People often think of Google and Amazon as setting the gold standard for search features. After all, they have millions of dollars and thousands of engineers to invest in the effort.
However, even the titans of the industry can’t get everything right. We’re obsessed with autocomplete, and we find it useful to study where other sites can improve. Here are a few areas where Amazon’s autocomplete misses the mark.
The first thing you notice when using Amazon’s autocomplete is that it’s not functional until the full page loads. That’s unfortunate — if you know exactly what you’re looking for, you want to start typing it in immediately. I often find myself waiting a few frustrating seconds for Amazon’s results to start appearing. Amazon themselves found that every additional increase in latency caused a decline in revenue, so a few seconds of delay before making autocomplete available has major consequences.
No Product Listings
One of the most useful features of autocomplete is the ability to jump directly to product pages and bypass search results altogether. This can save a lot of time if you’re searching for multiple items in a hurry. Furthermore, a/b tests with our customers have consistently shown significant conversion and revenue benefits to these direct links.
However, Amazon doesn’t show product listings in its autocomplete results, requiring users to navigate through a search result page to find what they’re looking for. This feature is very hard to do efficiently at large catalog sizes, which might be why Amazon isn’t there yet; but the benefits are undeniable.
We spend a lot of time honing our algorithms to automatically correct spelling mistakes that users make while searching. Amazon’s autocomplete, unfortunately, misses a lot of the common cases that we work hard to correct:
Phonetic misspellings: a search for “cindle” on Amazon returns results for “candle” but not “kindle”. You’d think that Amazon would account for this common phonetic transposition for one of their top-selling products. Similarly, a search for “barby clothes” shows suggestions for “baby clothes” instead of “barbie clothes” even though it’s relatively unlikely someone would add an extra “r” while typing “baby.” Along the same vein, a search for “vacum,” the common misspelling of “vacuum,” tries to autocomplete to results like “vacume,” or “vacumn.”
Keyboard proximity typos: typing “lego mindsgorm” on Amazon returns no results, even though “t” and “g” are next to each other on standard keyboards and are common transpositions of each other.
Punctuation nuances: a search for the popular “Canon 80d” camera returns a good assortment of suggestions on Amazon, but “Canon 80-d” doesn’t.
Omitted characters: If you’re hastily looking for a “northfce jacket” and forget the “a” in “northface”, you won’t have any luck.
With all of Amazon’s engineering and processing power, you’d expect them to offer some personalization features for their users. However, even though users often perform the same search repeatedly before purchasing an item, Amazon doesn’t boost a user’s recent searches to the top of their suggestions.
And with the massive amount of data that Amazon has on all its customers, you’d also expect the autocomplete suggestions to be ranked differently for different users. However, they aren’t.
Lack of Diversity
Advanced autocomplete systems try to walk a fine line between showing the most popular items and also showing a good amount of diversity. If a searchterm ends up showing the same general suggestions with only minor differences, it isn’t particularly useful to users.
On Amazon’s site, for instance, if you type “call,” you’ll find that 8 of the 10 suggestions are for “call of duty.” While those might be the most popular searchterms, customers would benefit from seeing other choices if they’re not looking for that particular video game. This becomes very clear when you take a look at the actual search results page for “call,” where results focus on calling and mobile apps, none of which are featured in the autocomplete.
These are just a few examples of where Amazon’s autocomplete falls short of perfection. Even large companies like Amazon can’t devote all the resources needed to create ideal autocomplete behavior.