Paul Kelly is the Global Chief Revenue Officer at A Million Ads — this interview has been lightly edited for style and readability
PK: We power dynamic creative for for digital audio. So in display and social, it’s probably more commonly known as DCO - dynamic creative optimization. In other spheres it’s called personalization; but it is the real time use of data, largely contextual data, to optimize or version the creative for each and every impression.
JC: So give me an example of how that might work.
PK: Sure. So - I just placed an order for dinner from one of the food delivery services. There are many available. However, as you and a lot of people I’m sure recognize, not all restaurants are available to all listeners. So if you are Uber Eats or DoorDash or Deliveroo, and you’re running a national digital audio campaign, and you’re trying to prompt a quick and convenient Tuesday night meal, our platform is able to, based on the time of day and more importantly based on where the listener is, we’re able to highlight the local restaurants that are available to that listener - which, as you might suspect, increases the likelihood of an order being placed.
JC: That’s normal geo-targeting that virtually everybody offers. But you offer a little bit more than that, don’t you? Because you’re combining lots of different data points to produce creative, which is a bit more dynamically built.
PK: If it’s a national campaign, bear in mind. So there are more than 40,000 ZIP codes in the US. So, you can target those 40,000 ZIP codes, but without a dynamic creative studio like A Million Ads, you’d be trafficking 20,000 audio ads, which obviously isn’t viable.
Targeting is really about the right person at the right time: but that’s only two of the three. Right person - right time - right message.
So if there’s an opportunity to version the message, then targeting is insufficient. If you are Geico with a simple message then maybe it doesn’t need to be versioned, but in most cases that isn’t the reality, particularly with increasing audience fragmentation.
We are able to decision, as we call it, from any source of data that is present within the media plan. So that includes audience segments - for a retailer, you might be targeting homeowners, new parents and fashion buyers and a large retailer might run an ad addressing those three segments with housewares, furniture, baby care. And of course, fashion jewelry, for example. AMA is able to talk about the wonderful mid-century furniture just to the homeowners. Talk about the baby care items just to the parents, and so on and so forth. Right person - right time - right message.
JC: So, you’re creating all kinds of different bits to go into the creative so that people end up hearing something which has been put together automatically for them.
PK: That’s quite perceptive James, you’re right. If you’re in display, it’s probably not that hard to create hundreds thousands of different image files. But for for audio assets it doesn’t work quite the same way. Each campaign is served by a single VAST tag. And within that VAST tag, there are components that in any combination could yield one of hundreds of thousands of different potential variations. So you’re right, the studio effectively stitches it together. We do this every single time we get an ad call and we serve an impression. It works really well, so kudos to the to the creative team we have here, and the technology itself.
JC: There’s a lot of work in producing all of those different parts. Is this only really viable for large companies?
PK: I’ll challenge the first assumption. If I run a regular 30" audio spot, that takes one hour in the recording session. In 90 minutes - just 30 minutes more - we can get everything we need to run thousands of different versions. The work involved is actually not a barrier whatsoever. To your point, the platform is is is built for scale. That’s why automation and the ease of deployment is so important.
If you are an advertiser that only needs three versions to address three different audience segments or different regions of the country, then you don’t need AMA; it’s probably fair to say that the use case for for a scaled solution such as AMA is not as readily apparent. But if you are a national advertiser, and are looking to drive outcomes, that’s different. Dynamic creative personalization has got a demonstrable impact on driving lower funnel outcomes for traffic to stores - orders on the food delivery app that I just mentioned, bookings for travel, and so on. Our clients have have come to us - particularly retail as a very large category where obviously there’s a high volume of promotional messages and products which are dispersed and change based on the type, the week, the days, and of course the location. So with that volume comes the the opportunity for an automated scale solution such as ours.
JC: I guess you can even go further if you know certain things about a particular listener, then you can potentially build that kind of thing into the ad. If you know my Starbucks order, and you know my name, and you know whether or not it’s hot or cold at the moment, then I guess you can make a very, you know, a very tailored ad for me.
PK: It’s a great point. There’s plenty being said about brands using their first party data and how valuable that is. AMA uses a completely privacy compliant toolset, so we don’t need to see anybody’s name. But it is possible for Starbucks to provide a unique identifier to an audience segment that will tell the AMA studio to play oatmilk as opposed to the bacon, egg and cheese. So one might infer that that is a group of vegan Starbucks loyalty members. And so quite understandably, they are going to respond more favorably to to creative that acknowledges that choice.
I’ve been with the company a little over two years, and with a primary focus on the US. So the company is headquartered in London, but we’ve had a focus on the US, given that market’s scale and diversity both from a geographical and cultural perspective, all of which kind of provides challenges for brands.
JC: It sounds as if you’re not necessarily American, though, Paul. Would that be right? What’s your background?
PK: I’m Irish - I was born in Dublin, but schooled in England and started my career in London: so I’m a transplant here in the US.
JC: Studio AI was launched recently. What’s that?
PK: It does two things. Going back to your point about the work involved in creating many different assets or versions or lines. When I gave you that example of the 40,000 ZIP codes. Studio AI will make it very easy for someone to version creative for every ZIP code in the U.S. and it will pre populate all of the fields, making it kind of very clean and simple for recording and uploading into our studio where we assign rules so that the correct version is configured and served every time for each listener. So Studio AI takes a significant amount of manual data entry work out of that process.
But then it also plays a role in helping to optimize certain aspects of the creative as and when clients need that. Most of what we do is “pre-optimized” - so I live in a specific location and Chick-Fil-A, which is my favorite restaurant, is not available to me here, so I don’t need to just put out a bunch of ads to figure out that it’s probably not the right thing to play to me. But things like promo codes, certain call to actions, those are things where we’re able to test and optimize mid-campaign.
JC: A Million Ads has been doing this for quite some time. What have you learnt? What works, what doesn’t?
PK: When I joined the company, we talked almost exclusively about personalization in the sense of enhancing the creative - making it “better”. It was proposition that one would think would resonate most with a creative agency. I’m thinking of a campaign we ran with Under Armour. It’s 5:00 in the morning in Chicago and it’s windy out, and we’re able to use that to inspire the listener to do the morning run. They’ll get more done and before 9:00 than most people get done in the day.
What we didn’t realize until probably a couple of years ago was what the content management infrastructure in big companies looked like, and the amount of assets they had - the burden that was placed upon internal teams trying to to sequence all of these messages across different channels. It was eye opening, and it provided a totally different lens with which to understand, interpret or take advantage of of our product. That that was a really important insight which kind of helped us develop particular products specific to categories.
We developed a dynamic audio product that effectively did the job of printed marketing circulars, such that they were no longer necessary. Messages could be run in audio with a greater level of agility from the clients. It can be versioned by different regions or market. Target, which is a huge retailer in the US, has been a longstanding client of ours. I mentioned before about the single VAST tag. I think Target have had the same VAST tag for three years now. But we’ve run hundreds of thousands of variations, which we can get out in the wild in under twelve hours. We wouldn’t necessarily have come to that product on our own without making a deliberate attempt to better understand the different challenges our clients have.
JC: I think the fact that you have mentioned the VAST tag a couple of times shows that that this works with anybody - there’s no specific additional thing that that any podcast hosting company needs do other than support VAST tags for DAI.
PK: We are agnostic when it comes to media and inventory - deliberately so. We have no horse in that race, so to speak. And it’s, of course, up to the clients and their their representatives to determine where the their creative runs and what media they buy. So, yes, you’ll find AMA wherever VAST tags are found.
The bulk of our business to this point has been within streaming audio - like live radio stations and things. But we’re starting to see a lot of interest now for running dynamic creative within podcast inventory, which is a new emerging trend this year. And obviously one that we’re excited about, as there’s a juicy middle between fantastic hosts read ads on one side which are labor and resource intensive, as we know. And then there’s a regular 30-second ad on the other extreme, which from a from a listener perspective, probably isn’t the most ideal. So there’s a large middle ground there. We think that we can be one of the people that provides a more native creative format for, podcasting that would fit the listener experience which podcasting has done an amazing job of preserving to this point.