How does Blask search for new iGaming brands?
Blask was created with the goal of providing our partners with information on all iGaming brands in every country. In this article, we will share the story of our unsuccessful solutions and how we eventually learned to use AI models for easy and fast brand discovery.
Why is it important to have data on ALL the brands in the market?
Blask believes that a comprehensive market analysis of a country is only possible when you have a bird’ s-eye view of not only the most well-known and popular brands but also every small player. This allows you to promptly identify rising stars and learn from their experiences.
Compare your brand’s metrics with those of your closest competitors.
For young or small brands, it’s too early to compete with the top 3 players in the market, so it’s better to focus on brands of a similar size. First, overcome your nearest rivals, and then switch to competing with more serious competitors.
💡 For instance, as of June 2024, Blask is aware of 168 iGaming brands in the Brazilian market alone.
Obtain comprehensive market data and accurate information on its fluctuations.
Since Blask analyzes all brands and their market share, we can provide the most complete and reliable data. Suppose a brand ceases operations, or a significant sporting event occurs. In that case, you can track how it impacted the popularity of iGaming activities in the country and which brands gained new players.
Learn from the best and adopt their best practices.
At Blask, you can sort brands by YoY (Year-over-Year) or MoM (Month-over-Month) growth, giving you a list of brands that demonstrate outstanding growth percentages compared to others in the chosen country’s market. Assign your marketing department to study the advertising activities of these brands and replicate their strategies!
Our journey in Brand Discovery.
Blask has come a long way from manually adding brands to a fully automated process of discovery and validation with over 90% accuracy.
It all starts with keywords and Google search results.
First, we gather all relevant keywords for the iGaming industry in English, such as “online casino”, “sports betting”, “online betting”, and others. Then, we translate this list into the language of the country where we want to find brands. Finally, we extract from Google Search a list of all websites that appear for these keywords in the specified country.
As you can imagine, a search for “online casino” can yield an iGaming brand’s site, a referral link advertisement, a blog post reviewing different casinos, an article about the dangers of gambling, or even just news about a new brand’s launch. This list contains too many irrelevant sites. Thus, the next step is to classify each of these websites.
First try: parsing HTML code of websites.
Blask started with the simplest approach and achieved approximately 70% accuracy in recognizing iGaming sites.
We extracted text fragments from the HTML code and initially analyzed them using a simple algorithm that counted the frequency of specific keywords. Later, we switched to a large language model for more accurate identification.
The main drawback was that perfect HTML code only exists in textbooks. In real-world websites, the HTML contains too much irrelevant clutter, making it difficult to find useful text.
Second try: text recognition through OCR.
The method of text processing and website classification was identical to the previous approach, with the only difference being how we obtained the text data. Instead of parsing HTML code, we took a screenshot of the website and used OCR to recognize the text on it.
This approach improved the accuracy of classifying websites in English but performed significantly worse for other languages. Additionally, some text can be obstructed by pop-up banners.
Therefore, Blask’s R&D department began searching for other ways to classify websites.“Our R&D team is always on the hunt for better ways to classify websites. We tried OCR for text recognition from screenshots, which worked well for English sites but fell short for other languages. We’re now diving into more advanced techniques to ensure our system stays sharp and accurate across all markets” — CEO of Blask Max Tesla
Third try: text description of website screenshots using a neural network.
Text analysis doesn’t provide a 100% guarantee of correct recognition because it ignores the visual elements of a website, which a human can instantly use to determine if a site is an iGaming brand, an affiliate platform, a personal blog, or a news outlet. Therefore, we turned to artificial intelligence for help.
💡 We chose LLaVA (Large Language-and-Vision Assistant). This neural network takes a screenshot of the website and a text prompt like “describe the image” and generates a textual description of the image.
What is LLaVA?
However, with iGaming brand websites, the results were even worse than with HTML parsing. The neural network described each screenshot and created beautiful stories like “I see a football in the right corner, and there are casino chips in the center,” but this wasn’t enough to distinguish an online casino site from a blog about online casinos.
Additional drawback: LLaVA turned out to be too slow. The response time on a server with an NVIDIA A100, one of the most powerful server GPUs, was about 20 seconds. Analyzing 5 000 websites would take 100 000 seconds or 1 666 minutes or 27 hours. This would be justified with an accuracy of 90% or higher, but it is completely unacceptable at around 60%.
The current version of the Brand Discovery process.
Blask still begins the process by compiling a list of keywords and parsing a list of websites from Google search results. However, we now use two neural networks for site classification.
Step 1: Recognizing iGaming-specific elements through computer vision.
We trained an AI model to recognize all elements characteristic of iGaming websites:
- Images of casino chips
- Logos of games and slots
- Betting odds tables
- Images of live dealers
- Images of card games
- Sports attributes
To train the model, Blask created a dataset from screenshots of websites that are definitely iGaming brands and classified each element of these sites. We then augmented the dataset by creating copies of the screenshots with slight distortions, such as altering some elements by 1-2%. This helps expand the original dataset, improving the accuracy of training and, consequently, the accuracy of element detection.
After the computer vision model produces a list of all detected iGaming-specific elements in the given screenshot, we can analyze their frequency and provide a fairly reliable verdict.
💡 The idea is that an iGaming brand’s website will have many such elements, whereas an informational article, such as on Wikipedia, will only have 1-2 images.
Step 2: Supplementing data with text analysis.
Now that we know the website is definitely related to the iGaming industry, we need to identify the brand’s website and discard affiliate sites and detailed review articles.
This is where the second neural network comes in. Its task is to analyze the text for the presence of referral links, game or brand reviews, and vocabulary that is not typical of iGaming brands.
This additional check increases the accuracy of recognition to over 95%.
Step 3: Final human review.
Neural networks provide a high level of accuracy, but still not 100%, so the final decision is made by a human.
Blask sends all websites with at least a slight confidence that they might be brand sites for manual moderation. The issue with less than 100% accuracy involves two main ideas:
- An affiliate site or news portal might be mistakenly identified as a brand site.
- A brand site might be mistakenly identified as an affiliate site or news portal.
In the first case, the only consequence is extra work for a human, but in the second case, we risk missing a brand. Therefore, despite the high accuracy of our neural networks, we still entrust the most critical work to humans.
Blask strives for maximum accuracy in all its data, so we cannot accept even the slightest chance of missing a brand.
Using this approach to find all brands in a new country, Blask only needs to allocate GPU resources to run the two optimized AI models and then send a short list of websites to a human for final approval.
Blask regularly initiates the process of searching for new websites to add data about the newest and smallest brands promptly.
How to find all brands without Blask?
In short, you can’t.
It will require a lot of effort and even more time. Connections within the industry will also be beneficial. And, of course, you’ll need luck — because in iGaming, you can’t work without it.
We can highlight the following methods:
- The top 5 brands are usually known to anyone who has seen internet ads or watched sports broadcasts.
- A basic Google search will show another 20-30 brands.
- After searching on Google, Facebook, YouTube, and Instagram, keep a close eye on ads — you might see an advertisement from another brand.
- You need to build contacts in the iGaming industry and constantly ask everyone which brands they know.
- You can become a regular visitor at iGaming conferences and network with brand representatives there.
- You can spend several weeks manually sorting through hundreds of search result pages.
Finding all iGaming brands in a country’s market is a very difficult task. And after that, you still need to gather data on each of the brands. If you need to do this for multiple countries at once, it’s easier to climb Everest. And in the end, it’s not enough to gather data just once; you need to continuously keep it up to date to always stay ahead of the competition.
We strongly recommend delegating the brand search to our two neural networks. With Blask, you only need to analyze the data and build your brand based on data-driven decisions.
Explore the iGaming industry with Blask’s AI models.
With Blask, you can collect all possible insights on the iGaming industry. You will always have the most comprehensive list of brands across all countries. No one else has gathered such extensive data.
If your brand is just entering the market, it will be especially useful to see all competitors, from the largest to the smallest, to carefully track your own progress.Frequently asked questions.
Discover every iGaming brand before your competitorsBook a demo
Read more
