retain and really get to that spot where you find enjoyment in your day. I think when we get down to it, there's kind of a role and a swim lane that each of these technologies play. Automation will play a role and automation will be this block and tackle rule based system that's very predictable and will do its work. AI will tend to be more of that probabilistic just in time runtime advisory, possibly doing agentic work if we allow it. And then finally there's that human touch which is that last mile refinement, that thing that allows us to bridge that gap between our brands and their shoppers, build that connection and build it kind of in a human way. So I think we all have our swim lanes and I believe the human swim lane is the most compelling one. Love that slide for them.
- [11:51]
A
And I actually want to touch upon that human aspect of it. But where do you think that marketers are getting things wrong when it comes to automation? Whether it's their perspective of automation, their understanding of autom, where do you feel they have a gap in understanding the benefits of doing so?
- [12:09]
C
It's funny, I see two extremes and they are polar opposites. There's one side where when someone hears the word automation, they hear adopting someone else's opinion and strategy and that's really not what it's supposed to be. Automation is based on rule systems. If you have a good framework that allows you to, to define and describe those rules, you can execute anyone's strategy their way, but give them the efficiency, scale and speed that they're looking for. So the last thing I wanted to do, and this is, you know, this is kind of going back to the experience at the previous company, was we, we had a philosophy about how we wanted to merchandise automotive inventory. It was our secret sauce. It was the thing that we believed in. And when we looked at automation solutions back then, it was basically going to be someone else's solution. And you might see, you might see some of that today when you, when you adopt kind of these hands off campaigns that you are releasing control to the, to the strategy. We've always believed that automation can work within anybody's strategy. Express your, your intention, express your rule systems, express, do that through the automation platform and then just get the efficiency, scale and speed out of using software to do that for you. So the last thing I want to do is basically say this is fluency's way to merchandise auto inventory. I mean we might, we might have a perspective, but we want this platform and this, this solution to work for anybody's secret sauce. So conflating automation with someone else's strategy is, is the first thing that I think I see. So that's like more restrictive. And then on the polar opposite side, I've seen a lot of confusion that says automation means AI and they're not one and the same either. They complement each other well, but they're not one and the same. You know, automation is very deterministic, rule based systems with whatever you put in a known outcome will come out. AI adds a little bit of a risk variable that you have to kind of work extra hard to manage. By its nature it's probabilistic, which means like a person, there's a bit of subjectivity in it. Automation can take advantage of AI and vice versa. You know, you can use AI to help you configure your Automation or you can use automation up to a point and let AI help you with a decision tree. But automation on its own is very deterministic with known predictable outcomes. I remember when this question came up, I had just done a full talk about automation and that the immediate question was, well, what if it makes the wrong choice? And the reality is this is it's not making choices, it's, it is executing your choices. And that's, that's very deterministic in how automation works. So I think if I could reassure somebody, if you love your decision paths, if you could draw a flowchart and say when I see this, I would do this. If you can draw that flowchart, you can automate that flowchart. And I think that that's probably the most visual way that I could express what automation really does. It'll walk that flowchart for you.
- [15:14]
A
Yeah, dig in. I would love for you to dig into the AI aspect of it because I think it is interchanged quite a bit. You hear it in market like we're AI or is it automation, is it machine learning? And many times people are using the term interchangeably between various solutions in market. I'd love to hear your view on the AI aspect too.
- [15:33]
C
Well, I have spent the last three years like really immersing myself in AI and I do love its potential, but I'm also, also like I'm pretty risk adverse for our clients. I mean there's a, there's credit potential and you know, like we just said, AI has some unpredictability in what it does, which is both a benefit and a risk to manage. Some of our clients are in highly regulated sectors like ones governed by say the Fair Housing act, for example. Right. And when you're, when you have those highly regulated industries, you can accept a lot of risk or you're going to get fined and that will most likely transfer to me and I don't really want to get fined either. So this is where, you know, I think people tend to be really important and they, they have that, that governance piece. So I think what we, we find is 90, there's a 95% value and 5% risk. And we'll just play with those numbers. Say two years ago it was 8020 and now it's gotten to say 9010. And I can see the path with the most recent models to say 955. We could be in industries where that 5% could strike out the value of the rest of the 95%. And I don't really want us to be there. So I think bringing a good orchestration infrastructure into AI, good governance, good controls and systems, good context management and I don't know if that's foreign language to anybody on the, you know, on the cast, but context is kind of king, where we've always said content is king. Context is kind of king in the AI world. Giving the LLMs, which read from beginning of conversation to the end every time you ask a question, giving it good information so it can make the best decisions. That's super important to get, to get even closer to that 95.5percent that, that we see. So then you have observation systems that sit on top of that to help you mitigate that last 5%. There's a lot of things that we have to do, but it, it isn't just free out of the box because it can feel like that. If I went into ChatGPT today and asked for advice, it will give it to me. But if I ask for advice a hundred thousand times because I, I'm, I make a hundred thousand decisions, there's going to be 5% of those decisions that could be damaging. And we want to make sure we have good mitigation strategies in place for that.
- [17:55]
A
Yeah, I've experienced that. I've experienced the hallucinations, I've experienced what is stated as a fact, but it is truly not a fact where it just makes something up. And then you ask, what's the source for this? And he goes, oh, there's no source for this. Is this reliable? And they're like, no, we just made this up. Basically you're like, what is going on? I cannot use this without spot checking it. So having those, the gatekeeper and like watching what it's pulling from is really important today because sometimes it just makes stuff up.
- [18:23]
C
It's really nice and apologetic when it does too though, isn't it? I mean it'll, yeah, it admits it
- [18:27]
A
after you ask it. That's the problem. It's like, tell me you're making it up.
- [18:30]
C
Oh, you're right, that was completely wrong. And nice catch. You're like, oh, yeah, but I, I didn't want to have to catch that. So, but, but the reality is that that's still there and that's not that different than people. And I, I do want to like acknowledge that there are, if we had hired a, a brand new employee, there's a chance they could make a mistake. Also, in, in business, you end up putting checks on things to make sure that you can limit the scope of a of a bad decision. And I think that's the same with a, a software platform that works with AI is think of it like, think of, of it like an employee, one with at some level a spectrum of, of tenure and skill level should be smart. It looks brilliant, but at the same time it can make a mistake. So put this, the infrastructure in place that will allow us to, to make sure that the, the impact of a bad decision, of a wrong decision is, you know, it's controlled and it's, it's well understood.
- [19:25]
A
So I guess I have two, two more questions for you because I'm kind of, I want to geek out, but I also don't want this to be 45 minutes to 3 hours long. When have you seen humans kind of out do or outperform AI? And is there like a common theme that happens there or is it general across the board? You could, you could find one or the other kind of outperforming the other.
- [19:48]
C
So this is an interesting topic to get into because I think there are some pieces, you know, we, as advertisers, our main job is to form connections between the brands and the companies that we represent and the consumers and the shoppers that they want to talk to. So we're trying to build that relationship. There's a feeling of human to human relationship that I think people are exceptionally good at. Now let's, we'll just go back, we'll rewind time for just a second. I think humans are really good at taking accountability and probably a little bit better at adhering to compliance rules and governance. Now automation is excellent at that. So when I say that there's a space for scalability and compliance as well, but it's more on the automation side and I'd be a little less comfortable with the AI side. I think AI is really good at 247 decisioning at a pretty high level. Maybe not quite as good as a human would be, especially a talented, seasoned employee that knows the space. We did a lot with brand safety and brand compliance where we were at our last company and there were people that knew. Let's just go with Subaru's brand rules and regulations cold. There was no ambiguity. You could spot it from a mile away and people are really good at that. With AI, you're gonna have to provide it really good context. It's gonna have to almost learn that in the flow every time and it's gonna interpret slightly different and it's gonna make, it's gonna make some mistakes. I don't worry. The hallucination Concern is definitely real but it's not, it's not the driving force right now. It's the quality of the context that I, that I think is the most suspect in AI and we'll grow and get better with that over time. What is the right way to inform AI to. To prompting. Make sure the context window because it does degrade as you give it more and more context. We can retain an awful lot as humans in our heads but AI has about a, depending on the model, a 200k token context window up to a million. But usually if you get up around that 200k you see some degradation performance. I use Claude code an awful lot and you can get there pretty quick. And we do too. Yeah, it takes some time to recognize when it's exhausted it's memory. So I would say when it comes to adhering to compliance and controls I would still want a human at the helm and then those little human touches that we put on piece. So think about that last mile. Let AI and automation do this block and tackle work. Let it get us configured and structured in the way that would be a relatively well performing campaign. And then we take that human subjectivity and that, that flair and that connection building that we are really good at and do that part of it. It's that strategy, it's that, that human connection point that I think is really valuable that people do exceptionally well.
- [22:43]
A
How long does it take to set something up like this? So let's say today I'm like Eric, sign me up, I'm good, I'm ready to move forward. Is this like a multi month project? Is this like hey, we can start piece by piece and start with one or two various rules or automations and then expand as we learn I guess how do they teach you or teach the system? What's that process there to get set up? You. You think I've used fluency.
- [23:08]
C
You are hitting at the, the core challenge for that. Evolving as we speak. I mean I love the ask this question when you have a rule system like this it like if I, if I were to make a parallel. Take a, take a CRM system like HubSpot or Salesforce, those are pretty long implementations. Oftentimes especially if you're going to get workflow automation set up in there, which is what a lot of the value will tend to be. So you're going to record information but you're also going to set these triggers and workflows up and they take a sophisticated user a reasonable amount of time to get set up. The way that you'd like them. And that's been the history of the company up to a certain point right now with AI assisted building. And that's kind of what we're focused on right now, is we can get the predictability of those rule systems, but use AI to help us set them all up. And that's the promise there is. To really shorten that build time and reduce the technical acumen required to get there, you become more in the reviewer phase. We have this part of the product called blueprints, and that's one of the heavy lifting tools that works through pulling in data from different sources, merging it all together, understanding the composition of an advertiser, knowing what the strategy is, and then turning that into campaigning. That's the blueprint infrastructure that's very deterministic. It works with rule systems, but it takes a very strategic lens and some time to configure. With AI, we can do it from a requirement stock. This is the program rules, this is the brand compliance, this is the goal and the intention for an omnichannel campaign. And it will actually build the search campaigns, then performance max campaigns, the DSP campaigns and the TikTok campaigns. It'll build all those things for you, automatically defining them as that blueprint side, getting you back to the deterministic outcomes that you can just then review. It's a little easier to review something than to. Than to map all that data together. You can say, oh yeah, I can see that it's pulling in my housing unit volume here. I can see that it's pulling in the city from the same data source that I want it to. It's a little easier to review it than to set it up. So that's got a lot of promise of reducing that timetable. Right now we've got an amazing team that does our implementations and they have already, even with the existing process, reduce that timetable from about three months down to. I hope I don't get this wrong, apologies team if I do. But I think it's about a month and a half to two months right now. But with AI assisted generation, we are seeing a lot of the things that take us those multiple days to do being done in a matter of minutes and then you just kind of. You have to go through that review side. So this answer is emerging and evolving as we go. We all have the expectation with AI that it will be an easy button, but we want to make sure that it's an easy button that really reflects what our clients want. Doesn't feel cookie cutter. It doesn't feel like everybody's got the exact same solution and that it expresses their value systems and their value props the way that they want to express those value props. I think we're at 140 large agencies and I have yet to see two that want to operate the same way and split platforms customizable and adaptable enough so far to allow us to adapt that. And that's those adaptations tend to be where the time sync is, but AI has been giving us a lot of promise there.
- [26:39]
A
Eric and I'll pivot a little bit as we kind of close out the episode. Where do you think things are heading? I hear a lot about agentic AI that seems to be taking over every possible media outlet left and right. I actually just posted the other day, I said, you know, one day the heavens will open and we will find out that your CTO just has a subscription to Claude. Some people loved it, some people hated it. And I said, too bad, you have no idea who I am, so deal with it. But I wanted to ask you where. Where do you see automation going, AI going? Where do you see the industry moving that you're like generally just very excited about that you think is going to be good for us overall?
- [27:21]
C
There's a couple things in there that I'd love to pick out, and I'm going to get to that answer because there's one specific area that I am personally excited about. I think there's a couple things that you can forecast. Like the increased adoption of AI is an obvious one. That agentic statement that you just made, that agentic AI, those two words paired together an awful lot. But Agentix software has actually been around for quite a while. And I would make the argument that agency in software means making the ability unattended or unattended, the ability to make a decision. And the automation workflows that we've had for years are a version of agentic software. Now, they aren't agentic AI, but they are agentic software. And I think that you'll find this hybrid stable of agents that will be both deterministic. The automation side is the deterministic. The AI side is the probabilistic version of agentic software. And they'll work together, they'll coexist, they'll intermesh. And I think that's gonna be a really exciting space for us. What I personally am really, really excited about being in the space for about 20 years. I will say even 20 years ago, the conversation was around incremental engagement and personalization and ads and I've never really seen anybody do it practically. I know that we have the tools to kind of do it with audience signals. People have done this action, I can then show them, I can put them in a different audience and I can exclude them from other things. But that, that whole inclusion exclusion has been really challenging for us to do well. And there's a creep factor that goes along with it. So although it's been technically available for us, I would say one, the consumer has not been ready for it and the practical tech is just, it's just been out of reach. But if our job is to make those connection points between brands and consumers, we can see the effect of it on websites today. Websites have personalization and people are, are embracing that. And then the other side of it is everyone that's engaged with AI. If you go and you ask Chad GPT right now to tell you about yourself, there's been this new expectation that it's, it's paying attention to who I am, it's learned from who I am. And, and I think that if like this is still speculative, but as we embrace it with AI and as we embrace it with web experience, that personalization is going to be more and more of the expectation and more welcomed by the consumer. And if we do it well and don't violate anybody's trust in advertising, we can use AI and automation to help configure those complexities. The journey that someone walks through omnichannel with, through different points in their consumer experience to do some amazing personalization and incremental engagement, walk them through that process and build that bond much better than we have in the past. I'm really excited about where that could go. Like I said, it's been a 20 year promise that I haven't seen really done practically. And I just think it's, I think it's right there for us to take now.
- [30:11]
A
Eric, thank you so much for joining me on the pod. Honestly, I feel like we could probably geek out for like four hours. So we got to find a way to make this happen. Not on a recording where we could sit down and actually chat about it.
- [30:22]
C
Atg, this was an absolute pleasure. Thank you so much for today.
- [30:25]
A
Yeah, thank you, Eric. I really appreciate you and thank you again for being here.
- [30:28]
C
All right, talk soon.
- [30:29]
A
Talk soon. Bye bye.
- [30:32]
B
Thanks for tuning in to another episode of the AdTech Godpod, a podcast for the people about the people. Stay connected with me for more insights, trends and interviews in the realm of ad tech. Don't miss out on the latest updates so follow me on X Instagram and connect with me on LinkedIn. Don't forget ATG Slack community has insights, networking opportunities and jobs. Keep the car, keep conversation going and stay at the forefront of adtech innovation.