As you probably know by now, at Trendio, we’ve built best-in-class AI technology to match brands with high-performing affiliates. We relentlessly launch campaigns for our brand clients, to promote at scale their products on video social networks. Our commitment to performance is upheld through sophisticated algorithms that precisely identify creators best suited for a given product and brand. This intricate challenge demands consideration of multifaceted attributes, quickly becoming unwieldy without advanced technological solutions.

Indeed, achieving this scale and efficiency is virtually impossible without automated systems and robust code. A manual approach to selecting and engaging with the optimal set of affiliates would consume countless hours for a single campaign, rendering scalability and efficiency unattainable. Through continuous learning and overcoming various challenges, we have refined our methodology.

In a previous article, "What makes a great affiliate on TikTok Shop" we explored the key factors predicting a creator's performance and efficiency for brands.

The article you are ready delves into the data science and operational insights that underpin our "Creator Selection & Matching" process.

Data: The Cornerstone of AI Performance

The efficacy of AI models is directly correlated with the depth and quality of the data pool they access. Social media platforms like TikTok Shop and YouTube Shopping are invaluable sources, providing extensive data points on their affiliated influencers. These creators, once eligible, promote products through video and live formats, primarily earning through affiliate commissions. Crucially, these platforms increasingly make rich information about creators, their content characteristics, and audience impact publicly accessible.

Let's examine the key data dimensions we map from TikTok Shop and YouTube Shopping.

TikTok Shop Creator Attributes

TikTok Shop stands out for its comprehensive exposure of affiliate information. Our mapping includes:

Creator Profile:

• Category of products
• Followers count
• Type of content (Video or LIVE)
• Agency or Independent creator
• Average commission rate
• Location
• Brand collaborations

Audience Demographics:

• Age
• Gender

Performance Metrics:

• Sales
• Items sold
• Average views per video
• Average views per LIVE
• Engagement rate
• Estimated Post Rate
• Brands they’ve worked with
• Number videos posted
• Number LIVEs streamed
• Promotion Performance Score

YouTube Shopping Creator Attributes:

While YouTube's service is a more recent entrant in this specific domain, it provides an extremely valuable dataset. Our key mapped dimensions include:

Creator Profile:

• Category
• Subscribers
• Avg views
• Number videos posted
• Location

Audience Demographics:

• Age
• Gender
• Top countries

Fortunately, APIs facilitate programmatic access to this information, enabling its integration into a scalable data pool for processing.

Our primary data sources are, of course, the native social networks themselves, supplemented by additional external databases. Undeniably, native platform data consistently surpasses external sources in terms of quality and exhaustiveness.

Operational Excellence: Ensuring Data Integrity and Freshness

Size of the data pool

Beyond the sheer volume of data, the scale and quality of our operational processes are equally critical. To support thousands of brands launching tens of thousands of campaigns, each involving hundreds of affiliates, a truly expansive and meticulously maintained creator pool is indispensable.

Trendio currently boasts a database of over 2 million TikTok Shop creators and more than 100,000 YouTube Shopping creators.

Data pool quality

We are committed to comprehensively qualifying each creator who enters our system. Upon initial identification, we begin with a foundational set of attributes, progressively enriching each creator record with data from diverse sources.

Furthermore, the "freshness" of information is paramount. In the dynamic social media landscape, creator profiles evolve rapidly. A viral content piece can propel a creator onto a rapid growth trajectory, quickly rendering their existing profile data obsolete within our systems. Identifying rising stars is crucial for capitalizing on optimal opportunities.

To address this, we have developed proprietary in-house processes that continuously update and refresh our information on a weekly basis. Our expertise lies in optimizing data retrieval speed, frequency, and query precision to maintain database stability and prevent system overload. Concurrently, we have implemented robust monitoring tools to continuously assess system health and data integrity.

Process Quality Surrounding the Data Pool

In addition to the core database, we employ sophisticated procedures to process and consolidate data, generating accumulated values that further enrich our predictive models.

Concurrently, we run processes that gather real-time information on creator activity and impact within our campaigns, including sample requests, product endorsements, and content publication. Their performance is tracked using a proprietary algorithm that facilitates state-of-the-art attribution, enabling brands to precisely understand campaign contributions and significantly boost the efficiency of our AI-driven recommender engine. This ecosystem-specific data provides the most accurate insights into the profiles of impactful creators, information typically unavailable directly from social networks and requiring bespoke in-house development.

Ultimately, this comprehensive intelligence is seamlessly integrated into the dashboards we provide to our brand clients. Data from four distinct sources converges in these dashboards, offering an unparalleled, holistic view of video-commerce campaign outcomes.

How do we refresh our creators and how often do we do it?

We refresh all our creators' profiles weekly. We consolidate our tables to store accumulated data daily. We collect the engagement of the creators and the impact of their videos daily.

In parallel, our AI algorithms are retrained periodically (we will keep the exact value secret for the time being) and we pre-load clusters’ results daily.

AI Algorithms: The Apex of Predictive Matching

As noted, the richer and fresher a creator's profile, the more accurately our recommender engine can identify optimal matches. Our advanced AI algorithm employs multi-level clustering techniques, leveraging dozens of integrated features to categorize creators. This enables the algorithm to analyze a previously unworked-with creator and "predict" their potential success in promoting a brand's product. This represents a highly complex and sophisticated process, given the multitude of dimensions it must process to operate efficiently and reliably.

We could elaborate further on the intricacies of training and deploying these models within our infrastructure, or the complexities of the involved MLOps but as Rudyard Kipling used to say at the end of his short stories: “That is another story”.

Curious about secrets we didn't reveal? Stay tuned, our next article might just hold the key to our Secret Sauce.