Staying within brand guidelines can be a challenge, and as the use of AI in marketing rises, this will become increasingly difficult. Rob May, founder of BrandGuard, explains how solving user challenges transformed his platform from what was initially an advertising platform into an entirely different product that uses AI to identify branding issues.

He shares his career journey, how the rise in generative AI drew him back into the start-up space, how different AI models work, and the impact he believes AI will have over the next five years.

Listen to the podcast now via the links below:

About BrandGuard

BrandGuard is an AI-powered brand governance platform that helps ensure brand consistency in customer facing assets, such as advertisements, generated by both humans and machines.

About Rob

Rob May is the founder and CEO of BrandGuard and is a leading figure in the field of generative AI and brand safety. With his extensive background in entrepreneurship and angel investing, Rob brings a wealth of knowledge and expertise to the table.

Time Stamps

[00:43.3] – Rob discusses His career journey and why he founded BrandGuard.

[01:47.5] – Rob goes into detail about BrandGuard, its beginnings and what it does.

[12:33.0] – Rob explains some off the issues with branding in AI content.

[16:10.0] – Who can use BrandGuard? Rob discusses what businesses can benefit.

[18:45.6] – Rob shares his thoughts on how AI is going to change marketing.

[22:59.8] – Rob’s contact details.

Follow Rob:

Rob May on LinkedIn:

BrandGuard website:

BrandGuard on LinkedIn:

Follow Mike:

Mike Maynard on LinkedIn:

Napier website:

Napier LinkedIn:

If you enjoyed this episode, be sure to subscribe to our podcast for more discussions about the latest in Marketing B2B Tech and connect with us on social media to stay updated on upcoming episodes. We’d also appreciate it if you could leave us a review on your favourite podcast platform.

Want more? Check out Napier’s other podcast – The Marketing Automation Moment:

Transcript: Interview with Rob May – BrandGuard

Speakers: Mike Maynard, Rob May

Mike: Thanks for listening to Marketing B2B Tech, the podcast from Napier, where you can find out what really works in B2B marketing today. Welcome to Marketing B2B Technology, the podcast from Napier. Today, I’ve got Rob May, who’s the founder and CEO of BrandGuard joining me. Welcome to the podcast. Rob.

Rob: Thanks for having me.


It’s great to have you on. So, you know, you obviously founded BrandGuard. But let’s start off by taking a little step back and finding out how you got to the point of, you know, wanting to found the company. So, can you tell me a little bit about your career journey?

Rob: Yeah, so I am a electrical engineer and a chip designer by training. So that’s where I got my start. And I always knew I wanted to get into startups. So I, after a couple years of doing chip design, I joined a startup, then joined another startup, then started my first company back in 2009. And so that in 2014, the first one did really well, otherwise, I don’t know that I’d still be doing it. So then I started a second company, the second company did not go well. Then I went into VC for a couple of years. And I saw this generative AI wave happening and decided I had to come back out of VC and do a little more operating. So started my third company, which actually spawned the technology that became BrandGuard. It wasn’t the focus initially of the third company, but it led to the creation of Vanguard.

Mike: That’s great. And obviously, all the great people started electronics engineers, I don’t know if you know, I started my career as an electronics engineer as well during board level design.

Rob: Exactly. That’s awesome.

Mike: Can you tell me a little bit about how you decided to apply AI? And what brand God does?

Rob: Yeah, well, we started with the idea that you could use generative AI to create hyper, personalised marketing and scale. So think about the idea that, you know, you’re going to sell a pencil, you know, if you were going to sell them a nice mechanical pencil, you have a couple of personas, you’re going to write ads for those personas. But what if you could speak to everybody differently, right, a 19 year old college student who’s really into mechanical pencils in the Pacific Northwest, may want a very different image and freezing in their ads than a 75 year old, you know, writer who loves to use mechanical pencils for nostalgic reasons, who lives in, you know, southern Florida or whatever. And so, imagine if you could really use your vi as Chet GPT, here’s 1000 different personas to buy my pencil write me a different ad for everyone. So that’s kind of what we created. And it worked really well. But there were two problems. One problem was that it became obvious that the platforms were gonna do this in cells. So Google and Facebook, were going to build in this functionality. And the second problem was that we would show this to CMOS, and they would say, see, I’m not a brand person, I’m an enterprise software person.

So I didn’t realise that, like, if you’re a brand person, you obsess over minutiae about how things look and how things are phrased. And so you might say, if you sell bottled water, you might say, we’ll say purified, but never filtered. Right? Or you may say, you know, you have a certain imagery of the model that you’re using in an ad. And you may say, like, no, no, she can, she can have a wrist tattoo, but not an arm tattoo she could like, are people that were catering to look like this and not like this, and like, these minor things matter, you know, she, she would sit with her legs this way at the table, like all these, all these little things. And so what would happen is we would show our tool to these markers. And they would say, well, that’s great, you’re gonna create 5000 Hyper targeted ads for me, that’s awesome. I have to review them all you’ve created work for me. So we took a step back, and we were like, well, could we teach machines to understand brands and branding? And it turns out, you can, it’s a very hard problem, what we what we did was, we built a series of tools you can think about, it’s not like a machine learning model. It’s dozens of machine learning models. And we ingest brand guidelines, previous versions of, of content that a brand has produced. And then we built what we call a brand governance platform that takes these things, breaks them down into models, and in the models check, is the stuff you produced on brand is it meet the brand guidelines, you know, it started as a feature of this ad product, like we’re going to create ads and the ads are on brand. And it just became the whole platform. We don’t do any ad generation anymore. We just whether humans create the ads or machines create the ads, we just run them through our series of models. We provide scores and feedback and analysis and all that kind of stuff. So it’s it’s pretty cool technology.

Mike: So you’re doing that checking of 5000 different ads that the person who’s responsible for brand didn’t want to do. Yeah,

Rob: or even we, you know, even even people that just are don’t have an AI process and are just doing hundreds of ads per month. We frequently hear but so let’s say you’re using an agency and the agency designers are working on lots of different projects. They don’t have your whole Till 10, or 40, or 80 Page brand guidelines memorised in their head all the time that they’re working on, you know, they make mistakes. And we constantly hear that about a third to half the time people are looking at content saying like, no, this doesn’t meet the brand guidelines, go back and do it again. And what they want is they want a tool that takes that first pass. So now the humans would say, don’t send it to me for approval until it’s past BrandGuard.

Mike: Okay, so you’re actually doing that first pass before it appears with humans? I mean, one of the things that I think is interesting from this is, do you think this world where everybody gets a personalised email is actually going to happen? Or do you think enterprises they actually want to preserve their brand? They want some consistency on brand. And they actually don’t want these hyper personalised emails being sent out?

Rob: Well, I think they’re, I think they’re trying to do both, right. So we’ll, we’ll see if it works. But the way you can think of it as like, they might use similar phrasing, like, obviously, you know, Nikes tagline is just do it. And they’re not going to change that tagline for me and you or anybody else. But a lot of their imagery is like people running in the Pacific Northwest, where there’s Nike, or in Colorado, where a lot of athletes train, but might they benefit from showing people running in, you know, the beach, or, you know, I’m in New York and downtown Manhattan, Central Park, you can see that having an impact without changing a lot of what Nikes trying to do. So So I think there’s gonna be a lot of experimentation to get there, I think it will move the needle. But there’s going to be a counterweight, right, which is, humans get tired of these things. I mean, every time something becomes a best practice, like, oh, man, these notifications, and these NPS scores, kill me. Because the more products you use, the more people have you that just want you to take a one minute survey. And it’s like, I can’t take 91 minute surveys in a day just because I interacted with that many products. So we got to find better ways to give and get feedback and interact with customers in ways that respect what they’re doing. So it’d be interesting to see how these multiple forces evolve in this scenario.

Mike: Yeah, I mean, I think it was interesting. I’ve now got visions of everyone in the UK getting pictures of people running in the pouring rain from Nike. Yeah. I hopefully won’t get to that. Anyway, going back to BrandGuard. So something you mentioned earlier that I found quite interesting was that you fed the AI system, you trained it on the brand guidelines, but also on past content. So presumably, one of the things you found is that there are explicit brand guidelines, things that, you know, are written down are very clear. But there’s also kind of tacit brand guidelines that are kind of held within the heads of people. Is that what you’re trying to address and understand?

Rob: Yeah, so there’s a lot of lot to unpack in that question. We see people with 150 page, well defined style guides, and we see people who barely have anything written down except a handful of brand guidelines. One of the things we hear a lot of times from agencies is, can you help us help the clients better define their brand, because they know what they want, they haven’t been explicit about it. We’re building a module into BrandGuard called Brand Builder that allows you to define and capture those rules. But we already do a creative job of capturing three simple rules, like the spacing around the logo has to be displayed, a logo can’t be turned this way, right. And there’s more complicated rules, I think we’ve seen some fun ones, like no images can show a child using technology without an adult present. That’s a hard brand guideline to teach a machine. And so we have an entire synthetic data pipeline that will create pro and con images that put them into a model so that the model can learn that rule. But yeah, you know, it’s one of the interesting things about this space, compared to a lot of other use cases of AI is there a lot of ways where AI is going to get better than humans. But in this case, humans sort of define the brand. And maybe we’ll get to the point where AI can make suggestions about how you might want to move your brand, which directions which attributes or values, you might want to focus on more than others. But by and large, humans will define brands, brand values, training and datasets for the brand related models. And so I think it’s a really good place to be if you’re working in AI for that reason, because like you said, so much of it is in people’s heads and, you know, you need workflows to sort of get that out and capture as much of it as possible. We also do it through the regular feedback, right? You could we could score something high or low and you could dispute it you can say no, no, this should have been scored a different way and here’s why.

Mike: So it’s interesting to continually retraining that model. I’m interested when you try and build that that style guide in the AI if you like in the AIS head I don’t know if that’s the right way to express it. Does it help to have things that are off brand and on brand or do you just feed it the past content that’s been approved?

Rob: It helps to have things that are off brand as well. So a lot of times we’ll pull some public competitor data, you know, from Nike, we would pull Adidas, just to contrast because that’s, you know, I don’t know how deep you go in the AI space, but these things are basically mapped to a mathematical space that focuses on similarity. And so if you can say, these things shall be similar to each other in this mathematical As a nice should not you can think of us as drawing a brand boundary in that mathematical space around what’s on brand? And what’s not.

Mike: And I mean, you’ve mentioned a few things. But is there any limit to the kind of content that the system can, you know, assess for compliance with brand guidelines? Can it go through to tweets and, and things like that, as well as articles and images?

Rob: I would say it’s built first and foremost for marketing materials, primarily advertisements, but we can do a lot of stuff, you know, tweets, tweets are a little bit harder, because they’re so short. And the less information that you have, the less accurate you’re going to be about if something’s on brand or not. Twitter, social media platforms are also an area where you try to be a little maybe more kitschy than you would be in other, you know, types of marketing materials, you’re trying to be funny, you’re trying to tie to memes, we can pick up on some of that, you know, is this a meme that your brand should want to tie to or not? We do a lot with some of the Instagram influencer use cases, we’ve been asked to do some some things we’re not we’re not working on this actively. But we’ve looked at doing PowerPoint presentations. If you’re a consulting firm or real estate firm, you’re doing a lot of presentations to people about things, right, you want to make sure those are all on brand for your firm. And you know how people get in and walk around with PowerPoint and change everything. So even if you have templates, it’s not right, we’ve been asked to do product packaging and, and other use cases like that. There’s a big use case around licencing as well. So if your sports team and you’re, I’m licencing you my logo so that you can use it, I probably have to approve the product shot and the marketing materials around it. And that’s very time consuming if you’re doing a lot of licencing. So we’ve we have some customers that have that use case as well.

Mike: Presumably, what you’re doing is you’re coming back with a score rather than necessarily, yes, it’s on brand. No, it’s off. I mean, there’s always Shades of Grey. I mean, how do you do that? Do you literally provide a score? We do

Rob: we provide a score and some feedback on specific models. So you can decide what to do with that we give you an overall score. But sometimes it could be like, everything’s great. But you know, maybe you have a rule that the logo always has to be in the upper left hand side of the page, and it’s in the bottom right. And so maybe it scores at 9%, everything’s good, but the logo totally fails, we highlight that information for you. And then you can drill down and see where the asset fail.

Mike: That’s interesting. I mean, I’m intrigued, you know, people are starting now to use generative AI to create some content marketing content. Do you see humans as being better at staying within the brand guidelines? Or would AI actually be more likely to stay within those guardrails?

Rob: Probably humans. And the reason is that the way that most of these generative models work, and this may change, right, people may come up with a better way that these generative models work. But today, the way they work is you take this world of information and you compress it down into a space. So you can think about a you can think about a song that’s compressed, and it’s lost some of its fidelity. So think about these ideas, or these images with these words that have done that. And now when you ask it to generate something, it finds an area in that mathematical space that we talked about, and it expands it by introducing some randomness. And so by that randomness, you can never tell what’s going to come out. It’s a big problem. One of the one of the early examples that we used to do is we would prompt chat GPT with the Tesla style guide and test the rule number one is do not use the word luxury Tesla’s not a luxury brand. It is a high performance brand. And then you would we would ask chat, GBT, right, some ads for me to sell Tesla’s to rich people and the first one every time they would come out and be like, blah, blah, blah, don’t you love luxury, even when you prompted it with the brand guidelines, because luxury and rich are so tied together statistically, in these models, which is how these models work, it’s hard for them to break, you can’t make it part of the model generation itself. So you need filters over top of it. And I just we don’t think it makes sense for every generative AI company to do their own filtering regarding your brand. Because now as a brand manager, if you have 30 tools in your stack, and you have to go through and be like, Okay, well, you know, I’m using open AI and Jasper and WordPress and HubSpot, and figma and Canva. And I have to manage my brand governance piece at all of them. And they’ll have slightly different models. So it’s not consistent like this, it’s not going to work. It’s why we’ve really tried to integrate it with everything because you need one tool that’s like this represents my brand to an AI. So we’re very heavy on the integration side. We work with figma and Canva, and a whole bunch of other tools today.

Mike: I’m gonna guess we started that that answer talking about, you know, some of the issues around generative AI and it getting a little bit of peace, you know, partly because of the randomness. I mean, how consistently good can I be at enforcing brand guidelines? You know, we hear a lot about hallucinations in generative AI. Do you have the same problem in brand God? We

Rob: don’t because we are not generative models we are what’s called discriminative models. So we are choosing between things we are not creating things and the hallucinations come from the randomness sits inserted in the creation process. So that’s why we sit on top of all these generative models, we can get really, really good. But we can only get as good as the data that we’re given to discriminate. And as you know, like brands, an area where sometimes even people on a company, senior people may argue over some aspects of the brand. And if something’s on brand or not, there’s somebody you know, we see people that have companies that have like usual, big lovable nerd is a brand voice concept. It’s like, well, like what does that mean, that’s open to interpretation. So that there will always be a little bit of that, we try to focus on providing easy, quick, automated rejection, for stuff that doesn’t meet the brand guidelines, and human in the loop approval for stuff that does or may be on the margins.

Mike: That sounds good. And I thought it was a great explanation of the difference between what you’re doing and guarantee of AI, I think it’s all too common for people to you know, see my eyes just one thing when it’s lots of different things. One of the things I’m intrigued in this, you’re actually effectively building custom models for each and every customer, which is obviously time consuming. Does that make brand garden expensive products? Is this like only for the largest enterprises? Or is it something can be used by a broader range of customers?

Rob: Well, that most of the process for training models on a per customer basis is automated. So we’ve gotten pretty efficient at that. So even though we do build different models per customer, they’re based on similar workflows, you input your data, and we can we can sort of get there. So that doesn’t really drive the cost as much as how much inference you want to do, which is how many things do you want to test to see if they’re on brand. So it’s, it’s a product that starts at about $20,000 a year for small to mid sized customer, and goes on up to you know, mid six figures, maybe for really big brands that to a lot of stuff have multiple brand hierarchies. I think over the years, this will become best practice for everybody. But right now we primarily see most of our customers are, I’d say, like fortune 5000 brands right there, the brand matters a lot to them, we’ve had CMOS tell us, they can estimate how much revenue they lose if an ad goes out with the wrong font. So, you know, really big companies with a lot of data on the impact that brand guidelines have on their brand and on and on customer perception. So that is the majority of our market now. But I do think it’s coming down market over time. And

Mike: I mean, one of the questions, I think people, you know, interested in the product might wonder is how would they go about evaluating the product. I mean, obviously, you can’t just run a, you know, one week test, you’ve got to build the models, is there a way for someone to experiment without having to commit to a year subscription? We do, we have test

Rob: accounts you can play with. So we use flex brothers a lot, we have a Brooks Brothers demo account where you can read the brand guidelines, you can upload stuff, and that’ll give you a general feel for how the tool works. And then what we normally do is we normally move to some sort of paid pilot that might be like 10, grand, maybe a little bit more, depending on how big you are, where we take in some of your data and train up some models, that process normally takes about 48 hours to get that going. And then people can try it out for a couple of months and play around with it. You know, the bigger challenge tends to be internally, how do you build it into your workflows? Your workflows have probably been mostly human based approvals? How do you migrate those over to a tool like this?

Mike: That’s amazing, because I mean, 48 hours to get up and running seems very quick. So you know, sounds like it’s actually not a difficult tool to evaluate and test and play with them. Certainly, the Brooks demo account sounds fun. Yeah,

Rob: it is, it is pretty easy to get going. And I just mentioned, there’s multiple ways to use it. There’s a web app, there’s a Chrome plugin, there’s an API. So we have people that use all those

Mike: awesome. I’m, you know, you’re obviously a big believer in AI. And, you know, you found an area of marketing that really benefits from Ai. I mean, how do you think AI in the next five years is going to change marketing,

Rob: I think you’re gonna see every marketing stack become more automated and more AI powered. And I think what that’s going to do so if you look at a lot of the research around AI, it doesn’t improve the top, it brings up the bottom. So here’s a very interesting example, think about chat, GPT. Chat GPT does probably not make your world class writers much better. Maybe it’ll inspire them here and there with some ideas, but it makes your poor writers average, much, much better, right? So so take the bottom half of writers, it makes them average, take the you know, next quartile makes them a little bit better take your best writers, it doesn’t do that much for them. So now, what does that mean? If you translate AI into your automated marketing stack, it means that if you look at your marketing, operational excellence, and your creativity and all kinds of stuff, all the people at the bottom are going to now be up here. And so your, your difference between the best and the worst is going to shrink, mainly because the bottom comes up. Not that not because the top comes down. It’s going to be easier to be a competent sort of performance marketer or, you know, brand marketer just from these tools. It’ll always be hard to be great because you have to have something special. You have to have a process or an insight or things that other people don’t have But I think a lot of what’s gonna go away is a lot of your operations are going to be automated, it’s gonna do a couple things. Number one, it’s gonna make brand strength more important. So focusing on building the brand, really honing those attributes and values and how they connect to the customer. And what they mean in the mind of the customer is gonna be really important, even for smaller companies that maybe thought less about their brand before. And then I think the second thing is, marketers are going to become more and more strategic planners, trainers for the AI models, strategists continually being creative and coming up with new ideas to test and innovate and, and stuff like that, and less of the right me 10 more Google ads for this persona.

Mike: That’s fascinating. I mean, one of the things we’d like to ask all our guests is, what advice would they give if someone said, Should I go into marketing someone just leaving college or entering college? I’m intrigued, it sounds like it’s potentially gonna be tough, you know, particularly if you’re in that, that bottom half of marketing ability to, to really stand out, would you would you say that marketing’s a career that’s got a lot of opportunity going forward? Or do you think AI is going to make it more and more difficult to stand out?

Rob: I think it’s already playing the stack. I think brand marketing and content marketing, PR com stuff like that is gonna matter a lot more. I think your fast turn stuff like social and performance marketing is going to be more and more automated away. So which is interesting, because if you’d asked me, you know, seven years ago, I’d say well, oh, God, you want to be the person who masters Google, and Facebook advertising, right? Like that’s drives so many people’s leads. Now, I think you’d be the opposite. I think you want to work on being the most creative, the most experimental, the best at using these AI tools to test and experiment and prove or disprove hypotheses about your customers.

Mike: That’s, that’s such a fascinating way to look at it. And I think a very positive view of some of the opportunities. Another thing we’d like to ask everyone is about marketing advice. And what’s the best bit of marketing advice that you’ve ever been given? Oh,

Rob: that’s a good question. I mean, this is a hard thing to pull off. And not every brand can pull it off. But there was a book that was written probably 20 years ago now called Purple Cow by Seth Godin. And he made this great point that like, you know, if you’re a farm that for whatever reason, produces this freakish Purple Cow, you don’t have to market it, everybody talks about it, because it stands out. Now. There are certain categories that people just don’t care about as much. And it’s hard to stand out and be remarkable. But if you can make a product that that’s, it’s that remarkable, it really markets itself, and that, that that matters quite a bit. And so I think when you can find those opportunities, you should really, really lean into them, because they’re very special. Great advice,

Mike: I love it. Rob, you’ve been really generous with your time, you’ve given us not only a great explanation about managing brands, and how brands can help, but also think given us a really good overview of, you know, some elements of AI. If people are interested to learn more, either about BrandGuard or contact yourself, what’s the best way to do that, feel free

Rob: to visit our website And then you can email me, I’m just rob at brain I can’t get to everybody sometimes. But I you know, I do try to set aside a couple hours a month to talk to people that are interested in AI making career transitions, you know, we try to one of our core values as a company is to be helpful. And that includes people in the AI ecosystem and marketing systems. So we, you know, like I say, can’t get to everybody, but I did try to set aside some time to answer questions and help us stuff like that for the, you know, even strangers that email me.

Mike: But that’s amazing and very generous. Rob, it’s been great. And you know, anyone who’s struggling, managing content and making sure it meets brand guidelines. I think, you know, going visiting brand garden AI would be a great next step to take. Thank you very much for being on the podcast, Rob.

Rob: Yeah, thanks for having me. This was fun.

Mike: Thanks so much for listening to marketing B2B Tech. We hope you enjoyed the episode. And if you did, please make sure you subscribe on iTunes, or on your favourite podcast application. If you’d like to know more, please visit our website at Napier B2B dot com or contact me directly on LinkedIn.