HubSpot AI: Building a SMART SAAS Platform

Are you interested in learning more about the use of AI in HubSpot?

The new HubSpot AI functionality is making the tool better for HubSpot users around the globe.

HubSpot is using AI to build a smart SAAS platform for sales, marketing, and service professionals. In this interview we talk to Kevin Walsh from HubSpot about what the AI team is thinking, doing, and dreaming about now and in the future.

About the Expert

Kevin Walsh

Kevin Walsh is the Product Group Lead for artificial intelligence at HubSpot. His teams are responsible for building HubSpot's machine learning as a service platform and for integrating machine learning models across HubSpot's product lines including Contact deduplication, Adaptive page testing, SEO, and others. The teams also implement machine learning models for HubSpot's internal teams like customer support customer success, and sales. Prior to AI, Kevin worked as the product manager for several data products including HubSpot's Salesforce integration and HubSpot Insights, the enrichment service. Before working at HubSpot, Kevin worked in catastrophe modeling analytics for Guy Carpenter. As an undergraduate, Kevin studied economics at Northeastern and also studied data science through Northeastern's graduate College of Computer and Information Science.


Full Transcript

Kevin Walsh 0:00
We use artificial intelligence to help our customers provide a better experience for their customers. We do that through data hygiene, through security and through personalization within our app and within our services. And ultimately, it's just one more tool to help provide a better customer experience. Are

George Thomas 0:15
you a HubSpot user looking for fun and interactive education that helps you be better at HubSpot? Then Welcome to sprocket talk. That's right. It's your boy George B. Thomas from sprocket bring you yet another HubSpot educational video. Now today we're going to talk about AI, HubSpot, and all the way that it's being built into a Smart SAS platform to help you as a HubSpot user work faster and smarter. If you're a sales marketing or service professional looking to learn more about HubSpot and inbound strategies to help your business grow, then guess what you're in the right spot. Make sure you subscribe to keep learning more and use the comment section below to ask us all the inbound and HubSpot questions that you may have. Now, let's get into the good stuff and start out by learning more about our guest today. Kevin Walsh from HubSpot

Kevin Walsh 1:07
have been at HubSpot for six years I actually started my product career working on the Salesforce integration. So I've always been working on large infrastructure problems here at HubSpot, and as HubSpot decided to make a bigger investment into data pipelines and machine learning. The timing was just right for me to begin working in that area.

George Thomas 1:26
The first thing on my mind that I wanted to ask Kevin was why is AI in SAS platforms specifically HubSpot excite him? Why does he wake up in the morning ready to find really cool ways to implement AI technology into the HubSpot platform for us, the HubSpot users

Kevin Walsh 1:43
it's great to be bringing these sort of state of the art products and technologies and applying them to helping our customers grow their businesses and provide a better experience for them. So it's exciting for me because I get to work on all of our different product lines and in collaboration with lots of different product teams. And so working in sort of that like consultative approach internally at HubSpot is is really rewarding. We've, I like to think that our teams have done a great job of establishing how collaborative projects at HubSpot should be done. And we've had a few successes over the past year. So it's been it's been motivating for sure.

George Thomas 2:18
If you're like most people, you're wondering, well, George, what's in it for me? Well, I'm the same way. So I asked Kevin, why is it important to HubSpot users like myself and those watching the sprocket tears that HubSpot focuses on the use of artificial intelligence in the HubSpot platform?

Kevin Walsh 2:34
Yeah, I mean, I think the best analogy these days is sort of where like cloud technology was, you know, 1015 years ago where the fact that HubSpot is using machine learning shouldn't be the first thing that comes to mind. It's just another tool to help people ultimately help our customers provide a better customer experience for their customers. And that's where we focus. So the fact that we're using machine learning like we do have, you know, a state of the art machine learning platform and worldclass machine learning team, but it's really more important that you're attracting people to your website, engaging with those people and then helping them be happy customers after the deal. And that's what we've leaned into is that it's important to use AI for your business today, because it will help provide a better customer experience for your customers. And if you're not providing the best possible customer experience, like somebody's probably going to come and eat your lunch. So machine learning and artificial intelligence are just additional tools to help you deliver on that more important value of the better customer experience.

George Thomas 3:31
Next, I had to ask Kevin, if there were any specific tools that they were leaning into with AI and that he himself thought, wow, this is cool, nerdy as well as useful. And here's what he had to say.

Kevin Walsh 3:43
Yeah, the the two big flagship projects that we had last year, were CRM, deduplication and adaptive page testing for the CMS. So CRM deduplication is interesting because we've taken a problem that exists It's in what's called n squared exponential time, where, in order to D duplicate a contact, you have to compare that contact to every other contact in the portal. And so even if that only takes a couple milliseconds, if you're comparing one contact to millions of contacts, and then all of those other millions of contacts to each other, it will take, you know, thousands of years. And so the innovation that we had to allow for contact and company duplication in the HubSpot CRM was that we can actually duplicate millions of records in just a few hours. So reducing that search space was the real innovation. And then in the app today, there's a screen where people can accept or reject the things that we think are likely to be duplicates. And we consider that feedback to make our recommendations in the future better. So reducing that problem from thousands of years down to a few hours for even portals that have millions of records was the real innovation there and we have a lot of really strong HubSpot Connect partners that offer great solutions for deduplication but the heart piece of that is they're going to cap out around a few hundred thousand records because it becomes untenable without machine learning to actually do that. So we're really proud of the content duplication and company duplication product for the CRM. The other big sort of flagship projects we had last year was adaptive page testing for the HubSpot CMS. And that allows you to create up to five page variants for landing and website pages. And then automatically, we are going to adjust the traffic being served to those variants based on which pages are converting the highest. So it's an improvement over our traditional AV testing, in that you get to run up to five variants instead of just two. And we're actually going to help you increase your conversion rate more quickly, because we're dynamically adjusting which traffic is being served to the best pages over time. And you can sort of set that and forget that. But we're also showing you in the app, you know, here's how the variants are performing. Here's the conversion rate for them. And this is what the traffic shift is today. So we're also really pleased with with how that product came out. And we're looking forward to trying to apply that same technology to other parts of the app, probably somewhere in email, and potentially somewhere in workflows too. But don't hold me to that just yet. At this

George Thomas 6:09
point in the interview, I had to scratch my own curious itch. And I explained to Kevin, that I had seen something very interesting happening in the HubSpot tool over the last couple of months. You see, a couple months ago, I was working in the blog tool, and I went to go to Settings However, instead of just general settings, it took me to the blog settings. So I tested this for contacts, landing pages, and Yep, each time I was taken to the specific settings area that I needed. So I was curious and asked Kevin, why are you using AI to add this type of functionality to HubSpot?

Kevin Walsh 6:43
Yeah, I mean, personalization through the app will be is more and more important for SAS products, whether that's, you know, commercial SAS, like HubSpot, or if you look at Spotify and Netflix, I think are often held up as sort of like the pinnacle of personalized experiences. It's the difference between Having configurable dashboards versus the product just knows what you're looking for every one of us when we log into Spotify or even Facebook, we're going to see different content based on what we engage with in the past. And something like a music library is a little bit more constrained so that we can just show you. We're all seeing songs, but we may see songs that are more relevant to us. We definitely aspire to make HubSpot, which serves lots of different business personas. If you're a sales rep when you log in, you should see the most relevant things to you and we should help guide you through that experience. And on the same side, if you're a content producer, we should also show you the tools that are most relevant to you it's a little bit trickier because that intense space of like what do people come into HubSpot to do is much broader than it is for something like Facebook, Netflix or Spotify, which are usually just, you know, I want to watch TV or a movie or I want to listen to music, personalizing those experiences much more constrained, but we definitely look to those consumer grade products and want to make HubSpot sort of that turnkey world class experience like that. And you can expect to see more of that.

George Thomas 7:58
Look. In my mind, AI is difficult So I wanted to make sure to ask Kevin if there were any hurdles they had faced while taking this journey through making HubSpot, a smart SAS platform with the use of artificial intelligence,

Kevin Walsh 8:10
I would say in general, what we've learned about CRM data is that it's often too messy to use. So this was something that continues to be a struggle for certain applications, like lead scoring, and depends on the data in your CRM. However, if the data in your CRM isn't quality enough or isn't consistent enough, then any models that you produce on that data are also going to be shaky and inconsistent. So two of the things that we've that we continue to work on to help address the messy data. problem in CRM is, well, like I mentioned, contact duplication and company duplication. The real behind the scenes value of that is we're helping people have more consistent data in their CRM. So it's more useful if you have fewer duplicate records. That means your automation is better, your segmentation is better your reporting is better. So how With data hygiene on the deep side, we also help by providing enrichment data through the HubSpot insights database, so that if we want to build predictive models to do something like a company need application, we can enrich the company records that you have so that we have more data points to compare, knowing, like these industries are the same, or these addresses are the same, or even the business names are the same. So data hygiene, enrichment data, there was one more and we're also in beta right now. We're working to help people parse signature information out of their emails. So that, you know, if I send you an email, George says at the bottom, like Kevin Walsh, product group lead at HubSpot, you know, that's information that I've willingly given to you as someone I'm doing business with. So we're helping users like you to get that information from the body of the email into the CRM so that it's automatically added. So that way, you have my title and you have my business name, right in the contact record. And if you're a manager or if you're a marketer, having automated data entry from your sales team is huge because it means that you can can create lists based on seniority or you can understand are the deals that I'm tracking? Do they have the right people on them, things like that. So the initial frustrations around CRM data is really messy. And it's like blocking us from doing a lot of use cases have actually presented an opportunity for us to then actually use machine learning to help with data hygiene, which is a problem that affects every customer. So we know that any work that we do there is going to immediately value be valued by all of our users, and it actually unlocks other possibilities for us to then use that cleaner data set to do

additional machine learning projects in the future.

George Thomas 10:35
As they listen to Kevin I kept thinking, is there anything we can do as HubSpot users to make the platform smarter? So that's exactly what I asked him. I mean, hey, inquiring minds we need to know,

Kevin Walsh 10:46
at the individual level. To be honest, there isn't a ton that any one person can do to their portal unless you're doing like a complete machine learning projects yourself. If you have a data scientist on staff and like you're building custom data sets to build custom model Then definitely like individual attention to data hygiene is key. Our approach is that that shouldn't be the focus of most of our users who are small and medium enterprises. And so we're trying to do the work to make that easy for you. We don't want sales reps to be spending their time making sure that their data is clean, we want to make sure that they're engaging with their prospects and being helpful in a sales process to close the deal. And we want to make sure that content marketers are spending their time creating valuable content that is like rich and engaging. So the sort of maintenance work that's required to do machine learning, our hope is that we can solve most of those problems for users. And you shouldn't have to think about it much in the same way that you don't think about, you know, engaging very much with your Spotify or your Netflix recommendations, or even your Google search results. All of those are highly personalized experiences that don't require a lot of input from the end user. And that's what we're aspiring to do at HubSpot.

George Thomas 11:53
Sometimes folks get a bit weirded out by the thought of AI or artificial intelligence. So I knew I really knew We needed to ask Kevin if there were any myths about AI, and the use of AI in HubSpot that needed to be debunked right here. And right now,

Kevin Walsh 12:08
sure, I think the two biggest myths that we run into and sort of applied machine learning for business today. The first one is, we always hear about how much data we have. And so it's like I have this mountain of data, I just want a machine learn it. And that really doesn't turn out to be the case. Machine learning requires like meticulously clean, organized data sets that represent some real world interaction. So if you're using the HubSpot CRM, but you're not tracking your actual revenue in the HubSpot CRM, it's really difficult for us to know where your net new revenue opportunities are going to be because the machine systems don't actually have that visibility into you know, the real world business logic that's sitting on top of it. And if you just have disorganized data as the it's the old garbage in, garbage out, machine learning isn't going to fix that. So that's that's sort of the first one is just like lots of data. That must be a good thing is really not the case. More often than not, you're better off to have less data but also very, like purpose driven clean datasets for a very specific machine learning question. The other big myth that we see a lot in sort of product marketing materials is this concept that artificial intelligence is smarter than you. I think today artificial intelligence is really just a more sophisticated kind of automation. And analogy that we use at HubSpot is sort of this 10 experts things if you give people a task, whether that's sales reps, or even interns, and you said, I want you to do this particular task, like identifying an email where the signature is 10 people could do that really consistently. That's a great application for machine learning. If you're trying to do something that 10 people wouldn't do the same way. Like I want you to write the paragraph of this blog post for me, you could give 10 expert content creators that task, they would all write

a blog post for you, but they would look very differently. So the output

of those models is very inconsistent. So in general, we think of machine Learning as a more sophisticated tool for automation, we don't think of it as this crystal ball or this like ultra intelligent being who's going to be smarter than you are at your own job. They're just going to help you make sure that you're doing all the tasks that you need to do that sometimes slip your mind like data entry, and like task automation.

George Thomas 14:18
I'm sure you're like me when it comes to wondering what's next. So I wanted to ask Kevin what the future looked like. Were there some things on the horizon that he was excited about that he was able to actually legally share with us? This rocket tears?

Kevin Walsh 14:31
Yeah, so like I mentioned earlier, we haven't beta right now, the signature parts parser, which is one thread of what we're calling never enter data again. And so we want to make it so that the information that exists right now in emails and in messages through our conversations product, and even on calls for call, transcripts, all of that information about who's on the call, you know, when should the follow up actions be which products are people interested in all of that should be automated data entry. It's not Things that are super challenging for sales reps to do to go and actually create the right deal with the right product on it with the right due date, those types of tasks we think that we can automate, and you'll see more of that over time. And the signature parser is just sort of like stepping our toes into that pool. Once we have more automated data entry, it's a rising tide that lifts all boats because your reporting is going to become much more powerful when you can segment based on seniority based on topics discussed during a deal. topics discussed on support calls, all of that will be able to provide a much better unified sort of customer view of how are people interacting with my business and what are they talking about without putting that burden onto your sales and service and marketing reps.

George Thomas 15:42
As we got to the end of our time together this thought came to mind what conversations are we not having that we should have when it comes to HubSpot and AI as well as whatever the missing conversation piece that would go along with it? Well, Kevin surprised me with his answer. Take a listen to what he has. had to say.

Kevin Walsh 16:00
Yeah, I think one thing that we don't talk about very often because it's hard to see in the product is we actually have a whole case of use cases around security and product protection which are vital to the HubSpot platform. But we actually use machine learning to help prevent phishing emails from being sent by HubSpot. We help it from people sending emails to lists that are potentially going to have a high bounce list. So it helps make sure that HubSpot is providing a best in class deliverability for our users that are following best practices, and also for people who maybe got a list from a place that isn't the most white hat, but they didn't actually know that we're going to kind of help them to make sure that they know, hey, do you actually have everybody's legal consent so that people can be complying with the various privacy legislation in the EU and the United States. So around email protection, there's there's a lot of use cases that you don't we don't talk a ton about because they're not like the flashiest sexiest use cases, but they're critical to making HubSpot, a world class email provider. In a similar class. We also have anti abuse systems that help with login making sure that when we see suspicious API or suspicious login activity, we're verifying with the account owner that that's who that actually is. And again, those things are usually at the top of a product marketing landing page. But they're critical to making sure that our systems are safe and secure for our users. And they're also respecting the privacy rights of the people that our customers are engaging with.

George Thomas 17:21
Wow, what a great interview. Do you have some ideas for how HubSpot could implement AI into their platform? If so, hit up the comments section, or do you have a use case where HubSpot save you time, money or other resources? Let us know about that as well. Hey, did you like this video, then head over to sprocket talk comm for more actionable and tactical HubSpot user resources, make sure you subscribe to the channel and hit that bell as well for those instant notifications, and so we know that you're part of this sprocket talk community. Until next time, make sure you're focused on being a happy, helpful, humble human. And as always, make sure you're doing some happy hub, spotting