It is intense days at Prisjakt's headquarters as Ehandel.se is on site to meet management and test the new technology ahead of the launch. Just three days after Friday's well-publicized stock listing on Nasdaq First North Premier Growth Market, where the company was valued at 675 million Swedish krona and the share was heavily oversubscribed, the price comparison site is implementing a comprehensive triple investment on Monday.
Today, a completely new, conversation-based AI shopping assistant is being launched to capture consumers significantly earlier in the buying journey. At the same time, the company is rolling out a new visual brand identity and a modernized logo.
Behind the technical shift are, among others, Peter Stark, Chief Technology and Product Officer (CTO and CPO) with seven years at the company and over 20 years of experience in product development from companies such as Sony Ericsson, and Ola Lindberg, Product Owner for the AI initiatives with six years at Prisjakt and a background from companies such as Ebay.
When Ehandel.se sits down with the duo at headquarters, Peter Stark explains why management chose to overhaul both product and brand simultaneously:
We felt that this is such a big change that it wasn't enough to just say that the product had a new feature. We actually needed to change the entire branding of the entire company and all products at the same time to signal that a new Prisjakt is coming out. And then it happened to coincide with the stock market introduction as well, so there was a lot going on at once. But a statement was needed that we are moving somewhere else.
During the work on the new identity, management even discussed whether the company should change its name completely, but concluded that the brand encompasses too much established trust to scrap. In the new commercials now being rolled out, the abbreviation AI is not mentioned in a single word, as the technology itself is not what ordinary consumers are asking for.
From fast tables to marathon training
For the millions of Swedes who have become accustomed to using the site as a final price check, the transformation raises a given question: is this the end of the classic table with the lowest price at the top? The answer from management is a firm no, as the assistant is instead added as an interactive layer on top of the existing service.
Peter Stark points out that the basic function for those who already know what they want remains completely intact:
If you just want to search for Iphone 17, you should get a list of Iphone 17 phones very, very quickly. We are careful to retain what is good and add a new layer to it.
The challenge that Prisjakt has struggled with for many years is that the product catalog now contains over 18 million products and a wealth of advanced filters, price graphs and comparison tables. For those who are not experts in a particular product category, the threshold has sometimes been high, but with the new conversational interface, the customer can start with a need rather than a model name, explains Peter Stark:
Before, it may have been that you wrote that you want a pair of Adidas running shoes, went to that page and clicked your way out. Now you can write that you are going to run the New York Marathon in three months and we start there instead. Then the assistant can help you find shoes among Prisjakt's 18 million products and all prices.
For Product Owner Ola Lindberg, the big lift is rather that the user no longer needs to leave the site to do their research on external forums or video sites:
Previously, you often had to do your research on Youtube or other sites and then come to Prisjakt with ready-made facts just to check the price. Now, the entire investigative part can take place directly in the dialogue. At the same time, we see that it takes time for users to get used to what the technology can actually do. Many are so accustomed to the search box that they just enter a single product name, before realizing that they can have a conversation and ask for guidance based on their needs.
By collecting expert tests, user reviews, price alerts and price history in one and the same dialogue, the duo believe that the path from vague idea to well-thought-out purchase is significantly compressed.
Undisturbed price list and free listing
Among Swedish e-retailers, the transition to AI-based advice has also raised questions about how products are selected and whether paying partner stores will be favored in the recommendations. Here, management states that the basic rules of the game that have applied since the start remain unchanged.
Peter Stark emphasizes that the sorting on product pages cannot be bought:
We do not change when we list prices on a product – it is still lowest to highest of all who sell. That does not change. The foundation and our entire credibility rests on that. We are based on the fact that you should be able to make a good purchase and then there must be complete and correct information.
Even stores that are not paying customers with direct links will, as before, be visible with their prices in the lists. Ola Lindberg believes that full coverage of the market is a prerequisite for consumers to trust the advice:
All stores can publish their range on Prisjakt. It doesn't cost anything. We show who sells it at what price, even if they do not pay a click fee directly. If you remove 40 percent of the products, how will the user know that it is the right product?
For smaller and niche e-retailers, the conversational model can also be an advantage. In traditional searches, the most clicked products have often cemented their places at the top of popularity lists, but when the customer describes a specific use case in the chat, even more specialized stores and brands lower down the lists have a chance to be highlighted.
Sponsored agents and new ad formats
Although the impartial price list is protected, the AI assistant opens up new commercial avenues. The click model (CPC) to the stores remains the backbone of the business, but Prisjakt is now drawing inspiration from how global giants like Amazon integrate advertising into their AI tools.
Peter Stark describes how the advertising business is expected to take shape in the new environment:
Overall, it is still advertising we work with and clicks to stores remain. But with the shopping assistant, completely new opportunities open up to expose brands and stores. We look a lot at what Amazon does with sponsored agents. Hopefully, the ads will be much more relevant to the user and integrated into the environment. Every pixel should be relevant.
Ola Lindberg envisions that brands and manufacturers, who have deep detailed knowledge of their own products, should be able to contribute their voice as a clearly declared complement in the conversation without infringing on the independent price comparison.
Own data as a backbone
Technically, Prisjakt has been working with machine learning to categorize products and prices since the late 2010s. However, the actual construction of the new shopping assistant did not really take off until the end of January this year, where the development team also used AI to write large parts of the code.
Management quickly decided not to try to build their own general language models from scratch. Instead, Google Gemini and smaller special models are combined with Prisjakt's own infrastructure and database with hundreds of millions of price points that are updated several times a day, explains Ola Lindberg:
The basic model almost becomes a commodity. We use Gemini for some parts, but we have built it so flexibly that we can replace components. The model itself is not the value – the value is the product you create around it with our data and user experience.
Peter Stark points out that it is precisely the control over real-time data that distinguishes a specialized shopping assistant from general chat services like Chat GPT:
Actually, it is only the intelligence and the ability to reason and generate text that we take from Gemini. All the data around it comes from us. We need real-time data and control our data, unlike general AI services that often sit on cached data.
Releasing a conversational engine with image recognition has also required extensive work with security barriers, so-called guardrails. During internal testing, the filters were set so narrowly that a developer who photographed his lunch and asked what kitchen utensils were needed to cook it was blocked by the system, before the balance was adjusted.
However, the boundary towards topics that have no connection to trade remains sharp, says Peter Stark:
We do not want it to start talking about politics or Trump, but stick strictly to shopping and products. But it is a difficult balancing act where the boundary should be drawn when the conversation starts in something everyday that then leads on to a purchase.
That is why the company has chosen to launch the service broadly now to be able to fine-tune the boundaries as consumers begin to interact with the tool in a live environment.
View of the new challengers and the future purchase button
The major investment is rolled out at a time when competition in the Swedish market has escalated. During the autumn, Ehandel.se has reported on how new challengers such as Fyndvi, Dubbelkolla and Prisgeek have entered to challenge with alternative business models without click fees.
At Prisjakt, they receive the new competition with calm and point out that the real threat would rather be if all product searches were swallowed by a single global tech giant, says Peter Stark:
First and foremost, we see it as a receipt and a validation that people actually want price comparison services. The biggest threat to us would be if the open web disappeared and there was only Google left. The more competition there is and the more people know that you can go outside of Google to look for products, the better it is for us.
Ola Lindberg adds that modern AI tools certainly make it much easier than a few years ago to quickly build the first 50 to 60 percent of a price comparison site. The big threshold, according to him, lies in the second half of the journey: to manage, clean and structure dirty product data in real time over tens of millions of articles where retailers constantly change product titles for search engine reasons.
When looking towards the coming years, one of the hottest industry questions is whether consumers will soon be able to complete the entire purchase directly within the assistant without leaving the site. Prisjakt is following the development of new standards such as UPC and UGP, but emphasizes that online stores should always be the sellers, explains Peter Stark:
The technology where you can actually have a purchase button on Prisjakt, but where the store is still the merchant of record and owns the purchase and customer relationship, is very interesting to us. As soon as that technology is available and mature, we want to look at it. We respect that stores want to manage their customer relationship.
To achieve the goal of being the obvious shopping assistant in the Nordic countries within two years, however, requires more than just new technology, it is about breaking a deeply rooted human behavior. Ola Lindberg states that consumers are now facing the biggest change in how they look for goods in two decades:
We are damaged by Google. For 20 years, you have learned to write keywords in a certain way, and actually almost nothing has happened within search during that time. Now it's about relearning and just writing what you actually want to do.
For Peter Stark, who has had the opportunity to experience a number of technological paradigm shifts during his career, Monday's launch in combination with the stock listing is one of the most eventful milestones so far:
I have a few years left until retirement, but to be allowed to experience such a major technological shift is incredibly fun. I remember when the internet came, when mobile phones came and when Google and Facebook were brand new. To now be able to reshape the entire company, turn the brand and go public at the same time is courageous of us, he concludes.