At TrailheaDX 2019, Salesforce's Michael Machado spoke with TechRepublic about the unique challenges of building Einstein Voice Assistant within the company's highly-customizable platform.
At TrailHeaDX 2019, TechRepublic Editor in Chief Bill Detwiler spoke with Salesforce Senior Director of Product Einstein Voice and Deep Learning Michael Machado about building the Einstein Voice Assistant and the many ways businesses can utilize the platform. The following is an edited transcript of the interview.
Bill Detwiler: With Einstein Voice, you've added the convenience of a voice assistant to the Einstein platform. Why the decision to add that in? Everybody else has voice-enabled assistance these days--what was it about adding that to Einstein that made it the right decision?
Michael Machado: We actually heard it from our customers--they really wanted to be able to interact with their data. There's 20 years of investment in the Salesforce Platform and Einstein Voice, and Einstein Voice Assistant is really about bringing new user experience to our platform.
Sales workers, service workers, marketers... all that data that they want exposed to them can be voice-activated. It can be input via voice and data can be exported via voice. We're thinking through a lot of different use cases, basically scenarios those workers are in, to really leverage that capability, that user experience. Your sales workers, your service workers, your marketers, all that data that they want exposed to them can be voice-activated, can be inputted via voice and data can be exported via voice. We're thinking through a lot of different use cases, basically scenarios those workers are in, to really leverage that capability, that user experience.
SEE: Data analytics: A guide for business leaders (free PDF) (TechRepublic)
Bill Detwiler: Talk about how it makes it easier--what specifically did they want to do? Use it on their mobile devices? Was it while they were sitting at a desktop? It seems like you've got people that are in the field. Say for example, if they're sales people in the field, and they want to be able to talk to their phone like they talk to a regular voice assistant: 'Tell me about this, about my contact.' Is that what they really wanted?
Michael Machado: I think the biggest thing is how your data is maintained over time. It's different for every scenario of voice product we launch. At Dreamforce last year, we launched three products. One was a voice assistant on mobile, which you mentioned. One was voice briefings, which is actually enabled via any smart speaker device. The final one was voice analytics, which is much more of a desktop experience where you're navigating dashboards and analytics through voice.
Each one of those has a different user and a different problem that they're trying to accomplish. For mobile, it's all about more granular data, more real-time data being entered into Salesforce. Mobile workers, field sales, field service--they don't have time to stop, open up Salesforce, open up their laptop and enter data. It's painful for them, it's a burden on them, but it's the most important thing to make sure your forecast looks right, to make sure your team is allocating resources the right way.
For voice assistant on mobile, it's really about thinking about how we can create new records, new tasks, new activities that you've done, streamline the way that they actually enter that data, so you're not finding the right record in Salesforce. We're letting Einstein not just let you voice dictate notes, but actually extract out meaning using our natural language understanding models to parse that data, sync it with Salesforce, and actually proactively say, 'I think that meeting you just had was with this customer. This is the contact you met with, and these are the actions you need to take because of that meeting.'
Maybe it's updating a stage, maybe it's creating a task for a friend. Those are the kind of actions that our field sales and field service workers are doing.
Now, if you look at analytics, kind of a different scenario. That's someone who might not be as accustomed to our product for analytics and they need to navigate dashboards, filter on the right region, make sure they're actually looking at the right forecast the right way. They can now use their voice to essentially extract away what might be a user interface they haven't worked with a lot to a voice-enabled interface.
And then finally, briefings. That's the idea of how you wake up in the morning and I ask Siri the weather or I ask Alexa the weather. I maybe play a song and I check the traffic. That same kind of concept is what we've brought to daily briefings, so your sales/service kind of VP can say, 'How's my sales forecast daily briefing looking?' And actually, we'll read out a dashboard for that user where they're able to work on the go, work from their home and kind of multitask, but be able to get information out of Salesforce where they're not really having to think about it. A very intuitive way.
Bill Detwiler: What were the challenges to be able to incorporate voice in the Salesforce platform? I mean, you think about the weather, you think about, 'Read me my calendar list.' Fairly simple task, there's data behind that, right? But it seems like dealing with customer data or dealing with analytics data, and dealing with those dashboards is pretty complex. There's a lot more complex data behind that, so what were the unique challenges putting voice on top of the Einstein platform?
Michael Machado: Salesforce, for 20 years, has been giving customers the power of customization. It's why we've been so successful. Our platform is really what's embodied Salesforce and made our customers successful. It's why TDX right now, there's hundreds of thousands of people here who their job is to really customize Salesforce.
As someone who works in machine learning, that makes our job a little bit more difficult, because we're not talking about a single model where we understand the schema for how we extract out weather, which is, 'Find out their location, find out the time range they want to know the weather for and the city,' and you're good to go.
For us, we have to think about how our customers customize Salesforce for their business process--which is different for every customer.
Bill Detwiler: And it's one of Salesforce's selling points--how it can be ultimately customizable, but it also means that when you design a product like Einstein, how do you keep people within a box?
Michael Machado: The expectations have grown so much. That consumer experience is natural for a lot of us. Like we said, we talk to a lot of our assistants every day, but we expect that consumer experience to translate to our enterprise software experience. That's the challenge that we've taken on, and we're thinking about different workers and different scenarios and trying to execute on those.
Field workers is a really big one for us, helping them log activities and doing that in a structured way that actually makes sure that the rest of the data still has the integrity we need. That's what we're trying to knock off first. We'll help you create contacts, help you create opportunities. The most common things that rep does, we want to make sure that's what streamline does, asking for the weather with Alexa.
Bill Detwiler: How customizable is voice for the customers? The admins that come in, the devs that come in, what can they do with voice to make it unique to that customer?
Michael Machado: Since we've kicked off this whole project--on the mobile assistant especially--it's really been about not locking the customer in any workforce. We know almost 80% of our customers store data in custom objects. From the very beginning, we architected our service to be able to support custom objects, be able to support custom fields and be able to update and log activities on those custom objects and fields.
That's kind of paramount to everything that our users do. Now, we also try to guide the user though. If you think you can do everything then you're going to have a bad experience, so we really try to put them in the mode of, 'Think of your most common workforce you're going through and have the admin set up templates,' (we call them). You can think these as sort of the routines that the agent goes through, and help them focus on that routine and log their memo and log their activities via voice and have the admin structure that for us so we know out of the whole domain of Salesforce and all your CRM (customer relationship management) data, where we need to be focusing. Is this a sales rep? Is this a service rep? Are they working on an account? Are they working on an opportunity? Things like that.
Bill Detwiler: What have you heard from companies that have been deploying this? What kind of feedback? The impetus for adding voice to the platform really grew out of the customer feedback. Now that they have their hands on the product, what feedback are you hearing about how they're using it?
Michael Machado: I think one of the most interesting things is it's all about data enrichment for us. How can you get more data into Salesforce? How can you get that data in real-time, closer to the actual event taking place? And then, how can you get more granularity? I think that was one of the most surprising ones because we had customers who said, 'I've got this one rep, he logs 100 activities a month. He's one of my most productive reps I have. But all I used to really know was he met with John at Acme. That's all I knew. After five meetings with John, all of a sudden, the stage closed or the deal closed.' But, now we're actually seeing--because voice is sort of this intuitive user experience—we're seeing reps come in and sort of stream-of-consciousness talk about the meetings they had.
That actually is really powerful for Salesforce and it's powerful for customers. They're starting to not just know the headline, they're starting to have the full article about really what happened in those meetings and why they're customers are being successful or why their reps are having troubles--that's really important for data enrichment.
Bill Detwiler: How difficult is it when you're building a voice integration, a voice assistant, to prepare for all the nuisances and the way people speak? You've got natural language processing (NLP), obviously words like 'client,' worlds like 'representative,' words like 'sale, contract,' things that are consistent across the sales roar. But, there's also going to be the individual ways people describe things. There's a lot of different terms that maybe people use in different regions, things like that. Talk a little bit about how you managed to overcome that barrier.
Michael Machado: I don't think there's ever a silver bullet for solving it for everyone. I'd be lying if I said, 'We figured it out. We're the only ones who know how to do it.' But I think because we're focused on sales, on service, on marketing, we have an advantage over a consumer application. They've got to focus on a myriad of different use cases, a myriad of different types of users. We can start focusing on specific industries, and that's a great way to go about it.
While we focus both on the ASR (automatic speech recognition) and actually tuning that to be more adaptable for maybe a scenario user is. Is there a lot of background noise if you're at a construction site? Is a hospital going to have a certain kind of sound clarity that we want to be able to get through? If I'm giving a speech Moscone Center right now, what kind of background noise am I going to deal with at a conference center? So, we can think about the ASR and the speech recognition and how we can improve that.
We also thinking about the NLP side. We allow a lot of our customers to bring in their own data, build their own custom model that's just for them. The advantage that gives us is they can train that model with their own specific terms and we can actually leverage that to make sure that customer's experience is tailored exactly to their use case. If you're in healthcare, there's a million acronyms. I only know half of them.
Bill Detwiler: Right. If you're in healthcare, pharmaceuticals, it's going to be a lot different than construction or retail.
Michael Machado: I used to have a cheat sheet to figure out all my acronyms that I had to know for my work day. That idea is why we bring that same customization that Salesforce does to the platform. We actually do that at the natural language understanding side, and we want to bring that to speech recognition as well.
Bill Detwiler: With voice, is there any reluctance? People don't like it when you move their cheese. There're people that still want to enter in information on keyboards, people that love their Blackberry and held onto them forever. Do you get a little bit of that reluctance or do you hear customers telling you, 'Well, so-and-so doesn't really want to use the voice platform because it's just new or they think it's dumb.' Or does it seem to be that it really is embraced widely because it makes people's jobs easier?
Michael Machado: I really think of voice being a way for making Salesforce work for the user versus them kind of having to work for Salesforce, work for their CRM. That's one of the most powerful things hearing from customers is, 'It used to take me five minutes to log in activity' or, 'Finding the right record in the right field in Salesforce was a navigational nightmare when I'm on the go and I can't even look at my phone.'
We're actually hearing more and more, 'I want this feature voice-enabled. Help me create contacts quicker. Help me create this thing quicker,' because once they kind of get into it and they get that experience down. Just like anything, there's a little bit of a learning curve. You have to get over the hump of your first time using Siri for instance, but once they kind of get the hang of it, they realize that it's a productivity boost like we've never had before. We kind of gamified it a little bit, too, and made it fun, and that's really exciting to see people actually enjoy working with the application and seeing how many activities they can log via it.
We've had a couple people run contests even with their customers, and their reps get excited. A little competition brings out the best of us I think.
Bill Detwiler: What's on the horizon for voice, for Einstein Voice?
Michael Machado: Einstein Voice, if you really think of it as an answer. I kind of mentioned the smart speakers, I mentioned desktop experiences, being able to navigate dashboards in the mobile application. One is kind of uniting a lot of those things together, so anything you do on mobile you could be doing via smart speaker. That's a really big initiative. Really think about how we can help level, we kind of call them templates, and skills, but take these skills and be able to deploy them over multiple channels. I think that's pretty powerful.
The other thing is telephony I think is a really interesting one. Probably majority of our reps are just sitting inside offices on the phone, call centers or inside sales. How can we enable voice via that? What does that experience look like and who is that user? Those are things we're investigating heavily right now. There's some hints around--if you're at TDX right now, there are hints of where we're thinking about that.
And then, just continuing to drive the mobile application as being really a voice first application where you're communicating not only to Salesforce but to other team members and bringing in that team sales, team service department together via voice.
Bill Detwiler: What about analytics and voice? Are we approaching the time where that manager or the director, can come in and say: 'Don't just read me this pre-programmed dashboard, but actually interact with the application through voice and have it perform a new analysis. Have it pick up the keywords and understand I want to look at the forecast for these two dates, change this range, or tell me the correlation between this and this. that someone hasn't built the dashboard for yet, but it's two pieces of data that you know you have'--how far away are we from that?
Michael Machado: You can see we're investigating, we're teasing it, we got pilot customers already trying to tell us what they want and we're getting more and more feedback around that. The voicification of data is really this kind of buzzword we're hearing about which is, 'I've got data that's sitting there. I need to understand what's happening with that data.' Whether that's via analytics or via reports or just the idea that someone can extract information, 'Tell me John's phone number really quick,' that data needs to be voice-enabled. It needs to be intuitively voice-enabled, and I think that's the biggest thing.
It's not easy, but we think it's important. We think we want to really democratize who can use these voice applications and think of the different personas and how they leverage it. Like in the scenario you mentioned, probably more like a VP of sales, VP of service really wants to understand and dig into the data a lot more, or even analyst. But you could even see some of these scenarios bridging into different personas, the admin themselves. How can the admin kick off a new work flow? Sending an email any time to one of my reps any time they haven't logged an activity in five months? That's pretty powerful, but if you're going through a click-path the whole way through, you really got to change your mindset. If you just got to take a business process and you can voice enable it, I think you'll see the new wave of enterprise applications going that direction.
- Salesforce's Parker Harris offers CXO insights for managing complex IT integrations and handing the inevitable crisis (TechRepublic)
- Developers: Salesforce open sources framework so you can build apps on any platform (TechRepublic)
- Feature comparison: Data analytics software and services (Tech Pro Research)
- Salesforce Service Cloud Einstein adds AI to CRM to boost customer service (TechRepublic)
- Salesforce brings more Einstein AI tools to Financial Services Cloud (ZDNet)
- Salesforce's $15.7 billion Tableau acquisition: Everything you need to know (ZDNet)
- Machine Learning: More must-read coverage (TechRepublic on Flipboard)