View from Initialized Is Building Amazon Quality Recommendations For the Rest of E-Commerce

Amid the pandemic, e-commerce now makes up almost one-sixth of U.S. retail sales and platforms like Shopify are fueling a Cambrian explosion of stores. co-founder Oliver Edholm began working as a professional AI researcher at the age of 15.

To compete with the e-commerce giants, these stores need their product recommendation engines to be just as good. The machine learning experts at are addressing exactly this need, and I am quite proud to announce that Initialized led their seed funding round with participation from Y Combinator, EQT Ventures, Liquid 2 Ventures, and Northzone.

Recommendations are a key contributor to e-commerce store success. They help the buyer find better results while browsing and enable more sales. Machine learning is a critical part of modern recommendation engines. They can not only provide far more recommendations that humans can produce, they also provide recommendations that anticipate buyers’ wants better.

But not all machine learning models are equally good, and the machine learning world has trended toward larger and larger models with signs that enormous generally trained models can be more successful than smaller models trained on less (or more domain specific) data. This trend is perhaps most notably demonstrated lately by the number of surprising results built on top of GPT-3.

It is no surprise then that the largest e-commerce stores like Amazon generally have the best recommendations given that they have the biggest data moats. They have huge inventories, purchases histories, and accumulated results of previous recommendations which they use to continuously generate better and better recommendation engines.

Smaller e-commerce stores depend just as much as Amazon on increased revenue through recommendations but the size of these stores naturally constrains the amount of data they can feed into a model. Worse, this low data environment requires data that is extremely specific to the individual store, forcing store owners to do deep integrations with recommendation providers, handing over sensitive sales information.

Image for post co-founder Anton Osika

This is where has taken a different approach. They create one huge model for all their customers. They can scan an e-commerce store, figure out what inventory it has, and infer the best recommendations for all the products without any internal sales or customer data. It’s dead simple for the store owner to include on their site via a few lines of code and without waiting for additional training sensitive internal store information.

It’s a lot of hard technical work to build such a model, but if we’ve ever met anyone up to the task it would be Oliver Edholm and Anton Osika, founders of Oliver is only 18 but has been a professional AI researcher since he was 15 and has deep experience in cutting edge machine learning models.

If you have an e-commerce store definitely give a shot and see how much better your recommendations can be.