Algorithmic Transparency: Consumers

CRM option 7-morethanfood_discovery_banner_1440x810.png
  • png

When you open the app, you might wonder why a particular restaurant or store is being shown to you. In this post, we'll explain how our recommendation and search algorithms help you find what you want or need among the thousands of choices available. The information is taken from the Wolt Algorithmic Transparency Report 2024.

Ranking & Recommendation Algorithms

The first screen that users of the Wolt App are greeted with is called ‘Discovery’. Discovery lives up to its name – it's a place where you can explore what’s available in your neighborhood.

You can also choose to view Discovery by only looking at a list of restaurants or a list of stores. Due to their differences, the recommendation system for stores operates differently compared to restaurants, and is detailed after the following section focusing on restaurants.

Restaurants in Discovery are ranked by: 

  • Consumer’s location: To ensure swift deliveries and to maintain the quality of the food, we prioritize showing you restaurants that are closer to your location when you open the app. We want to avoid displaying venues that are too far away, as this could impact the timeliness and freshness of your order.
  • Opening hours of the restaurant: We don’t want to show you a restaurant serving really good food if it is not open and available when you open up the app looking for something to order.
  • Time of day when you open the app: We humans tend to eat roughly around the same times of the day – breakfast in the morning, lunch around noon and dinner in the evening. We therefore think that you would be more interested in breakfast-related items in the morning and dinner options in the evening.
Discovery, Helsinki.png
  • png

Carousel Ranking Algorithm

On the Discovery page, you'll see different content types displayed in a carousel format. A machine learning algorithm decides where each item goes, favoring what works best in each city based on the number of impressions and the number of orders of an item. This changes multiple times a day. So, depending on your location, time and highest conversion rates in a city, you'll see different content in different positions. There's also a manual option for local operations to control specific placements. For example, if we want to promote relevant venues for Valentine’s Day or if there is sponsored content.

Collaborative Filtering Algorithm

We also try to make the content in Discovery even more relevant to you by basing it on what other consumers have ordered and what we think you might also enjoy. This filtering is based on consumers’ purchase behavior, a method more commonly called ‘collaborative filtering’. 

Let’s take a look at how this works in practice through the lens of two fictional consumers living in Wolt city – Alice and Bob. In the real world, the names of the consumers would remain unknown. The user data used by the recommendation algorithm is stored and processed through randomly generated IDs, ensuring the anonymity of the people involved.

Algorithmic Transparency Report 2024_page-0009 (1).jpg
  • jpg

What Alice and Bob have in common is that they have both ordered and liked the food from the same pizza restaurant. ‘Liking’ meaning that they have either ordered from the restaurant again, opened their menu, added it to their favorites or rated it high. From this, we can assume that Alice and Bob have similar tastes (at least in pizza). If that assumption is correct, then Bob might be interested in Alice's favorite sushi place. So let’s recommend it to Bob!

Now there are two different scenarios; either Bob orders from the recommended sushi restaurant and likes it (by rating it high, opening the menu, adding it to his favorites or ordering from it again). From that we can infer that indeed, Bob and Alice are similar in their purchase behavior and we can continue recommending them venues that they both like. 

If Bob doesn’t order from the recommended sushi restaurant or he orders, but does not like it (for example by not continuing to interact with the venue, such as by not ordering from them again). We make the conclusion that Bob is not that similar to Alice. If that is the case, we find a new ‘Alice’ for Bob to make sure we can help Bob find his favorite venues on Wolt.  

Algorithmic Transparency Report 2024_page-0009.jpg
  • jpg

Recommendation in the ‘Stores’ section

Determining the best way to rank the most relevant stores and shops is different compared to restaurants. For instance, predicting the time of day you might order from an electronics or flower shop is more challenging compared to typical meal times for restaurants.

In the Stores section we therefore apply a different recommendation algorithm that prioritizes venues based on their:

  • Popularity (partner sales figures on the Wolt platform)
  • Customer rating
  • Attractiveness (ratio of a partner's visibility to their sales figures). E.g. we support new venues (less than 60 days on Wolt) by boosting their visibility.

These main parameters are considered in equal parts. As with restaurants, we show only the closest franchise partners to your delivery address and prioritize any category lists displayed based on their popularity.

First-Time Users

How does personalisation work then for people who are new users? If they have not registered an account or have a registered account but with no purchases, recommendation will be based on a simple status model we call the ‘First Time User’-model. The model works with aggregated data, so no personalisation is applied, and is used on the Discovery and Restaurants page.

In order to rank venues for first time users, the model factors in: 

  • Venue delivery time
  • Venue popularity (partner sales numbers on Wolt platform)
  • Price level
  • Delivery price
  • Distance
  • Type of business operation model (e.g. brick and mortar vs. virtual)
  • Venue rating
  • The number of ratings and retention.

The importance of these signals is determined through an automated machine learning process. Essentially, the system observes how each signal impacts a customer's choices over time. For example, if many customers prefer items with faster delivery, the influence of the 'delivery time' increases. This learning process is ongoing, adapting to changing trends and preferences on real-world behavior.


Consumers have an option to rate the venue they ordered from after receiving their delivery. This rating system helps us to gain insights into their experience and preferences. We calculate the average rating (on a scale of 1 to 10) by aggregating the ratings provided by all consumers who have ordered from the venue in the past six months. This rating is displayed on both the Wolt App for other consumers and the Merchant App for the merchant to identify any areas that require improvement. The ratings are calculated and updated daily.

  • png

Recommendation in other parts of the experience:

  • In-cart: Once you are viewing your cart and ready to order, we may also recommend additional items that you might be interested in. This is based on item popularity, user purchase history, items in basket, and basket value. For example, if there is an item that a user has purchased frequently in the past 12 months, the system will recommend that item in case it is not already in the cart. How many times have you forgotten to buy milk while going to the grocery store? If it is a frequent purchase, we’ve got your back! 

Search Algorithm

Wolt Search is designed to help you quickly and easily find products and services that are relevant to you. Wolt has two search functions that work differently: The standard search in the Wolt App and on the website, and the partner-specific search for searching a certain item at a specific restaurant or business.

Standard search

Standard search allows you to search the entire Wolt platform for products, services, and partners. When you enter a search term, we match your search with our index of partners (e.g. restaurant name) and product data (e.g. product description). To give you the most relevant search results, we also use synonyms and correct spelling mistakes. 

CRM option 2-SearchBar_Celtra_Image_SoMe-Display.png
  • png

By default, the results are sorted by ‘Recommended’, which is based solely on the:

  • Venue’s opening hours
  • Text-based relevance of the search results (match of the search term with the partner and product data)
  • Popularity of the venue (partner sales figures on the Wolt platform)
  • Distance of the venue to the delivery address.

These parameters, determined by algorithms, weightings, and internal business rules, shape the partner's position in search results. Accordingly, the text-based relevance influences with most weighting the position of the partner in the search results, after all search results have been previously grouped according to the opening hours. The popularity of the partner influences the results with the second importance in weighting and the partner’s distance influences the results as the least important factor in weighting. Partners matching your query but not open or located farther away are displayed at the end of results.

You can also sort the search results manually using different options, such as the delivery price and the partner's customer rating. We also store your search history and suggest searches that you have used in the past. If you want to delete your search history, you can log out of the app, or manually delete previous searches

Partner specific search

With the partner specific search you can search for items that are available at a specific partner. We match your selected search term with the partner's product titles and descriptions. The results of your search are always sorted based on the text-relevance of the search results (match of the search term with the product data).


At Wolt, we are committed to ensuring that our products are accessible and usable for all users, including those with disabilities. We understand that for sighted users, visual cues often suffice to navigate and interact with our content. However, for users with visual impairments or other disabilities, alternative methods are essential for an equivalent experience.

To address this, we have implemented features like 'high contrast mode,' designed specifically for users who need a more distinct contrast between background and foreground colors. This mode can be enabled manually by the user or activated automatically if their operating system is set to prefer high contrast settings. This ensures that when a user opts for higher contrast across the web, will recognize this preference and adjust accordingly.

CRM option 4-3631325321_Wolt_Academy_Assets_Main.png
  • png

Furthermore, we are dedicated to ensuring that both and our iOS and Android applications comply with Web Content Accessibility Guidelines (WCAG) 2.2 standards. To achieve this, we are actively auditing our products to identify and rectify any issues that may hinder full compliance. Adhering to these guidelines serves a dual purpose: it not only guarantees that our content is easily navigable by users but also ensures that our products are developed with semantic integrity. This allows assistive technologies to effectively and clearly communicate all necessary information to users with disabilities.

We have also implemented training programs designed to increase awareness and understanding of accessibility issues among our staff. These programs cover a wide range of topics, from the basics of web accessibility standards, such as WCAG 2.2, to the practical implementation of accessible design and development techniques. Our goal is to ensure that every team member, regardless of their role, understands the importance of accessibility and how it impacts our users.

The information on Wolt’s products and algorithms in this report are based on our operations as of February 2024 in Austria, Azerbaijan, Croatia, Cyprus, Czech Republic, Denmark, Estonia, Finland, Georgia, Greece, Hungary, Iceland, Israel, Japan, Kazakhstan, Latvia, Lithuania, Luxembourg, Malta, Norway, Poland, Serbia, Slovakia, Slovenia, Sweden. For more information, check out our Transparency web page.

About Wolt

About Wolt
Wolt is a Helsinki-based technology company that makes it incredibly easy to discover and get the best restaurants, grocery stores and other local shops delivered to you. To enable this, Wolt develops a wide range of technologies from local logistics to retail software and financial solutions, as well as operates its own grocery stores under the brand Wolt Market.

Wolt was founded in 2014 and joined forces with DoorDash in 2022. DoorDash operates in 31 countries today, 27 of which are with the Wolt product and brand.

Contact details

Related topics

Receive Wolt news on your RSS reader.

Or subscribe through Atom URL manually