Jan 5, 2024

Table Of Contents Stas Tushinskiy is the CEO and co-founder of Instreamatic , a platform that offers AI-powered voice and audio marketing solutions to enable brands to better engage with consumers. You previously Co-founded Unisound, an audio ad agency. How did this experience lead you to conceptualize launching an AI voice marketing company? My experience at Unisound was foundational in understanding the evolving landscape of digital audio advertising. We were at the forefront of recognizing the growing demand and potential for audio ads in a digital space. A key takeaway from my time there was the realization that personalization and contextualization significantly enhance the effectiveness of advertising, including audio ads. This understanding became a cornerstone for the vision behind Instreamatic. At Unisound, we observed a gap in the market for intelligent, responsive advertising solutions. We envisioned using AI not just for targeting but also for creating a more interactive and engaging experience. This led to the idea of an AI-driven marketing platform, which would revolutionize how we interact with ads. Could you share the genesis story of launching Instreamatic? Originally, Instreamatic was born from a vision to transform how audio publishers monetize their content. Initially, our focus was on serving audio ads for monetization, which remains a significant part of our operations. As we delved deeper into the industry, we identified a substantial opportunity in AI for creative optimization. This realization was pivotal in shaping our direction toward integrating AI technology more deeply into our services. The convergence of our expertise in audio advertising and the advancements in AI technology was the catalyst for Instreamatic. We saw the potential to not only serve publishers but also to enhance the overall ad experience for users and advertisers alike, paving the way for a more dynamic and efficient advertising ecosystem. What were some of the initial AI/ML technologies that were used? We started with a simple classifier. It’s a supervised machine learning method where the model tries to predict the correct label of given input data. Then, we enhanced our classifier by using embeddings. Eventually, we didn’t limit ourselves to just NLP technologies. New ideas and challenges presented us with fresh obstacles and, now, our arsenal includes text-to-speech synthesis and zero-shot voice cloning. How has generative AI changed your technology stack and how do you deploy it? Generative AI has brought significant changes to both our technology stack and deployment strategies. Our current technology stack includes advanced machine learning libraries and frameworks that support generative AI models, particularly for text-to-speech synthesis and zero-shot voice cloning. We utilize high-performance computing resources to train these models, as they require substantial computational power. This involves leveraging GPU-accelerated hardware to handle the intensive processing demands. For deployment, we rely heavily on cloud-based solutions. This offers us the scalability needed to manage the heavy workloads of generative AI applications. We use containerization technologies like Docker and orchestration tools like Kubernetes to manage and scale our applications efficiently. This setup ensures that our generative AI models can be deployed rapidly and scaled according to demand. Our CI/CD pipelines are optimized for machine learning workflows. We use tools that enable us to automate the training and deployment of models, ensuring that they are always updated with the latest data and algorithms. This automation is crucial for maintaining the efficacy of our generative AI applications. In terms of data handling, we have implemented robust data processing pipelines. These pipelines are designed to handle large volumes of data efficiently, which is essential for training and operating generative AI models. We ensure that data is processed and stored securely, adhering to best practices in data security and privacy. Overall, the integration of generative AI into our technology stack has led us to adopt high-performance computing resources, cloud-based infrastructures, containerization for scalability, automated CI/CD pipelines for machine learning, and secure data processing mechanisms. These technical elements are fundamental to supporting the advanced capabilities of our generative AI applications. Instreamatic specializes in what you call contextual video and audio advertising—how do you define that? Contextual Advertising taps into current advances within generative AI to significantly alter what’s possible with video and audio ads. The result for businesses is increased brand engagement and ROI. Contextual Ads offer an essentially unlimited capability to continually generate and A/B test new creative content relevant to the listener’s specific context and environment. The fact is the advertising industry has been up against declining engagement rates across ad types for years. That’s probably no surprise to anyone, as consumers demonstrate increased screen fatigue and resistance to generalized advertising that relies on bombarding audiences with ad quantity to earn conversions. While ads that demonstrate more specific original content and higher relevance to the consumer earn higher engagement, the time and cost investments required to manually produce and manage separate ad copy for each individual consumer’s context are extremely prohibitive. Our contextual audio, video, and connected TV (CTV) ads are powered by AI to buck this low-engagement trend by enabling advertisers to make every ad hyper-relevant and precisely targeted to the consumer hearing it. Consider a traditional 30-second audio ad spot: a hired voice actor might record a few ad copy variations at most, not enough for the listener to be particularly surprised, or to necessarily capture their attention. Contextual Ads are capable of enhancing that traditional ad content, using generative AI to synthesize the same actor’s voice and automatically generate thousands of ad variations across a campaign. Contextual Ads are especially useful for revitalizing longer ad campaigns (in the 3-6 month range). Traditionally, these campaigns are very vulnerable to creative fatigue: audiences get the same creative over and over, inevitably leading to decreased engagement. Our technology solves this challenge by making it simple to refresh creatives weekly. For retailers with weekly-updated product offers, for example, our automatic ad generation is similarly ideal for keeping those campaigns current and fresh. How realistic is it for brands to expect AI to hyper-personalize ads? It’s now fully realistic, as demonstrated by generative-AI-powered contextual advertising . Contextual Ads can feature hyper-personalized details, including the listener’s location, the time of day, the name or type of app or platform they’re using, and the activity they’re engaging in, whether it’s listening to a podcast, playing a game, etc. Contextual Ads can even include variables such as naming local storefronts and addresses, local in-store promotions, promo codes (unique to each channel to enable performance measurement), travel destinations with specific offers, and much more. These ads can also name the nearest local storefront where a listener can interact with the brand and redeem the deal offered in the ad. This same targeting capability ensures that ad campaigns reach vetted audiences that are most receptive to the products and solutions being offered. These ads are all generated and delivered without recording new voice or voice-over content. Can you discuss the core offerings that your customers have access to? From a brand perspective, our Contextual Ads platform takes a single original voice sample and script, identifies the set of parameters unique to each individual listener, and uses our voice AI capabilities to seamlessly produce and serve audio, video, or CTV ads aligned with those specifics. For example, a Contextual Ad generated for a particular user could begin, “Hope you’re enjoying your podcast on this rainy morning in Chicago, I just wanted to quickly let you know that coffee is buy-one-get-one-free at Jake’s Coffeeshop all month.” Whereas producing that same ad creative with prerecorded audio and branching logic would be an all-but-impossible undertaking, the voice AI behind Contextual Ads prepares this creative on-demand—automatically and in real-time. From a publisher's perspective, AI-driven voice, video, and CTV Contextual Ads offer a game-changing innovation with no complex integration required. Contextual Ads work with all demand-side platforms (DSPs) and ad servers supporting VAST tags, offering instant scalability. Publishers can also leverage our ad network to reach more than 6 billion impressions globally at no platform cost: technical costs are included in media spend when publishing within network. Could you share some details on the process of launching an ad on the platform? Launching an ad on our platform literally takes just minutes. The brand or agency user simply writes ad copy with or without help from AI, then either chooses a royalty-free voice from our Voice Library or clones their own voice talent. Users can also upload any additional assets necessary (background music, video footage, banners, etc). The user finalizes the ad, and the platform provides versions ready to serve—either via the VAST tag (the industry standard for ad trafficking), or as downloadable media files ready to go for any digital and broadcasting environments. These AI-enriched ads not only increase the performance of video and audio ad campaigns by enabling hyper-personalization at scale, but also slash the cost to produce campaigns and reduce ad creation time from weeks to minutes. For campaigns with 50+ versions, users experience a ~10X cost decrease. Our technology offers similarly decisive benefits for single-creative campaigns as well. The platform is also a great instrument for sales teams to quickly produce ad mock-ups for their clients without engaging with production and creative teams at an early stage, since our AI can write copy and fully produce custom ads. What’s your vision for the future of AI advertising and marketing? I really do see a future where customers aren’t annoyed at (or tuning out) ads because each one is now relevant and more interesting to them, with brands are that much more capable of reaching the right audiences at the right moments with the perfect contextual message. That’s obviously a sea change from where the industry is now, but I do believe that’s where we’re headed—and AI, leveraged strategically, is making it possible. Contextual Ads are also going to continually get better at capturing listeners’ attention because they speak precisely to their context and their needs, especially in the privacy-first world where user targeting gets harder and harder—so context targeting is the only efficient mechanism for boosting ad performance. Our advanced generative ad AI can create unlimited new creatives to address each listener as an individual. The result is an uplift in listener engagement, greater ad ROI, and more meaningful customer connections for brands. Thank you for the great interview, readers who wish to learn more should visit Instreamatic .