Hamish Ogilvy, CEO and Co-founder of Search.io – Interview Series

Hamish Ogilvy is the CEO and Co-founder of Search.io, an AI-powered search and product discovery platform that processes billions of searches for some of the largest organizations in the world. They focus on increasing conversion and revenue of on-site search using machine learning. This is achieved using proprietary search technology.

What initially attracted you to the field of machine learning?

I started out in physics designing lasers, but I was also always pretty good at seeing patterns in data. That drew me into analytics and predicting various things where I spent several years. Machine learning was an obvious extension to better facilitate value extraction from data. That seems fairly normal now but at the time people actually laughed at me for having the job title “head of data”, so the world has changed somewhat!

Could you share the genesis story behind Search.io?

Search always annoyed me; it’s improved a lot but it’s still pretty bad for the most part. I always thought it should better understand context and personal intent. So it started with that thought, and years later we’re much closer. Larry Page famously said search is not a problem that will be solved in our lifetime, and I now fully understand why. But we’ve drastically changed the game and are a long way towards what we envisioned.

Search.io uses self-learning search technology based on “neural indexes”, what specifically are neural indexes?

Old school search technology creates indexes on specific keywords, much like the index in the back of a book pointing to the pages where the keyword appears. Neural indexes instead create indexes from specific neurons in specially designed neural networks. These neurons are designed to activate for related context and meaning as opposed to keywords, so it’s possible to very quickly identify related items, even across different languages. Self-learning allows these indexes to adapt and improve as people purchase items or other positive feedback events.

Can you discuss in what ways reinforcement learning is used to optimize search results?

Mainly this is used to reorder the search results to maximize a specific goal, such as clicks or purchases. This is a classic explore vs exploit problem where the best order is initially unknown. There was a famous saying that the best place to hide a dead body was on page 2 of google results! That’s because virtually no one even makes it to the bottom of page 1. So if you have say 10,000 relevant results for a query and only show 10 per page, some probabilistic randomization helps to rotate different ones in and elevate those that are resonating with the business goal.

Is deep learning used, if so in what ways?

Yes, in many ways. We use deep learning to turn language into vectors and then also vectors into neural hashes.

The platform enables businesses to control the built-in learning by enabling tweaks, could you discuss how this process works?

Self-learning typically uses events to “nudge” the neural hashes towards things that have previously resonated, such as purchased items vs non purchased. In some cases that can take too long and the business knows what should be a great result. So we allow them to tell us and this teaches the relevance to better understand intent. This basically fast tracks the learning.

What are some of the challenges behind deciphering user intent when a search is made?

Language is highly ambiguous. A “bank” could be a financial institution, a river bank, a plane turning, a basketball shot, etc. To make it worse, for someone from Los Angeles searching for a “jacket”, would also be a very different intent to someone searching from Boston. Male vs female, even more different. Language poorly encapsulates intent, perfection is actually impossible, so search is a game of probability.

What’s the process for integrating Search.io in a website?

It varies depending on complexity. For shopify stores and content based websites you can be up and running in minutes, whereas advanced ecommerce stores with complex catalogues and highly personalized results can take weeks to several months.

Is there anything else that you would like to share about Search.io?

If you’re a data heavy organization trying to increase online transactions, drop us a line. We’ve consistently massively improved the online performance of our customers via improved search and discovery experiences.

Thank you for for participating in this interview, as someone who has been studying search engines for over two decades this was quite fascinating. Readers who wish to learn more should visit Search.io.