Globally, we send daily 161 billion non-spam emails. While spam has been an issue for which engineers have developed solutions already for decades, writing these emails also represents a massive burden. With an average email length of 240 words per message and average 40 words per minute, typing the average email takes approximately 6 minutes. With 162 billion emails, writing these would take 16.2 billion hours to type, or 1.8M years of constant typing, every day. These statistics are probably a bit exaggerated (there is a lot of templated emails and so on), but you get the point: email writing is a huge time sink, and anything speeding it up has huge productivity potential for the world.
As writing electronically has increasingly become a routine, tools to speed up the process of typing have also become more common. The latest technology is already able to complete longer sentences for us. In this post, we compare our solution TypeGenie to the world’s most famous sentence completion feature: Gmail’s SmartCompose. While both tools aim at improving typing speed, the use cases differ slightly: TypeGenie completes sentences for customer service agents on Zendesk, while SmartCompose works on Gmail for generic business emails.
We choose to use two metrics for comparing the utility of these systems: the number of times that a user accepts a completion, and the number of typical words used in an accepted completion. We do this comparison in the domain of customer service, writing 5 very standard customer service tickets:
To better simulate a customer service agent’s work, we type freely answers to the tickets without being constrained to fixed word choices. As a comparison with an incoming ticket context would be technically a bit harder, we provide both systems with no context of the customer query.
We are pleased to see that both systems are able to help us, providing completions as follows.
Looking at the statistics in more detail, we can also see that TypeGenie provides more useful completions in all tickets except for the last one. The typical completion length is also longer.
The results clearly show TypeGenie as perhaps 4X more useful than SmartCompose for the use case of resolving customer service tickets. Naturally, this magnitude of improvement can make a massive difference in the productivity of a person typing.
On the other hand, the results are hardly surprising to a technical expert: while SmartCompose for Gmail uses a generic language model that uses only light customization per user, TypeGenie for Zendesk is not only personalized for customer service, but also for each individual company and each customer service agent.
To conclude: both SmartCompose and TypeGenie are helpful tools for someone writing numerous emails. For someone in customer service, TypeGenie can achieve significantly higher productivity gains as it adapts far more to the relevant company’s and user’s writing style.