Amazon Alexa Now Has 10k Skills, Including Europe

Just one week ago, Voicebot was the first to report that Amazon had passed the nine thousand skill mark. However, we noted an interesting anomaly. The trend had been over 200 new skills introduced per week. In fact, Alexa skills had grown at a rate of more than 400 skills per week in January. We documented that trend in as rise from seven to eight thousand skills took just 17 days. Then the progress slowed and it took 28 days for the next thousand skill milestone. The pace really started to slow down when the skill count passed around 8,850.

We offered several hypotheses that could explain this sharp reduction in new skill introduction. Then, Amazon suddenly announced 10,000 Alexa skills when there were fewer than 9,300 available in the U.S. skill store on February 23rd. What was going on? New counts from the UK and Germany were added.

The Increase in UK and German Skills

There was a reorientation in the ten thousand skill announcement that may have gone unnoticed by many people. The last time there was an official mention of Alexa skill count was on the first day of CES 2017.  Mike George from Amazon acknowledged that Voicebot’s reporting from the previous day was correct and there were 7,000 Alexa skills. However, on that day and in previous announcements Amazon had only been commenting on the skills available to U.S. users. That changed with the most recent announcement. The phrase was, “more than 10,000 skills worldwide.”

Last fall, Amazon started shipping Echo devices to the UK and Germany. Alexa has a unique natural language processing (NLP) voice model for British English and German. The idea that skills developed in German are not available in the U.S. is not that unsurprising given the small number of German speakers in the U.S. However, it was a little more surprising that skills available in the U.S. are not automatically available in the UK and vice versa. This has to do with the voice models optimized for both regions that ensure a good user experience. It is also likely influenced by the fact that many skills are country or region specific.

“Worldwide” Counts Shift Not Reported

The “worldwide” moniker didn’t show up in the official Amazon blog or the Wired article that was the first to report the new milestone. However, the term was included in an email to the Amazon Alexa developer community. The rapid rise of skills in one week was a function of changing what was counted. There is nothing nefarious here. Earlier reports of 7,000 skills obviously did not include UK and German skills not available in the U.S. The more recent announcement supposedly counted all unique skills worldwide. I say unique, because the total counts in the US, UK and Germany exceed 14,000 today. It is fair to assume that duplicates are being removed from the counts to provide the 10,000 count last week.

4188 Skills in the UK, But Just 548 in Germany

The more interesting story is that skill count in the UK is over four thousand, but only around five hundred in Germany. At 4,188 skills, UK users have a lot to choose from. The 548 in Germany still includes a good deal of variety but it has a long way to go. There are already complaints about this in Stackexchange and on Reddit.

A European Commission Report from 2012 found that 56% of Germans has conversational English proficiency. So, why not let the Germans use the thousands of English-language skills to supplement the 548 in their native language? It is probably partly a function of the voice model and its ability to function reliably with accents of non-native speakers. But, then again every accent imaginable is now resident in the U.S. The voice model is dealing with that today. Amazon should open up all skills to everyone, regardless of the dominant language in the place they happen to live now. If they don’t, this will be an opening for Google and others to gain ground on Amazon in non-native-English-speaking markets. Voicebot will be looking at this further in the coming months.

Editor’s Note: Thank you to Voicebot readers Rob O’Connor and Max Koziolek for their contributions to the research behind this analysis.