Questa estensione usa un’IA per aiutarvi a scrivere email migliori

Cosimo Alfredo Pina -

Ci sono diverse estensioni che permettono di migliorare l’esperienza di Gmail; tra le più interessanti c’è Boomerang che con cui potete pianificare l’invio di una mail e impostare una notifica in caso questa non venga letta o di mancata risposta. Una funzione semplice ma comoda, che da oggi viene affiancata da Respondable una feature molto più avanza, basata su un’intelligenza artificiale.

Sfruttando i concetti di machine learning, Respondable di Boomerang analizza il testo della mail (per ora solo in inglese) e assegna dei punteggi che indicano quale sarà la probabilità di ricevere una risposta. Gli aspetti valutati sono diversi, tra cui alcuni molto semplici come numero di parole e lunghezza dell’oggetto.

LEGGI ANCHE: Google sta insegnando alla sua IA a leggervi le immagini

La potenza dell’IA di questa estensione si esprime però con altri tipi di indicatori, come quelli che ad esempio indicano la leggibilità, il grado di cortesia e soggettività. Parametri difficilmente valutabili da chi scrive una mail e che Respondable promette di analizzare senza intaccare la creatività.

All’atto pratico però l’estensione dà anche consigli su come scrivere, indirizzando per esempio ad uno stile di scrittura più semplice e leggibile. Un concetto davvero interessante che ha richiesto 6 anni di sviluppo per essere reso fruibile. Peccato che, come già detto, per il momento la funzione Respondable di Boomerang sia disponibile soltanto per la lingua inglese.

Se però scrivete spesso email in questa lingua con Gmail o volete seguire il progetto, nel caso arrivi anche il supporto all’italiano, potete già scaricare Boomerang, compatibile con Firefox, Chrome, Safari, Opera e Outlook, direttamente dal sito ufficiale.


We write emails for all sorts of reasons. For some of those emails, like requesting information or asking a coworker to take action, it’s important to write messages that are easy to read, process, and respond to. Until now, there’s never been a way to know if your messages are optimized to get a response before sending them.

Today, there is! We are delighted to announce Boomerang Respondable – the first AI assistant that helps you craft perfect emails.

Boomerang Respondable uses machine learning algorithms trained on hundreds of millions of messages to isolate what factors impact response rates. It understands how different aspects of your writing combine to affect the quality of the emails you send.

Respondable brings that intelligence directly into your compose window to help you write more effective emails. Respondable gives you unique insight into your writing style and how it impacts the likelihood that the messages you send will receive a response. And it’s built right into Boomerang. Just reload your Gmail window, and the new features will appear. Or, if you need to reinstall Boomerang, click the button below:
Get Boomerang for Gmail
After reloading Gmail, Boomerang will add a new bar into your compose window. The display updates in real-time to show you the quality of your message as you write.

Share on Facebook Share on Twitter
Click the Respondable button to see insights into how Respondable’s algorithms analyzed your message. You can see how each aspect of your writing impacted the quality estimate and see where you might want to make changes.
Boomerang Respondable in Action
Share on Facebook Share on Twitter
For more details about each parameter, just click the graph associated with it. Boomerang provides a description of each measurement and an overview of the findings from our research. Each detailed view provides actionable advice, based on your individual message, about how to adjust the writing to improve the likelihood that your message will get a response.

Share on Facebook Share on Twitter
We’re incredibly excited about these new features! There are two major reasons why.

First, a lot of recent news coverage about artificial intelligence has focused on the possibilities that machines can someday replace humans. Respondable doesn’t try to write important emails on your behalf. Instead, it helps you unlock your creativity. It’s an editor that works alongside you – and one that you can and should ignore when your judgment says otherwise.

Second, we’re excited about the prospect of using machine learning to provide actionable writing advice based on data. Historically, authors of writing style guides have not had access to large amounts of data or had the tools to process it. Our company library includes a dog-eared copy of Strunk and White, and we don’t see ourselves throwing that out. Still, this is the first time in history that anyone has ever been able to get an unbiased, data-driven, real-world view into what kind of writing is *actually* effective. And it’s the first time in history that there has been a way to bring that analysis to bear on your own writing in an accessible way. We hope you’ll find it illuminating – as we’ve been testing it, we certainly have.

For six years, Boomerang has helped millions of customers get more responses to the messages they send. Boomerang can help you send emails at the optimal time, include a read receipt that boosts response rates by 25%, and remind you to follow up. The whole Boomerang team is grateful for your support. Thanks for helping us keep the lights on so we could take that one step further with Respondable.

Please take it for a spin, and let us know what you think. Or, for answers to a bunch of possible questions, read on!

Productively yours,

– Alex and the Boomerang Team

How do I get Respondable?
Respondable is included as part of Boomerang. It is immediately available for all Boomerang users – just reload your Gmail page, and the new features will appear. If you do not see Respondable, try restarting your browser. New users will receive Respondable immediately upon installing Boomerang.

Respondable is also available as part of Boomerang for Outlook – if you want to use it in Outlook, visit to install.

Can I turn Respondable off?
Yes. If you’d rather not see Respondable, just click the Boomerang button in the top right corner, choose Settings, and then uncheck the box labeled “Enable Respondable”

Is Boomerang reading my emails?
No. By default, Boomerang Respondable works entirely within your browser. It does not transmit any message data anywhere, for any purpose. If you choose to enable advanced machine learning features, Boomerang transmits message data using enterprise-grade encryption, analyzes the message data on a secure server, then transmits the analysis back via encrypted communication channels. After performing the analysis, Boomerang’s servers immediately discard any personally-identifiable information connected with the analysis. Boomerang employees do not have access to any message data transmitted as part of your use of Respondable.

Do I have to pay?
Access to Respondable for unlimited messages is included in the free Basic plan. Advanced features, including positivity, politeness, and subjectivity are available for users on Boomerang for Gmail’s Pro and Premium plans. Because these advanced features require several layers of machine learning, all calculated in real-time, they are expensive for us to provide. You can find out which plan you are currently subscribed to by visiting the Boomerang manage page. To upgrade to a Pro or Premium plan, click here.

Machine learning sounds complicated. Does Respondable slow Gmail down?
Thanks to new technology in HTML5, we were able to build Respondable in a way that does not slow Gmail down. All of the calculations for the basic version of Respondable are performed in a background worker thread that does not interfere with the Gmail UI. Calculations related to advanced features are performed in the cloud via a proprietary architecture we developed to allow us to deploy machine learning that operates in real-time. It also doesn’t slow down the Gmail UI.

If you’re interested in the technology behind Respondable, we’ll be doing a technical talk about its architecture at a later date, and we’ll also be writing a super geeky blog post about how we built it – stay tuned! If you’re really interested in the technology behind Respondable, we’re looking for engineers!

I added a rude word, but it said my politeness increased. Or I added a negative word, but Respondable’s positivity score increased. What gives?
You’ve discovered one of the limitations of artificial intelligence in its current form. In any individual email, the machine learning can arrive at answers that do not make sense. In those cases, trust your judgment rather than the calculation.

Politeness, positivity, and subjectivity rely on machine learning techniques that compare your writing to other samples of text that were used as training data. Because there’s no “objective” set of data that can translate how positive, polite, or subjective a set of words is, the machine learning techniques look for similarities between your writing and writing samples that come with an approximate numerical score for positivity, politeness, or subjectivity. One common source of positivity training data is actually movie reviews – words that appear in reviews for poorly-rated movies tend to be more negative!

As a result, if a bunch of movies about asteroids all got panned by movie critics, and words about asteroids were not present in other training data, it’s possible that the neural networks could conclude that those words add negativity to your writing. There are even some names (there is a very polite Tom somewhere!) that can mistakenly affect the measurements.

As such, you should treat the calculations as a general guidepost rather than a source of absolute truth. They will usually make sense, but not always.

How did you decide to build this?
Long story! I’m so glad you asked!

Last September, we brought our entire team to Maui for our fall workaway. Twice a year, we take our team somewhere outside the office to reflect on our culture and to work on projects that are important but not urgent. One of our projects last fall was to explore natural language processing and machine learning, as they can be applied to email.

We experimented with a bunch of technology, found that holiday support emails are the most positive, and then went back to the office to resume working on urgent things. And we posted a job listing for a data scientist.

Then, around the holidays, Chris from our marketing team came up with an inspired idea – let’s send a useful, thoughtful year-in-review email to our customers. It took us a while to build it, but if you were already a Boomerang user back in February, you may remember getting an email from us that showed a set of data-backed principles for writing effective emails. We also published the data as a blog post, had a ton of stories written about it, and made our first appearance on the Today show.

Several dozen of you asked if we could build a real-time version of our blog post that would apply to emails as you write them. We still had a healthy appetite for machine learning (and a data scientist), so we decided to see if we could bring the recommendations from the data to your fingertips.

I can’t wait to see what we’ll come up with at our next workaway.

Why did you pick these aspects to focus on?
Boomerang Respondable makes its predictions based on machine learning algorithms applied to a corpus of hundreds of millions of data points in public and proprietary (aggregated and anonymized) email datasets and other datasets.

Our research from the past several months isolated a number of factors that correlate strongly with the likelihood of receiving a response to an email:
Message subject length
Message word count
Reading Level
Question count
We looked at several other factors that didn’t make the cut. Some of them just didn’t correlate strongly enough to getting a response. For others, we couldn’t find good enough data sources to train machine learning algorithms effectively yet.

We’ll continue to explore other factors that influence response rates and incorporate them into the product in the future.

Can you tell me more about the machine learning techniques?
We used a variety of machine learning techniques for Respondable, including neural networks, deep learning, and decision trees/forests. Respondable layers these techniques together in a manner that provides a balance between consistency and quality of results, the ability to translate those results into actionable information, and the speed and performance to deliver results in real-time for a satisfactory user experience. We’ll write more in an upcoming blog post – once our engineers have a chance to rest their fingers from getting ready for today’s release!

Via: Venture Beat