Luca Gonnelli On How His Italian Agency Spin-Out Algo Uses AI To Automate Videos
May 10, 2023 by Dave Haynes
Tools that fast-track or reduce the costs of video production have been around for many years – often based on templates that can be brought into design software or extended to cloud platforms. These kinds of tools make it possible to produce a professional-looking video spot quickly, or with a newer breed of them, produce dozens, hundreds or thousands of videos really quickly.
An Italian motion design studio saw both the demands and possibilities for video automation, and launched a sister company in Turin called Algo. It has some similarities to what’s out there, but takes what you might call a hybrid approach. The design process is very much like a traditional agency, with briefs and storyboards. But once that phase is completed, Algo’s customers use the platform as a service.
If you have an electric vehicle and have used a Volta charging station, you may have seen motion infographics on the screen that used real-time data from Bloomberg to visually show local air quality conditions on the charging totem screen. Johns Hopkins University used Algo to develop a daily COVID tracker during the pandemic.
Algo’s main market is the business side of social media – so more Linkedin than TikTok. But it has already done and expects to see more work coming for digital signage and Digital Out Of Home screens. Automated spots can run on screens in much the same way as digital signage platforms tap into subscription news, weather and entertainment feeds.
I chatted with Luca Gonnelli, one of Algo’s founders.
Subscribe from wherever you pick up new podcasts.
TRANSCRIPT
Luca, thank you for joining me. We’ve not met in person and I’ve never been to Torino, but maybe one time, I’ll get there. Can you tell me what Algo does?
Luca Gonnelli: Sure. Hi, Dave, it’s my pleasure to be here. Thanks for having me. So Algo is a design studio, basically, it’s a creative studio specialized in data visualization and in particular in what we call video automation, which is basically software or a dashboard to create videos.
Okay, and what does that mean in real terms? So if I’m a digital signage network operator, obviously we’re talking about digital signage part of this, and you’re doing all kinds of work for different end users, but what’s that gonna mean for that person, for that end user?
Luca Gonnelli: Sure. We are working on different types of campaigns and it’s super interesting to talk to you about digital advertising, out-of-home, et cetera. That’s normal, not often, but sometimes that’s part of the equation, while, of course, the other part of the equation is social media and digital overall. So yeah, we really like to work on campaigns where we help our clients showcase data in a very meaningful way, and possibly very easy to understand for the end user, and also it’s a kind of way for companies to avoid sensationalism and try to communicate to their users in a respectful and positive matter about data. But of course, it’s a way for companies to profit from their data, and use data in their day-to-day communication, which is not often very common.
In terms of the types of campaigns we work on, it’s gone from completely autonomous ones like the campaign we did for John Hopkins University tracking the COVID pandemic where we were creating a video every day that was automatically tracking COVID based on the data, and this was only for online, but in some other projects Where we lean more towards the manual input of data. So sometimes we work with teams with our clients to empower them to create videos effortlessly without having to have video knowledge within the whole organization.
And so yeah, in some cases, for example, in a project for Volta and Bloomberg Green, we did just that, and the output was of course on digital advertising.
Would it be fair to say this isn’t necessarily what a listener might think of as a conventional video? It’s not people or landscapes or that sort of thing moving around. It’s more dynamic/motion infographics. Is that a more accurate way of saying it?
Luca Gonnelli: Yeah, that’s correct. Before being Algo, we also ran a motion design studio called illo. So Algo is a kind of a technological sister studio, and of course, our background is very much in motion design, but at the same time, Algo really can work with any type of medium, even footage, for example. So of course we tend to go towards a look, which is more graphic-oriented and more based on graphic design but at the same time, in some projects, we also have footage and photos and yeah, that’s of course less live-action than maybe average. But, of course, that can be part of the equation.
So when you’re using video assets they’re like an element within a presentation as opposed to you’re generating original video out of it. This isn’t an early-stage AI, MidJourney thing, or whatever, right?
Luca Gonnelli: No, but at the same time, we use AI a lot for many different things. So we really tend to define ourselves as Video + AI, because we are not a kind of stable diffusion type of algorithm. But we use many different tools for doing different things from the simplest ones, like transcribing an audio to even generative things like generating a picture or generating audio. This is, for example, a really good use case that we are doing right now quite a lot.
You mentioned that you have or this is a sister company to a more conventional motion design studio. Did you create Algo because this was work that kept coming up, and you thought, okay, we need to set up its own initiative to do this?
Luca Gonnelli: Yeah, absolutely. A few years back, we were seeing that communication, the social media world especially were asking for more and more video every day, and our clients could not anymore rely on one piece of advertising every six months, but they needed to be always on and always communicating so that’s definitely something that’s starting from our technological background, both me and my co-founder and some of them earlier team members have a technological background at the same time, it’s coming of course, from the needs of the market.
We were seeing that this was something that was coming in more and more frequently, and of course, also with out-of-home advertising, it’s great because you can have different content for different cities for different times and update everything across time and locations, which is great.
There have been video automation platforms around for a good 10 years, arguably longer than that, depending on how you look at it. But a lot of the early ones were template based and you would put inputs in and hit a button and it would render something and give you something back in five minutes or half an hour, or whatever it may be. What’s distinct about this?
Luca Gonnelli: We are very different in a way given that we decided to approach this from a very studio point of view, rather than being a product or a platform, which we are not. When a client works with us at the end, they have access to a dashboard. So there’s a kind of a product part of it, but really we don’t believe in the one size fits all template solution, and so what happens is that we want to remove the humans, the animators, the interns that are doing these things from the equation. But at the same time, we think that the designers and the animators at the very beginning of the project when you’re building something tailored made to the specific use case and to the specific location or to a specific client are really important aspects.
So yeah, we just tackle this as a service business and of course, it’s a service and then turns into a product because then the client has access to a dashboard and can create videos really like it, just like they would do on a SaaS kinda platform, but yeah, it always starts with a service.
So if I’m a financial services company and I want to do something like what Bloomberg did, I would come to your company and there would be a brief and everything else, but you would basically design a template that would be the working wireframe or armature to produce videos as often as needed and quickly or even automated. Is that accurate?
Luca Gonnelli: That’s super accurate, and yeah, most of the time, the starting point is really understanding what data can be used and what data the client has available and what’s their objective in their communication. So what they want to obtain from communicating, and so yeah that’s really it’s a work that we do together with the client. It’s rare that we get a brief and we start working. It’s more like, I have an interest in automating something. I have the data about this and what, what can we do together?
And so it’s really about helping them sketch out concepts and understand exactly how this could work. But yeah, then, of course, we get into the data analysis phase, the conception and storyboarding phase, then design, animation, and then all of the technical phase later to make this possible.
Do your clients have their heads around how all this works? Do they understand what’s possible, or do you get into these discussions and say actually we could do a lot more than that?
Luca Gonnelli: Since it’s not yet a super common thing to work on automated videos, we definitely help our clients understand what’s possible.
For example, could be the fact that maybe financial clients know that we can create a campaign with a weekly video that’s doing a recap of the financial markets, but then what they don’t know and what we try to tell them is that you can also trigger a video when something happens, so for example, if Bitcoin is up right now, plus 20% compared to yesterday, that’s the moment where you want to communicate. So we can automatically trigger and generate a video at that moment. That’s one example of how we try to make our clients understand the possibilities.
How much pre-planning and rules and everything else do you have to put into making that scenario happen? It’s not a smart thing where it’s just going to know “Bitcoin’s up so I better generate a video” – there are parameters, and everything is set, right?
Luca Gonnelli: Yeah. We connect to different sources of data. I think over the course of the last few years, we connected to really hundreds of different APIs and data points. But yeah, of course, what you do with the data is the interesting part, and so each time is really about deciding what these rules are and what rules are meant for the specific client. So it’s definitely a process and it’s definitely an iterative process.
So we start with an idea but maybe 20% up is not the best because maybe it won’t trigger very often. So we want to put that at plus 7%, and so yeah, that’s definitely a lot of back and forth, but it’s super interesting and it’s super meaningful when you start to see that videos coming out are really talking about the important stuff for the client are on top of the news. It’s super interesting.
Another example of this would be we are using an AI called Feedly to basically select articles that are relevant in a specific sector and create videos on top of those articles, basically transforming those articles into videos, and that’s another similar but very interesting approach where you completely give the AI the ability to create videos on different topics. The only thing you do is basically say, okay, I want to follow these new sources, I want to follow Bloomberg and the New York Times and the Financial Times, and then I will track these topics: Crypto, NFTs, and so at that point the AI will come out with videos that are trending and that has just been published and are interesting. So you completely give the AI the ability to create videos, which I think is very fun and interesting.
Is there any kind of gatekeeping in there?
What I mean by that is let’s say you’re using an AI tool and it decides it can generate a video about something and it’s not correct, which can happen, I think they call AI hallucinations or something like that, and it’s the wrong thing. So if I’m a financial services company, I obviously don’t wanna be putting out inaccurate information. Can they review everything before it goes up?
Luca Gonnelli: Exactly. So the first thing that happens in this particular kind of project is the fact that of course the video gets generated and the editorial team on the client’s side can review the video and can both edit the video if something just needs a little bit of correction or can skip the video completely. So there’s an option too, if it’s not connected straight away to posting. We have a connection to posting, but it’s normally after human review, which is always needed at this point.
So if you had really trusted lockdown data sources, like the financial numbers for a company or whatever that you know is secure. Those could be automated, but other things you’d want that just checks and balances on.
Luca Gonnelli: Exactly. When the AI comes in, it’s accurate 97% of the time, but of course, you want to make sure that 3% don’t get published so there’s always a manual check, which is needed. But actually, the interesting thing for the client is that you can have a kind of newsroom producing video content for you in really high numbers per day, and the only job remaining on your side is to just watch the videos and approve them or edit them in case you want to add something.
You’re in this interesting position where you’re a creative agency, but you’re working a lot with AI and you have all the discussion right now about what AI means for the creative process, does it remove the creator process to some degree, or is this good or bad, or you have a somewhat unique perspective?
Luca Gonnelli: Yeah, it’s been quite a lot, actually. Since we started Algo, we also have had animators and designers coming to us and saying, Algo is trying to replace my job, and we are always replying to them, the first project we did started because we wanted to work with a client in the sports sector for the Italian football league, which is a very huge topic, and basically, they wanted videos coming out every weekend for the whole season, and it was like a nightmare of a brief, and we decided to tackle it with automation. So we tried to save ourselves from doing this project manually, and so yeah, in the end, what I’m always say to people that are scared about Algo replacing them is basically the human needs to do the job of the human, which is the conception or the design, and thinking about the, how the design changes in the function of the data.
While, of course, updating the content, the template and super quickly and putting it out on social media, it’s something that our machine can do better and so we can get rid of that part of the job, which I don’t think people like, and on the AI side probably is something similar like of course, it’s crazy because you see these super high-quality images coming out and it’s getting to the video also quite quickly. I’m very positive towards technology as a person, and so I think that this will be a huge change but at the same time, it’s somehow very interesting and manageable in terms of what you can build with it. The whole change that’s happening is super fast and so it’s scary, but at the same, I feel that we are in a good position.
I believe every market around design, around creativity, is going to be much more saturated because many more people can access it, but at the same time, we’ve seen that in other markets. For example, if you think about it, creating a website that’s a super-saturated market compared to maybe video today. But of course, the most interesting and the most high-end shops producing amazing websites are still there even if all these Webflow or Squarespace or all these platforms came out to make it easy for anyone to get a website.
So I really hope that there will be, of course, a much more saturated market, but at the same time, if you are in the high-end space, that’s probably going to be more a value add than something negative.
When making notes ahead of this, I was trying to get a sense of the big attraction would be and I wrote down speed, scale, recency, and relevancy. The fact that you can have something that just happened up on a screen 15 minutes later or whatever it may be. What are the main attractions to this that you’re hearing from customers?
Luca Gonnelli: No, that’s definitely correct. The ability to scale up your production, so for example, coming to our Volta project I was talking early, the project that was being distributed to digital screens across the US with EV charging stations. The objective of the campaign was to provide a way for people that are charging their cars to not only see ads but also see this additional content, which is basically an air quality forecast of their city, so it’s connecting the objective which is living in a city with cleaner air with what you’re doing. So by being there and using the charging station, you’re participating in improving your city’s air.
It was a really interesting project. The videos were super short, and it was challenging to think about them in a way that they could work for people just passing by. From social media, for example, because of course on social media, people are scrolling all the time and it’s really difficult to get their attention the same, in a similar way, but it’s similar but different. So yeah, we try to work with that.
But definitely, in this case, we work creating content every day for the 12 different cities. So this is an example of the scale that we require maybe a few different people to work on this constantly just to produce this while Algo was working completely autonomously and yeah, the speed, that’s definitely, sometimes especially when working with sports or finance data, speed is important, and so yeah, we can get to have a video out maybe 30 seconds later than something happened, and so it’s really almost real-time in a way that that’s crazy, and so it’s also very interesting in some projects.
Is it reducing the costs of production?
I realize that you’re able to knock out a lot more stuff than you would normally, and a company like Volta or whatever, probably, even if Shell owns them, probably can’t afford just to have original videos produced for 200 locations every day or whatever it may be. But is cost a factor here?
Luca Gonnelli: It’s definitely a factor. Of course, we are positioning ourselves as a high-end solution. It works when there’s an opportunity to use a format and communicate through a reusable specific format. We work a lot to ensure the format is not perceived easily and yeah, when working on a video campaign, our objective is always to try to make it so that the end user doesn’t understand that it’s automated content. So yeah, becoming transparent. It’s always our goal. But cost optimization compared to working manually, it’s definitely an element of it.
And the more you produce, the bigger the output you have and the more that is fundamental, for example, sometimes we even work with campaigns where we produce content for a specific person. So imagine the kind of Spotify Wrapped type of campaigns, where you’re providing content specific for every single user of an application, and in that case, we’re talking about millions of assets, and so it’s definitely worth and basically the only way to produce these kinds of campaigns through Algo.
So you can do that kind of industrial-scale stuff then?
Luca Gonnelli: Yeah, absolutely. We use different technologies and one of them, which is based on a library called Lottie which Airbnb creates to incorporate animations into mobile apps and the web. We use that and with that, scale up to potentially create millions of videos per month.
Yeah, I saw on your website the reference to Lottie, and went a little cross-eye, what is that? And how about you explain it?
Luca Gonnelli: It’s super interesting. That’s an open-source library that was created by Hernan Torrisi and co-developed by Airbnb and basically, it’s a way for animators that are working inside of After Effects, which is the software that we are using daily together with others.
But yeah, it allows you to animate in After Effects, and you do that with all of the best tools that animators are used to working with, and then you output that as an SVG animation. So it’s code-based, web animation that can run in the browser or inside of a mobile app, a native iOS or Android app. So it’s a great way to come out with the tool that every motion designer loves and uses and gets to code and so that’s a super amazing way to scale things up and to reach numbers that, for us, were impossible by using only After Effects.
What are the file formats that you’re outputting?
Luca Gonnelli: All of the video file formats. So it can be mp4, it can sometimes be when working with TV can be MXF or anything, literally, so anything that can be exported from Adobe software. So static PDFs or GIFs, that’s also another format that maybe sometimes it’s not so useful maybe on digital advertising, but it can be exported. So yeah, we have many options.
So there’s nothing proprietary about it? You don’t need to write some sort of player software, or something like that to make it work?
Luca Gonnelli: No, Lottie is basically a JSON file with a JavaScript player, it’s open source, and it’s amazing. It’s nothing proprietary on that front.
So if I’m a digital out-of-home network operator or a digital signage solutions provider, software company, that sort of thing. How would I work with your company?
Luca Gonnelli: We could work on a project together either for a client or for themself, but basically, it’s about understanding what kind of data they want to talk about and what kind of solution, so it can be very free and pretty open, and then, of course, we would work on design animation and then on the output side, for example, for the Volta project, we were delivering those automatically to the screens directly. So we integrated it into the platform that they were using to deliver the videos to their screens but I remember that we also evaluated other options like going as a video directly or of course the Lottie thing can be a good solution as well, because of course, it’s outputting a very lightweight web animation.
We could of course start the project from maybe our dashboard that we built. Where the client can input the data and change and see how the design changes in the function of the data, and then yeah you just click a button for creating video, and the video gets generated in a few seconds and gets potentially delivered to the distribution servers so that, yeah, that things can proceeds mostly directly to the screen.
So it doesn’t sound at all like you get into a situation or a conversation with somebody who says, yes, we’d love to work with you, but it has to be done this specific way. It sounds like it’s pretty flexible.
Luca Gonnelli: Yeah, we tailor the solution for every project. We build something custom, and so yeah, there’s no particular way of doing things that it must be done in that way. We can really adapt, and we change technology, and we change the way we work, so that’s also part of the complexity. We are trying to make people on our clients understand that there are a lot of potential solutions that could happen.
But at the same time, of course, we have some previous examples which we can share. So it’s easy to see some real-life examples.
I have a feeling when you get the question of how much it costs that there has to be inevitably the qualifier of: well, it depends.
Luca Gonnelli: Yeah, it really depends a lot on the needs of the client. But yeah, the pricing works normally through a setup fee, which covers the whole project setup. Normally we start from a couple of months of work, and yeah, the pricing can also vary a lot in function of what kind of data we are using. There’s licensing of this data, or how complex the output is, if it’s more generative so we are actually designing our, creating an output, which is changing every time or if it’s that more relying on some rules that we’d predefined.
But we normally start with this kind of two months of collaboration where with our design team and animation team and technical team to build a project, and that’s a one-time fee covering all that, and then when the project after testing, a lot of testing after testing when things are going live you subscribe to a much smaller but recurring fee based on the function of how many videos you need to create or how many animations web animations, and so also that is very variable, but yeah, it’s a closer to SaaS kind of approach.
So typically, you might have a significant, depending on the brief, upfront cost to put it together, but after that, it’s just it just becomes an operating line item?
Luca Gonnelli: Exactly. The first year, you’re investing in creating this format, and the more you use it later, the more it’s going to be cost-effective. Of course, the one-time fee, it’s only due the first time. Normally Algo projects are running for around maybe two or three years and of course, sometimes we also do updates and work on refreshing the project after a while since it’s a video project, so we can always do that later. But yeah, normally it’s an investment in the first year, but then it’s paying off in the following ones.
With AI and all the generative stuff emerging at a dizzyingly fast pace, is it worrying or confusing or whatever to try to stay on top of this and stay relevant to when you’ve got all these little apps coming out saying, you can do all of this automatically. You don’t even need to have a photo library anymore, you can just generate it.
Luca Gonnelli: I really find it super, super exciting because I’m trying to follow it as much as possible. Of course, it’s moving very fast. But how the way we are approaching this is really to see which tool is the most effective in helping us obtain what we want to obtain for projects.
So just to give you an example we are using GPT4 right now on a project to basically summarize an article and turn an article into a video, and that’s amazing how you can just simply use the summarization feature which is super well done and so yeah, we are actually making GPT4 write the script for the video based on just a long-form article which we’re passing to it, and that’s the only suggestion. So you copy-paste the URL of the article, you click a button and you will see structures adapting function of the content that’s been analyzed by GPT4, and so that’s super interesting to see how this can evolve and how to use, for example, the next step could be using another AI service which is called Play.HD, which would be love, which is voice synthesis. It’s like creating human-sounding voices, really super realistic voices that are almost indistinguishable from voiceover actors to basically record voiceovers for the videos so they sound warm.
So just with those two things you’ve written a script based on a long, preexisting, long-form article and you have a voiceover for that, and so then we focus on the design side. But yeah, that’s super exciting. Of course, we’re not developing ourselves. We’re a small team, and we’re not developing our own machine-learning algorithms, but we’re literally using all the interesting ones that are coming our way for all the projects.
Really interesting. Luca, thank you so much for spending some time with me.
Just before we go, where do people find you online?
Luca Gonnelli: Sure. Our website is algo.tv and most of our socials are also @algo.tv.
Very simple. All right. Thanks again.
Luca Gonnelli: Thanks to you.
Leave a comment