How I Learned to Stop Worrying and Love the Search Bar

Leveraging A.I. for Online Search and Discovery

Search experiences suck (probably yours too)

We are stuck with technology when what we really want is just stuff that works.” D. Adams

The current state of onsite search in digital retail is somewhat depressing.

According to Baymard Institute 70% of site search engines are unable to return relevant results and include the top 50 grossing e-commerce sites in the U.S.

Considering that 30% of ecommerce total revenues come through the search bar and that search bar users are 216% more likely to convert, this does sound like ludicrous waste of potential revenues

We touched on how bad search can be in our previous post (the TL;DR version of our critique is just searching for shelves online). While the market of SaaS solutions is becoming pretty crowded, our main contention with current (from-decent-to-nice) solutions is that they got the challenge all wrong: it’s 2017 and search should be an A.I. problem.

In fact, A.I. can give your customers a much more rewarding experience than “just” search: at Tooso, we call it search and discovery.

One search technology to rule them all

When discussing onsite search shortcomings, there are obvious and not-so-obvious use cases that need fixing.

The first that always comes to mind is getting customers that know what they want to buy from you. If I search for ‘elegant high heel shoes’ I should get first shoes with a specific property (high-heeled) appropriate for certain occasions (elegant): this is what you get in fact on the high-end Italian fashion store Liu Jo, but it’s easy to check that things get pretty confusing on many eShops (hint: Liu Jo is using Tooso).

According to our data, up to 70% of searches are something very generic, like ‘shoes’ or ‘jacket’. When a customer lands on your website and types such a generic query, she may not know what to buy yet and/or she may not know what’s in store. In the real world, these two factors are often intertwined: the final buying decision is the result of preferences/desires given the total space of possible choice.

For this second use case, the challenge for the merchant is to help the end user navigate this (possibly gigantic) space without frustrating her (we have all experienced countless times ‘the screw-this-I’ll-just-go-on-Google-or Amazon’ moment).

Is there a tech that solves both problems?

From search to discovery: some options

“To boldly go where no man has gone before.” J. T. Kirk

Let’s put the scenario in context. The following is an imaginary shop where the customer lands and types a simple, generic query, like ‘suits’:

Tooso test interface showing items from Canali digital store.

What are the options to engage with this customer? Let’s start with three potential solutions and why they won’t really work:

  • Old school UX, that is filters, sliders, checkboxes somewhere on the page. PRO: easy to implement and very Nineties: we love the Nineties. CONS: clumsy on mobile (which already accounts for ~20% of total revenues), no adaptive or personalized interface.
  • Chat bots, that is chat-like interfaces overlaying the eShop main window. PRO: a dialogue is the most natural form of shopping experience as it mimics what happens in the real world. CONS: while still clumsy on mobile, the main issue is that “it’s a little too human a little too soon”. A conversational interface encourages distracting questions, and technology is just not there yet to keep you entertained: modeling a full conversation introduces a huge amount of complexity that is probably unnecessary to solve this problem (beside the fact that I personally cannot resist asking ‘is P = NP?’, but that’s because I am a horrible person).
  • Visual search, that is visual interfaces allowing you to upload a picture (or pick a product) and find similar items visually (such as the snap feature in Snap). PRO: intuitive and rewarding when narrowing down items based on style. CONS: it still needs a search tool and possibly other UI elements to effectively guide the user in a catalogue.

Unhappy with the available options on the menu, our customers ultimately turned to the search experts (yes, that’s us) for some new ideas; since there is only one piece of HTML we control — the search bar — we transformed that constraint into a (patent pending!) virtue.


Back to the future (of the search bar)

Prediction is very hard, especially about the future.” Y. Berra

Technologically, we built Tooso “back-end first”: it was always our plan to build a SaaS product and all SaaS products start from APIs. However, philosophically we built Tooso starting from a “front-end dream”, i.e. give people an awesome search experience leveraging the latest A.I. and machine learning tools.

In particular, we always loved the idea of using the search bar as our one-click tool to fulfill our needs; either because end user could express articulated needs (‘elegant high heel shoes’) and expect the search engine to understand them immediately; or because the search engine can anticipate our needs and help us to refine your shopping journey.

For example (using the ‘suits’ scenario introduced above), the bar can help us self-generating search autosuggestions of increasing complexity.

The important thing to understand about suggestions is that to be really effective there’s a lot of intelligence to it. If we are going to suggest the user a specific attribute we might want to ask ourselves a few questions:

  • ‘how many items do we have in the catalog with that attribute?’ (so, we need inventory data and a good product ontology)
  • ‘is this attribute pointing to the product that we want to sell?’ (so, we need business criteria, like marginality, best sellers, ongoing promotions and marketing campaigns etc);
  • and most importantly ‘is this attribute the right one for this user in this moment?’ (so we need data about the user’s browsing behavior, the user’s shopping history etc.).

Good suggestions have to take into considerations a whole bunch of different factors.

Smart suggestions will match the needs of the business and the preferences of the shoppers.

As the user clicks and start her search & discovery journey the search engine can go further and provide simple on-click discovery tags to refine the search query with additional information, say a color or a brand.

Tags help users refining their intent in an easy, adaptive way.

Each one of this click is gold: it tells us more about this person, what she likes, what she has in mind, why is she here and so on.

Once again, all the tags have to be dynamic and intelligent, they have to depend on different factors, or we’ll be back in the cage of filter and categories rigidly decided upfront.

How is that an improvement over other discovery processes?

  • It’s mobile friendly: as most eShops have already chosen a full width search bar in their app, nothing to be changed there!
  • It’s engaging, but focused: no confusing interface or unnecessary hurdle with artificial conversations that go south, as the customer is still fully in the context of the search interface.
  • It’s smart and adaptable: thanks to our superior A.I. technology, we can understand both the user’s intention and the underlying catalogue to always suggest the refinement most likely to drive the final conversion.

Our search and discovery tech is now used by our customers. We A/B tested and we know it works.

  • 15% increase in the conversions after search (people buy more often).
  • 20% increase in the average basket value after search (people rely on the search bar multiple times and buy more stuff).
  • 7% decrease in the post-search exit rate (more happy users, less ‘screw-this-I’ll-just-go-on-Google-or Amazon’).

And wait until we deploy voice versions (featuring support for speech-to-text and text-to-speech technology)!

See you, space cowboys

If you want to join the A.I. revolution in online shopping, don’t forget to follow us on Linkedin, Twitter and Instagram.

We would love to hear your thoughts on search and discovery, so feel free to reach out directly to jacopo.tagliabue@tooso.ai.

Acknowledgments

Many thanks to Maria Paola Sforza Fogliani and Katherine Yoshida for sharing with us their linguistic and fashion wisdom.

Like what you read? Give Jacopo Tagliabue a round of applause.

From a quick cheer to a standing ovation, clap to show how much you enjoyed this story.