Every year, almost 76% of consumers abandon a site after not finding what they are looking for, costing ecommerce companies over $300 billion.
Instead of driving customers away from your site, Natural Language Processing (NLP) in ecommerce can drive conversions and revenue instead. Successfully connecting people to the products they’re looking for is the heart of great ecommerce search.
Here are six ways an NLP ecommerce search engine can drive your KPIs and boost your revenue:
Table of Contents
1. Better Autocorrect
Autocorrect refers to systems that understand and correct common typing errors. Poor autocorrect leads to lost revenue as customers don’t get the results they intended, even if the product they’re looking for actually exists.
This is a common problem with older methods, which are slow and require a lot of computing power. Multiplied across millions of ecommerce customers, this extra time translates into lost revenue as customers become frustrated with the experience.
A modern NLP ecommerce search engine draws on first-party data to automatically understand these differences phonetically and typographically. It understands people’s common misspellings based on their distance on a keyboard and their varied pronunciations based on how they speak, even for unfamiliar terms.
It also provides features like autosuggest (type ahead) and visual autocomplete on top of autocorrect without manual involvement, increasing search success by helping customers find what they’re looking for faster.
Taken together, even an egregious misspelling of chocolate like “vhocolste” leads to chocolate products being offered along with visual autocomplete.
2. Better Synonym Detection
Say an American company sells beanies in Canada. How does it know that beanies are called “toques” in the Great White North? The answer is robust recognition and substitution of synonyms, which is another advantage of artificial intelligence in ecommerce search. Typically, only companies like Google and Amazon have successfully tackled synonym generation at scale. Others claim to bring you synonyms, but only fix spelling and punctuation, e.g. “ipod” and “I pod.”
Up to 70% of search engines are unable to return relevant results for product type synonyms like “rain boots” for “wellies”. So if you have rain boots and a customer searches for “wellies,” they are likely to get zero search results and leave. This is known as shadow churn, and it accounts for approximately half of all new customers leaving a site.
To get beyond this issue, an NLP ecommerce solution processes the relationships between terms in a way that is not merely keyword-based. It understands that when people search for “toques,” they may also be searching for beanies, as people think of “beanies” and “toques” as the same thing.
An NLP-based technology like Cognitive Embeddings Search (CES) learns from categories, product names, and text descriptions to solve the issue of synonyms and zero-result searches.
By combining this data with deep learning, this technology can represent the product catalog as a “sea of stars” chart. Each catalog item is represented as a “star” on the same chart. From there, the algorithm measures the distance to each of its closest neighbors and makes an inference about what the user actually intended. The result is fewer frustrated searches and irrelevant suggestions.
So whether your customer enters “beanie” or “toque” or “skullcap” or “knit cap,” they will get the same (or similar) hats.
3. Reducing Zero Results and Frustrated Searches
A customer searching for “jmpsut” on a clothing retailer’s website expects jumpsuits and not an empty page. Unfortunately, many customers do get these zero-result pages. They also get irrelevant results, which leads them to abandoning the search.
CES reduces zero results and frustrated searches, leading to more conversions and revenue for ecommerce retailers. It solves a number of persistent issues that users commonly have with search engines:
- Sometimes customers are unsure what to call a particular product in a new space. For instance, if they’re looking for epaulets on a jacket, they may type “jacket with shoulder flaps.” In such cases, an NLP-driven engine increases revenue by surfacing suggestions that customers are more likely to buy.
- When price shopping between different retailers, customers copy and paste long product names. Older search engines offer zero results because of their keyword-driven algorithms. But an NLP-driven engine can cut out the unimportant words and act on the ones that give material results.
So if a customer finds “ALMO Men’s Regular Fit Organic Cotton Melange T-Shirts” on Amazon and copy-pastes that product name on your website, you should return a number of organic cotton t-shirts. This leads to more successful searches that turn into revenue for your business.
Most search engines can’t associate generic searches with specific suggestions. If a customer in Minnesota types “not cold,” the engine should be able to associate that with “warmth” or “winter,” surfacing specific suggestions that the user either accepts or uses as a jumping-off point for more refined searches.
So a search like “I don’t want to be cold,” should provide something as specific as Castelli Warming Embro Cream.
4. Revenue from Long-Tail Queries
Long-tail queries are highly specific strings of more than 3 to 4 words entered by a user. “Long-tail” here refers to the large number of low-volume, high-keyword searches represented in the graph below.
While these queries are each searched a handful of times per year, they make up as many as 70% of searches on the web. Because they’re often more specific and reveal a higher level of customer knowledge and intent, they’re more likely to lead to a sale with the right result.
It doesn’t make sense to merchandise for these terms individually. But these users are a huge and untapped source of revenue best served with natural language processing
With an NLP-based technology like CES, addressing these rare search queries that add value over time becomes possible. By intelligently mapping even unlikely search terms to products despite no keyword matches, CES converts half-chances into real revenue at scale.
For instance, for a string like “no sugar added freeze-dried healthy foods,” CES can provide “organic dried banana chips.”
5. Voice Recognition
“Hi Alexa, buy me blue All-Stars, size 8.” You have probably said or heard a phrase like this in your house or in the house of someone you know.
A recent report shows that the value of ecommerce transactions via voice search will grow to $19.4 billion by 2023. Voice assistants like Siri and Alexa have already been around for a while, and people are using them more than ever before. However, people input information into a device a lot differently when they speak than when they type.
They communicate a lot more colloquially. A text search query like “red lipstick” may well be phrased, “Hey Siri, find me a red lipstick between 50 and 70 bucks” when voice searched. Most voice searches are not trained on your catalog and therefore miss even simple searches.
An NLP ecommerce engine that automatically maps voice phrases, subsequently sorts product order by user intent, and automatically applies filters can prepare you for a conversational search future.
6. Multilingual Support
With more shoppers than ever heading online, retailers have an exciting opportunity to reach more global consumers. Today, nearly half of the world’s top ecommerce sites greet customers in four or more languages—and some of the world’s largest companies like Apple generate more than 50% of their revenue from cross-border transactions.
While translation is important, communication is a two-way street. Does your search platform understand natural language queries in every language you otherwise support? NLP is an ideal solution in these use cases as it can “learn” any language.
Additional techniques like custom tokenization can specify how the NLP should break each language down into discrete units. In most Western languages, we break language units down into words separated by spaces. But in Chinese, Japanese, and Korean languages, spaces aren’t used to divide words or concepts. Custom tokenization helps identify and process the idiosyncrasies of each language so that the NLP can understand multilingual queries better.
| English | Japanese |
|---|---|
|
Red Jimmy Choo shoes in size 8 |
サイズ8の赤いジミーチュウの靴 |
Promises Kept
The ecommerce industry shows no signs of slowing down. Despite the pandemic, analysts project steady growth to $7.4 trillion worldwide by 2025. And with an NLP ecommerce solution like Krestor, your company will be well prepared for this future.
Of course, great search isn’t the only way ecommerce NLP boosts your bottom line. Its interactions with other technologies and its insights save time and provide escalating benefits to your company, thus driving your KPIs and increasing your revenue.
No abandoned searches. No broken promises. Nothing but net.
This blog post was adapted from the ebook Walking in Your Customers’ Jimmy Choos: Optimizing Ecommerce KPIs with Natural Language Processing:
Speak your customers’ language. Improve your ecommerce search KPIs.
Customers get frustrated when they can’t find what they’re looking for on your site on the first try. Natural Language Processing (NLP) and AI can help.
Get our free ebook and learn exactly what NLP is, how it works, and the key benefits you can expect when you implement it in your search strategy: higher conversion rate, higher revenue, and better customer experiences.
