57. Hoala Greevy: "It's a matter of sorting the data, training the data, and then using those new learnings to provide greater phishing detection."

Episode 57 of HIPAA Critical features an interview about AI with Paubox Founder CEO, Hoala Greevy.

Rather read?

Hoala Greevy: Yeah, sure. 

During the pandemic, we've been staying in touch with our customers, answering them and asking their questions about trends we see in the market. One of the big trends for us is we've seen email use go up and not down, in fact, quite a bit. 

One example is, many of our mid-market and enterprise customers have taken advantage of the pandemic, in a sense, to catch up on projects they've wanted to do for some time now. One of them's been replacing the physical fax machines and converting those fax numbers to efaxes or email, they still need to have a fax number because of the third parties they deal with. 

But now these things are coming in as email attachments. Again, that's more email coming in and fewer printers are being used, more emails being sent. This, of course, flows with a hybrid remote environment. 

Then therein lies the opportunity for email AI. It is taking that influx of new email, and providing workflow automation for our customers. This is also known as robotic process automation or RPA. 

That's a really exciting one. There is taking the email, not all of it, certain types of email, and are doing workflow automation, business process automation.

Hoala: Sure. 

We do Zoom Social Mixers once a month, any listener to our podcast would know this. You mentioned this at the end of our podcast with an open invite.

In a July Zoom Social Mixer, one of our customers said, “Hey, I have this problem where our corporate voicemail system sends an email, audio attachment if that person doesn't pick up the phone, on their work line. I'm afraid that they'll open these attachments in public on their smartphones because some sensitive stuff will be in their voicemail. What can you do about that?”

Hoala: Yeah, especially if it's sent to a doctor or physician. Who knows, right? It's a regional healthcare provider, hundreds of 1000s of employees, lots of opportunities for risk there. 

He saw it as “How can Paubox help us reduce our risk for this particular problem?” 

So what we ended up doing was we set up our email AI for each customer that wants it and we look for the presence of an audio file, we open the audio file, we apply natural language processing to it and we transcribe what's said in the message into text. We insert that text at the bottom of the email sent by the voicemail system and the attachments are still part of the email. 

But now the end-user gets to just scroll on their phone and read the voicemail.

Hoala: Yeah, who wants to download and wait? 

When it's just easier to just scroll on your phone, and you see the same thing. What's great about this solution is the transcription is so accurate, that they don't want to open it anyway. Because the fastest thing to do is just scroll and read. So it's reliable, it's secure and easy to use. And it fits all three big ideas at Paubox: security, reliability and ease of use. 

He immediately knew that this was something we needed to build. And we built it within six days, rolled it out, and got happy customers using it. And that's our first foray into AI. 

Hoala: Oh yeah, sure. 

We have a behavioral health organization in Florida. They're looking to automate some of their referral business workflows. This is for providing services for their patient population. They get referrals from other social workers, the court system, any variety of places for their clients. 

They need to get that information and put it into their EHR system, which is specifically made for behavioral health. This is basically emails coming in from someone at that org, opening the email, and then transcribing, or copy-paste, manual input, basically, into their EHR system. 

What we propose is, hey, we're already providing you folks with inbound email security, ransomware, ExecProtect, phishing protection, etc. We're seeing all the email flow through anyway, why don't we set up a system whereby we identify these referral-based emails coming in, will parse those emails and input that data into your EHR system? Simple idea, not easy to do, because hey, this is healthcare and it is complicated. 

And we're heads down on that now. I think there's a lot of applications for other EHR systems out there. So pretty exciting stuff. Our customer can't wait to see this thing done. Because they've got a set of dominoes sitting behind it of other workflows they want to automate.

Hoala: Yeah, so I guess if we look at our own business Paubox, there's very little of business correspondence that happens just in our physical mailbox, right. If we hire an employee in a new state, we’ve got to go register with that state's unemployment system. 

Oftentimes, that's a paper-based workflow. But other than that, every invoice, every receipt, all that stuff flows in on email. Whether it's tying to an accounting system, a billing system, a CRM, system, EHR, all this stuff's happening in email. We're going to rely on our customers and customer feedback to kind of point us in the direction of where to go next. 

It just seems like there's just so much opportunity here. There's a human opening that email and doing manual work, which we've defined as a human endpoint, which is a term I learned about in a book on email AI I recently read. 

Again, a human endpoint is when you have two systems or two sets of software that can't talk to each other. There's a human in the middle providing duct tape, i.e. manual input data entry to glue them together. And why not use software for that specifically, email AI software?

Hoala: Yeah, this is a perfect example for AI to provide value here. 

As we see phishing and ransomware campaigns continuously evolve, it gets very hard for a human mind to programmatically insert logic into an email security system to keep up with these threats. They're just seeing there gets to be at some point too many connections for the human mind to have in their head and to make the correct assumption at scale. 

This is really where machine learning and AI come in handy, where you're providing correctly labeled data sets, in our case, what's a good email, and what's bad email, and then providing enough correctly labeled data where the machine learning algorithm can provide deeper insight that a human wouldn't be able to do. 

We're heads down in that currently as we speak. And, you know, the clever part about this whole thing is, we see tons of this every day coming in because we provide inbound security. 

It's a matter of sorting the data, training the data, and then using those new learnings to provide benefit across our entire platform of customer base to provide greater phishing detection, because that's what our customers want. The spam kind of a subjective thing kind of depends on the recipient. But we know 100% of people don't want phishing and ransomware attacks.

Hoala: Oh, yeah, I mean, if done correctly, they can and should be this is the perfect use case for AI. We look to provide leadership in that direction, especially for email AI as a concept. There's very little being done with it right now.

Hoala: No, I think it will be a nice complement to them. 

If we look at if we think about artificial intelligence, what it's really doing is providing more and cheaper access to prediction, the democratization of prediction, if you will. 

In the case of our voicemail transcription robot, the NLP is providing a prediction on what that text, that voicemail is actually saying. And if the prediction is accurate enough, then that's providing new insight for the human. And then they can make a judgment call from there. 

There are always going to be humans needed to make judgment calls. But if they can leverage AI to provide them more accurate and cheaper predictions, then they'll be more valuable and more highly paid in their job. 

When it comes to compliance officers and IT managers, I think they will learn to view this as a complementary toolkit or another feather in their hat, if you will, something you'd want to take with you across jobs.

Hoala: Oh, yeah, for sure. I have skimmed an article recently where they provided some theoretical ways to do this. And that would not be very much of a stretch for us to provide that as a value add to our customers as well. Once we get more comfortable with email AI as it relates to phishing, why not include a phishing simulation?

Hoala: Yeah, that would be a nice add-on for us to provide for sure.

Hoala: We are talking about this often in our monthly Zoom Social Mixers. So please join us if you're a customer or currently talking to our sales team and thinking about becoming a customer. We actually do listen and take action on the things discussed in them.