AI

How AI search enables ecommerce companies to boost revenue and cut costs
facebooklinkedintwittermail

Consulting powerhouse McKinsey is bullish on AI. Their forecasting estimates that AI could add around 16 percent to global GDP and labor markets by 2030, or about $13 trillion. With inflation and recession concerns still high, many e-commerce companies are looking for ways to increase profitability while also doing more with less. AI-powered search has the potential to help companies do both. In this blog, I’ll explain how.

The search imperative

On-site search can have a big revenue impact for both B2C and B2B companies. Studies show that consumers on B2C retail commerce sites frequently start their journey with search and spend 2.6 times more compared with non-searchers. For B2B commerce, an enormous 92% of purchases start with search!

Today, ecommerce site managers must spend considerable time optimizing search results for their buyers. That’s because search bars are most often powered by keyword search engines. As the name suggests, keyword search is based on matching the search query with keywords in the search index. Keyword engines work very well for “head” queries where the search and index terms match, but any variations in terms can cause the engine to fail.

long tail volume
“Fat head” queries represent the top 10-20% of searches, and “long tail” queries typically include the bottom 50-70% of site searches. Keyword search works well for the fat head, but long tail is better served with AI or in combination with keyword search, sometimes called hybrid search.

The problem is you can’t write enough rules, synonyms, or keywords to account for every possible query variation for every product in your catalog, especially if your catalog includes thousands or even millions of items items, as is often the case in B2B commerce. You might be thinking that what I’m describing sounds a lot like a long tail search problem — and you’d be right! Searchers arrive at your site with something in mind. However, the way they formulate a query could be quite different from the way your search index is set up.

Unlike keyword search engines, AI-powered search actually understands the meaning behind a query. The difference between keyword matching and AI understanding is immense. It’s much more human-like.

Here’s the important part: AI search works well for catalogs without additional effort for query matching. AI search handles those long tail queries for you automatically. No need to build a synonym library, write rules, or stuff your pages with keywords while hoping to get a keyword match; with machine learning for search retrieval, it works out of the box while saving you time and money.

How AI works

Artificial intelligence is the fancy term; most technologists prefer to speak about the underlying machine learning algorithms that can power a wide range of technology, from inventory prediction models to self-driving cars. ChatGPT is known for its use of large-language models (LLMs), but there are many other technologies or algorithms for machine learning. Every AI system is a multi-step process of transforming information into machine-readable data for faster processing, and then applying different models to interpret the data and deliver predictions. Models can be tweaked to improve performance and confidence levels.

The data in your search index plus the information you’re collecting from your customers — geo, past purchases, past searches, browser, member status, ratings, returns, etc. — is fuel for the fire. Some machine learning technologies called learning-to-rank algorithms will automatically improve results over time by learning what works and what doesn’t. Other AI solutions can be used to power recommendations such as customers who bought x also bought y which can improve customer satisfaction and also generate a higher average order value.

End-to-end AI search

There are many different AI technologies in play, but the one that has the biggest impact on search relevance is called vector embeddings. Vectors enable the search engine to understand similarity between terms. For Algolia, vectors are just the first step. We apply another technique called neural hashing to compress vectors, which enables our customers to scale search and make frequent changes to their catalogs and schema without negatively impacting results. It also allows us to combine AI with keyword results at the same time for even greater relevance. The result is radically faster, more accurate, and scalable AI-powered search retrieval.

NLP example
How natural language processing (NLP) parses and analyzes queries to help the search engine be more successful.

The entire process — from query to result — is a multi-step sequence:

  • Query processing: First, we apply a machine learning functionality, natural language processing (NLP), to pre-process the query (see diagram above).
  • Retrieval: Then, machine learning powered by neural hashes identifies the most relevant results, and orders them from most to least relevant.
  • Ranking: Finally, dynamic ranking is applied to optimally order results to improve site performance — clicks, conversion, purchases, etc.

All of this happens in near real-time (whether you have autocomplete configured or not). In fact, most queries are returned in less than 20 milliseconds, which is 5x faster than the blink of an eye. It has to be fast, too; Amazon and Google both showed the importance of speed for customer experience and revenue.

One of our early Algolia NeuralSearch™ alpha customers — an ecommerce site with about 3 million SKUs — saw an immediate 6% conversion rate increase within the first month of adoption. They didn’t need to do anything either. AI automation all-but eliminates the manual work and business processes once required for search result optimization. It was just the tip of the iceberg; we expect that number to climb even higher as we ready the product for general release.

neuralsearch demo banner

The smarter search revolution

Companies like Google and Amazon have spent billions of dollars and tens of thousands of hours building their AI-based search, recommendation, and personalization solutions. Now, with advancements in AI algorithms, implementing smart search can be done in a fraction of the time and cost.

We’re excited about the impact that AI-powered search is going to have on B2B and B2C sellers as they push for digital transformation, cost reductions, and faster automated decision making. Companies can reallocate the endless hours they spend optimizing search, while at the same time improving results to drive higher conversions and a better user experience.

For more information, take a look at our very own AI-powered search solution — Algolia NeuralSearch — or contact our team of experts.

About the authorMichelle Adams

Michelle Adams

Chief Revenue Officer at Algolia

Recommended Articles

Powered by Algolia AI Recommendations

What is AI-powered site search?
AI

What is AI-powered site search?

John Stewart

John Stewart

VP, Corporate Communications and Brand
What is end-to-end AI search?
AI

What is end-to-end AI search?

Abhijit Mehta

Abhijit Mehta

Director of Product Management
A simple guide to AI search
AI

A simple guide to AI search

Jon Silvers

Jon Silvers

Director, Digital Marketing