Wed, 11 July 2018
Today on the Salesforce Admins Podcast we’re joined by Molly Mahar, Product Designer of User Interface and User Experience at Salesforce, to help us demystify the Einstein machinery behind bots.
Join us as we talk about the types of bots that are out there, natural language processing, and how you can get your data in order to train your own bot with Einstein.
You should subscribe for the full episode, but here are a few takeaways from our conversation with Molly Mahar.
From astronaut to film production to UX.
When she grew up, Molly wanted to be an astronaut. “I went to this really nerdy science and tech high school,” where she learned about science and programming. “Probably some of the choices I’ve made in my life, since then, were still trying to reach for that astronaut, but getting there in different ways,” she says. After high school, she studied film and working in the industry for a long time. “When UX became a bigger thing,” she says, “where you need to combine technical knowledge and creative abilities into problem-solving for people.”
Today, Molly is a Product Designer at Salesforce working on Einstein bots. “I help work with our researchers to determine what our Admins and other customers need to be able to do in the product,” and then works with developers and product managers to figure out the how, what, and when of it. She also is in charge of figuring out visually how it will look, and the workflow and process behind it. She makes the magic happen.
Rule-based bots vs. natural language processing.
“You hear about AI everywhere right now, and it’s going into so many different products that you use, and you may or may not know that it’s within those products,” Molly says. Bots, as we discussed last week, are not AI but can involve AI. You can have a bot that is entirely rule-based, “which is basically just a computer program,” she says. You can also have an AI-based bot that uses natural language understanding that can interpret what a user is asking for without having everything explicitly programmed into it.
“You basically take a lot of data of the way that people say things, and you plug that into a very complicated algorithm that creates a model, which is a way of understanding the world, and through that model the computer is able to make predictions about what a customer is asking for,” Molly explains. Instead of computer programming rules, it uses a lot of examples from the data it has. You can mix these rule-based and algorithmic techniques, too.
No matter what you do, it’s important to remember that a bot is always going to be a work in progress. If you think about how many times you’ve misunderstood a text message, you can see where a bot can run into problems. “Make sure you’re always checking in on what customers are asking your bot, and where they might be problems so you’re ready to make updates as needed,” Molly says.
Getting started with Einstein bots.
If you’re looking to get started with Einstein in your org, how do you get started giving your bots data? “There’s a number of sources: chat transcripts from Live Agent, case records that have ways that people are asking for these things,” Molly says. However, “people tend to talk differently when chatting with a human than with a bot,” she says, so you need to keep that in mind. “We’ve built the ability to have packages of data for intents that map to your use cases, which you can find on AppExchange,” she says, and there will be more on the way.
That said, it’s not going to be an easy process to perfect your bot. One of the differences between Salesforce and other bot vendors is that “we want Admins and companies to maintain control over exactly what the bot is learning from,” Molly says, “it’s important that you trust the data going into it.” They find that, ultimately, the work you put into adds more value in the future. Salesforce gives you the ability to specify things like product names or order codes to help it cut through the lingo associated with your specific product or service.
Thinking about ethics and bots.
Ethics in AI is also an important thing to keep in mind because you want to make sure that the data you’re giving your bot represents all of your customers. “There are two points to that,” Molly says, “is all of your data representative of all the different types of people who are asking things?” And, “is the way that you are using a bot providing different service levels to different customers in a way that you might not want to happen?”
This is why a bot is a constant work in progress. You need to be aware of what’s happening and constantly asking yourself if something like that is happening within your org. Curating data is a big way to help with this, and that’s why Molly and her team prioritize control, even if it takes a little more work.
We want to remind you that if you love what you hear, or even if you don’t head on over to iTunes and give us a review. It’s super easy to do, and it really helps more Admins find the podcast. Plus, we would really appreciate it.
Love our podcasts?
Subscribe today or review us on iTunes!