Years of top-notch features development: edrone’s AI portfolio

At edrone, we’ve been at the forefront of this trend in e-commerce, developing a wide range of AI-based features that help our clients achieve their goals using the same technology as the biggest players in the industry. Our portfolio is full of innovative solutions that leverage the power of AI. 

Time to sum it up!

In this article, we’ll take a closer look at some of the key AI features we’ve developed over the years, showcasing how they can benefit businesses of all sizes and industries. So whether you’re just starting with e-commerce or looking to take your business to the next level, read on to learn more about edrone’s cutting-edge AI features.

Voice Search Widget

Let’s start with our Crown Jewel: Mobile Search Widget, the Voice Interface. 

Imagine being able to shop online without even having to type painstakingly on another and another version of a query or once again set an additional filter. 

With edrone’s Voice Search Widget, that’s precisely what you can do. Our R&D’s child allows you to browse and purchase products using nothing but your voice. Simply speak your search query and let our advanced natural language processing (NLP) algorithms do the rest. So whether you’re looking for a specific product or just browsing for inspiration, our voice search technology is the ultimate shortcut to finding exactly what you need. And with recent recognition at the Mobile Trends Awards in the “Technology boosting Mobile” category, you can be sure that our widget is both cutting-edge and highly effective. Developed over three years by a team of PhDs and top NLP specialists, the Voice Search Widget is the future of voice commerce. 

Mobile first

In today’s mobile-first world, nearly 70% of e-commerce traffic comes from mobile devices. However, the conversion rate on mobile is often lower due to clunky search interfaces that require tedious clicking through facets and filters. That’s where edrone’s Voice Search Widget comes in. By providing a voice interface for product search, our widget delivers better search results, shortens customer paths, and boosts mobile sales.

Using the Voice Search Widget is simple and fast. Customers can simply speak their search query, such as “I’m looking for the perfect accent for my room. I like the Scandinavian style, but my wife prefers boho. Please show me what you’ve got.” Then, the widget’s Automatic Speech Recognition (ASR) technology quickly recognizes the query and passes it through the Natural Language Understanding (NLU) algorithms to detect all relevant pieces of information, including product type, price, size, color, gender, and context.

Once the customer’s query has been processed, the Interactive SERPs feature displays products matching the query on the screen, allowing the user to change, add, or remove search criteria by voice for an exploratory search experience. The Conversational module also guides the customer and asks for additional criteria if the query is ambiguous, similar to having a conversation with a salesperson in a brick-and-mortar store.

Significant speed boost

Voice search is not only fast and convenient, but it’s also the most natural form of communication. In fact, it’s 3.7 times faster than classic search methods and is the preferred form of interaction for Gen Z and Millennials. Plus, it provides higher accessibility, breaking through the digital divide and enabling people of all ages and abilities to shop online easily.

By incorporating edrone’s Voice Search Widget into your e-commerce strategy, you can stay ahead of the curve and provide your customers with the most convenient and natural shopping experience possible.

So why waste time typing out your search queries when you can just speak them into existence? Check on edrone’s Voice Search Widget and experience the ultimate in e-commerce convenience!!! 

Marketing Machine

Edrone’s Marketing Machine is an interactive recommendation feature that analyzes the behavior of each customer who visits a store’s website, identifies their interests, compares them with those of other customers, and displays tailored offers to each client. As customers browse the website, their behavior is analyzed to create personalized product recommendations, creating a unique experience for each visitor.

This approach, known as Segment-of-One Marketing, allows for a website that adapts to each customer, resulting in tailored offers continually refined with each visit. This technology is used by major companies such as Amazon and Netflix. Its implementation can improve the customer experience and increase sales by retaining their attention with personalized displays that align with their interests.

Interactive Recommendation Frame

edrone’s Marketing Machine is an interactive on-site recommendation frame. edrone’ system analyzes the behavior of each customer entering the store’s website, it gets to know his interest, compares it to the interests of other customers with similar interests, and, thanks to that, it displays tailored offers to each client.

1. The customer enters your website — they builds his profile with every click.
2. Based on his behavior, we create recommendations for best-fitting products.
3. Thanks to the principle of Segment of One Marketing, the website will have as many versions as customers and visitors.

We treat each client individually and create a tailored offer for each person. Therefore, with each visit and even refreshing the site, the offer will be better suited to the visitor.

What are the advantages of using it?

• Implementation of the same technology used by giants such as Amazon and Netflix.
• Make your visitor’s experience improve every time he returns to your store by personalizing your displays to what interests him.
• When showing what your visitor wants to see, you retain his attention; the more attention you have, the more sales you make.

Collaborative Filtering

Machine Learning at its best! This recommendation style focused on all users’ activity. Based on 2005 introduced Slope One recommendation algorithm. 

Slope One uses the concept of item-based collaborative filtering to make recommendations. It is based on the assumption that if users frequently rate two items similarly, they are likely to be similar.

The algorithm works by calculating the average rating difference between each pair of items that the same user has “rated”, by visiting the product card, adding to the cart, or purchasing it. This gives a measure of the “slope” between the two items. The algorithm then predicts the rating of an item by taking the average of the user’s ratings for the things that are similar to it, weighted by the slopes between them.

The Slope One algorithm is relatively simple and computationally efficient, making it suitable for large-scale recommendation systems. However, it has some limitations, such as the fact that it only considers the average rating difference between items and does not consider other factors that may affect the similarity between them.

• Product view
• Add to the cart
• Purchase

There is no better choice for your 404 error page display. This tactic ensures an almost seamless shopping experience, even for the most impatient customers.

Market Basket Analysis

In machine learning, this is a sub-technique of “Affinity Analysis”; however, with respect to eCommerce, we should say “purchased products analysis.” The recommendation is based on products that are often bought together.

• While browsing the product, the client sees recommendations for the displayed item.
• While the client is watching a category, the recommendation algorithm displays products based on a few items in that category.

Market Basket Analysis is a data analysis technique used to identify the relationships between products or items that customers tend to purchase together in a transaction, such as a shopping cart or a receipt. It is also known as association analysis or affinity analysis.

Market Basket Analysis aims to identify the patterns of product co-occurrence and use these patterns to make recommendations for product bundling, promotions, and cross-selling.

Deep Dive

Market Basket Analysis typically involves the following steps:

• Data Collection: The first step is to collect transaction data, which includes information about the products purchased in each transaction.
• Data Preparation: The transaction data needs to be cleaned, filtered, and transformed into a suitable format for analysis. This may involve removing duplicate transactions, handling missing values, and creating a transaction matrix.
• Association Rule Mining: This is the process of identifying the relationships between items that tend to co-occur in transactions. Association rules are generated based on the frequency of itemsets and the level of support and confidence.
• Rule Evaluation: Once the association rules are generated, they need to be evaluated based on their relevance and usefulness for the business. This involves identifying the rules that have the highest lift, which indicates the strength of the relationship between items.
• Application: The final step is to apply the results of the analysis to make business decisions. This may involve creating product bundles, designing targeted marketing campaigns, or optimizing product placement in the store.

Quick Fact: Market Basket Analysis is widely used in the retail industry, but it can also be applied to other domains, such as healthcare, finance, and telecommunications.

Email Recommendations

In the same way, we recommend on clients’ sites; we can embed frames with products within your mailing! Simply pick the product, and it will appear in your mail, allowing clients to quickly convert to the website and purchase recommended product (maybe including some cross-selling thanks to Marketing Machine 🙂 ) 

Product Linker 

Optimizing a website is a comprehensive task that requires a significant amount of effort, especially when it comes to activities like internal linking that can be time-consuming and require specialized knowledge. Fortunately, we have a solution to help streamline this process.

Imagine being able to save time and focus on more critical tasks instead of manually searching for products that match every text on your blog. Our smart tool can automate this process for you, generating relevant product suggestions to enhance your content.

Tool first of it’s kind

Our tool, Linker, automates the process of internal linking in online stores, making it easier for SEO specialists to add links to their website’s content. Linker suggests up to nine relevant products for each paragraph, regardless of industry or product range.

Linker uses the SentenceBERT algorithm, a variation of Google’s BERT, to provide precise product matching and help increase conversions. With 68% of purchases originating from search engines, linking relevant products can shorten the conversion path and improve overall sales.

Long Story Short: Using our functionality, you can save time, improve the efficiency of internal linking, and enhance the user experience on your website.

How it works?

1. The linker will use product descriptions to learn about your offerings. 
2. After entering expert text, it will divide it into paragraphs…
3. It will analyze each of them for matching products.

Why Product Linker?

1. Our research proved that even for a skilled SEO expert, it takes around 6 minutes to prepare links for an article. 
2. 6 minutes alone does not sound like much, but you probably have around one thousand articles, which gives us probably more than one hundred hours of work. 
3. Moreover, some links expire or a new product version is released. In other words, the job needs to be done again.

Here comes product linker, our new feature in edrone with AI-generated product suggestions for your content.

1. We will need now a blog post. I select paragraph, or larger portion of text, and copy it. Now the product linker comes into action. 
2. We paste the copied text. And in a few seconds our AI algorithm will suggest products per paragraph of your text.
3. If you like the results, you can now copy URLs to clipboard via this button and link them on your blog post for the given paragraph. 
4. If you find some of the results dissatisfying simply regenerate links. This will also improve the future accuracy of your results.

Tag Propagator [BETA]

The technology behind Product linker, has various applications. On of them we experimented with is Tags’ Propagation

Business Value:

Store search engine optimization for advanced search criteria, among other things:

• Symptomatic – the customer knows his need but does not know what exactly will satisfy it. 
• Thematic – the customer would like to find a product in a specific group; products are not directly related to each other.
• Based on compatibility – the customer is looking for complementary products or products that match the one they already have. 

Improving on-site SEO activities:

• Labeling and tagging, products and on-page content.
• Internal linking – Improving Domain Authority, Page Authority, and especially Topical Authority index.

Application: Tag Propagation

The promoter bases its effectiveness on finding similarity of meaning between content on the page.

Propagator can support the process of tagging (a.k.a. labeling, tagging, labeling) products in an online store. In most cases, this is a manual process, requiring a significant amount of work, proportional to the number of products in the store and their categories.

How does it work? An example would be an online bookstore. Valentine’s Day is coming up, and you would like to improve the search results for this query on your site. 

1. Some items in the in-store store are labeled “Valentine’s Day” and others are not, although as much as possible they should be. 
2. In the first step, we use correctly labeled products. They will serve the propagator as a reference/reference point.
3. The basis of tag propagation is product descriptions. 
4. Based on the descriptions, the propagator learns “what the description of a book looks like”, which is a great gift for Valentine’s Day.
5.  It then checks which items should, and don’t yet, be so tagged. 
6. The suggestions are then verified by the store manager in a CSV file.

Using the propagator, we can fill in any missing product tags. It will work analogously for the tags: Fishing, Bicentennial, or Science fiction.

Additional Tagging

Propagator can also help with Creating new tags. Example: Tool store and Topical Search applied. 

The spring season is approaching. The Garden Tools tag will help with activities related to promoting relevant equipment. Just tag a portion (minimum 10) of selected products, and the propagator will find more matching items. 

SEO

The propagator can also greatly improve SEO activities. It can also significantly improve activities related to search engine optimization of the site. 

Image Generation

Applies to: Mail, Drag’n’drop

Artificial intelligence has not one face. Therefore, the next functionality allows you to generate any image based on a text prompt. 

You will do it in a very similar way to headlines or call to action. Enter a short description of the image you would like to be included in your email—the more accurate the description, the better. Feel free to experiment, too; if the illustrations don’t match what you would like to convey in the message, try describing the image in other words. 

This is much faster and cheaper than spending long minutes (sometimes hours) on stock illustration portals. 

Header and CTA Smart support

Applies to: Mail, Drag’n’drop

In this functionality, we use the API of OpenAI, an organization that has been talked about recently due to the popularity of ChatGTP.  

Based on a few slogans and concepts describing your message’s subject, we will generate several proposals for catchy headlines. It only remains for you to choose the one that will best influence your customers. You can also select the statement’s tone, allowing you to match the headline to the content even better. 

Nothing prevents you from modifying the suggestions or from generating further variations based on the modified ones.

Button prompts

Well, that’s not all! All of the above options are also available in the button text box. We know that writing such short forms can be tiring in the long run, and creativity under time pressure has its limits. 

Simply type in a few slogans that describe what you want the user to say, set the tone of the statement and enter the generated Call to Action proposal.