Using machine learning to answer emails.
As you might already know, I absolutely love emails. When people Facebook message me about my website or about my conventions, I tell them to email. Why? Well because I need a written record in order to remember things!
Not everyone loves email as much as I do. In fact, there are some days I question my own love of email. Why would I love getting 100s of messages a day when each one comes with more work? I guess I just love being busy!
Email has now become the most-utilized communication method for business on earth — email is the reason for thousands of dollars of economic opportunity. Businesses of all sizes thrive on email.
Think about it — even large businesses that you will never interact with via email have email exchanges inside their office. While you’re not likely to email someone who works at LinkedIn, the employees are certainly emailing each other!
Many people spend a ton of energy on their emails every day, spending a huge chunk of their life examining, noting, and sifting through their emails. In fact it is often a huge cause for stress for some people.
Businesses thrive on email.
- Ryan Kopf
According to McKinsey Global Institute, over 28 percent of essential business is done entirely by email, so answering email is essential. Email supports customers, structures business deals, and more. That all sounds like too much for a machine, right?
There is a term called ‘email overload,’ where you have a lot of information rambling into your inbox and you are fighting to remain caught up on all the messages. Important messages get buried. Non-priority emails get answered first.
Maybe AI or technology should be able to help with that, I thought. And of course it can, that’s why I made some software for email automation. OwlReply is something I built for my own conventions, and now others can use.
One researcher, Ahmed Hassan Awadallah, said “PC based insight and AI are moving business and people in a whole new direction”.
And today AI can be used for email. Today specialists have mined email data to make a dataset of key email factors, for example, the message length, the number of unanswered messages in an inbox, and whether a message was human or machine-made. This dataset helped setup an email learning model for answering emails.
The model can improve the ability of a computer to understand an email’s inner contents. For instance, email clients could (and now often do, in the case of Gmail) utilize such a model to remind clients about messages they have given up or even ignored, sparing them the exertion they would have spent looking for those messages and lessening the probability of missing important emails. In fact, Gmail is often doing this today with “bumps” of emails reminding you to reply.
Prediction Systems utilize a heuristics-based point of view with an algorithm that picks top words of reference, and gives them measured weight.
What’s that mean in plain English?
Important words are prioritized, and those emails are put on top of your inbox. That’s one way AI is already helping people answer email.
What OwlReply does is not quite “AI” but it is machine technology. OwlReply helps to create a small predictive model of not just keywords you have received before, but of which keywords you are likely to receive again. If people email your business about weddings during a certain season, you’re more likely to be suggested auto-responses for weddings. Same goes for any business.
Of course, the AI isn’t perfect and is still being worked on.
Our plan, one day, is to advance AI and machine learning so that it can help answer every email, but we’re currently focused on reducing the length of time taken in the preprocessing step. Pre-processing is currently time consuming and machine-expensive, and getting this to a more efficient state is key to answering emails before humans get tired of waiting and toss their email-ridden computers in the trash and go to the beach instead….
I might just see you on the beach instead ;)
Scholarly Citations:
Microsoft Research Blog about using machine learning to answer emails.
https://www.microsoft.com/en-us/research/blog/email-overload-using-machine-learning-to-manage-messages-commitments/
Wired Research about How Google answer your messages.
https://www.wired.com/2016/03/google-inbox-auto-answers-emails/