For this episode of the Salesforce Admins Podcast, we sit down with Marco Casalaina, SVP, Product Management, and GM of Einstein at Salesforce. We go over all things Einstein and what you should do with that all-important free prediction.
Join us as we talk about how to translate what you need to know into an Einstein prediction, how to know if you have enough data, and understanding what to do with the recommendations you get.
You should subscribe for the full episode, but here are a few takeaways from our conversation with Marco Casalaina.
Marco started at Salesforce way back in 2005, working on what is now the Service Cloud as one of the original developers before eventually moving into product management. “A lot of old school admins who are listening to this podcast might know me as ‘Werewolf,’” Marco says, which was his handle on the old Answers and developer.force.com discussion boards where he has literally thousands of posts. While Salesforce has gone through a lot of changes since his developer days, some of Marco’s original code is still a part of the platform. “Whenever you click to dial a phone number…I wrote that code myself, it’s still there,” he says.
Today, Marco is the GM of Einstein, an AI that adds intelligence to the apps you build and use in Salesforce. With over 45 different Einstein applications and platform capabilities available today, “but the philosophy behind it is that it should be accessible to a Salesforce administrator,” he says, “you shouldn’t have to know how to code to use Einstein and you certainly shouldn’t have to know how to use algorithms.”
If you’re like a lot of us, with all the Einstein options available it’s hard to figure out where to get started. Marco recommends taking a look at the pre-built applications. Many of these are included in the base Salesforce licenses, and even more importantly, they give you valuable insights on your data you can use right now. This includes smart features like opportunity scoring and case classification, a new feature in Winter ‘21 that can automatically read emails and classify them based on what it already knows about your previous cases. This feature is super easy to set up but can potentially save your organization tons of time, and it’s included in the base service cloud license.
To support you, there’s a new Prediction Builder module on Trailhead to help you get a handle on everything. “When it comes to Prediction Builder, you as an administrator—you’ve got to apply yourself a little bit,” Marco says, “you’ve got to think through what is it exactly that you’re trying to predict.” That means getting specific and boiling it down to KPIs you can put into the application.
“You can predict a yes or no question or a number so you have to reduce it to that,” Marco says, “then the next question is, do I have enough data to make this prediction and is that data in Salesforce?” Einstein needs a minimum of 400 records to make a guess, but with algorithms, the more the merrier. What you get is a probability, and it’s up to you to translate the results and figure out what to do about it.
So you get one free prediction to make the case to your organization that they should invest more in Einstein—what should you do? Marco highlights a few common predictions, including late to pay, prospect scoring, and the likelihood of churn (which is basically the likelihood of something not happening). Next Best Action is another powerful freemium feature that is only getting better, so check it out.
One of the coolest things about Einstein is that it’s constantly learning and retraining itself. Lots has changed in 2020 and businesses need to be nimble—you may find yourself selling to a new market, or in a new way that you hadn’t considered before. Unlike other lead scoring methods where you need to write a set of rules, Einstein automatically makes predictions itself based on the data coming in. It’s only getting more and more powerful as time goes on, so now’s the time to make the case to your organization that they need to get smarter.
Mike Gerholdt: Welcome to the Salesforce Admins Podcast where we talk about product, community, and career to help you become an awesome admin. This week, I am super excited to be diving into the incredibly interesting world of Einstein. Don’t be scared. This is one to listen to. Wow, this is fun.
So we are talking with Marco Casalaina, who’s the SVP of product management and GM of Einstein. He’s Einstein, right? This is neat. So let me tell you about what you’re going to hear in this episode. In this episode, you’re going to hear from Marco about one of the important questions I had. First of all, as a Salesforce admin, how do you start with Einstein? Where do I begin?
And then we hear freemium, you get one prediction free. So I asked him, “Where do I start? What should I do as my first prediction?” and Marco tells us where to go. And then, buckle up, got to stay around to the end and I’m going to tease it out now because he tells us what’s on the roadmap for spring ’21 and it’s cool. Like literally, I don’t have words. So with that, let’s get Marco on the pod. So Marco, welcome to the podcast.
Marco Casalaina: Well, thank you for having me.
Mike Gerholdt: This is your first time on. I feel like you’ve done some live social stuff with us, and I know you’ve been on stage, so admins have seen you, but refresh my memory, what’d you do before you became the overlord of Einstein?
Marco Casalaina: The overload, you’re the first person to call me the overlord of Einstein, but-
Mike Gerholdt: I got to pick new terms every time. I mean, it makes for a fun business card, right?
Marco Casalaina: That’s right. Well, yeah, so I’m the GM of Einstein now. So what did I do before? Well, I was at Salesforce before, really. So in reality, I was one of the original developers of what is now the service class. I was hired at Salesforce in 2005, 15 years ago, believe it or not.
Mike Gerholdt: Wow.
Marco Casalaina: As a developer. I was hired to write the CTI tool kit, our original telephony integration, stuff like that. And then I became a product manager. So a lot of admins who are kind of the old school admins, who are listening to this podcast might know me as Werewolf. So I was Werewolf on the discussion boards on both the answers boards and the developer.force.com boards, and I was super active, and I still am pretty active on the boards as a matter of fact. I have literally thousands of posts on the boards. You’ll still find Werewolf code and Werewolf posts here and there.
Mike Gerholdt: Wow. Okay. If we dig deep enough in the Service Cloud will we find Werewolf in the code?
Marco Casalaina: I never actually put it in the code, but it is true that my code though-
Mike Gerholdt: Missed opportunity, man. Missed opportunity.
Marco Casalaina: That may be, but I am pleased to say that my code itself though is still there. So even though the original CTI toolkit is now dead and gone with open CTI and all these kinds of things, whenever you click to dial a phone number, I wrote that code myself. It’s still there.
Mike Gerholdt: Well, that’s great to know. But we’re here talking Einstein now. So tell me a little bit about Einstein, because before we pressed record, it’s kind of a broad term, use it for a lot of things, but what does it mean in your world?
Marco Casalaina: All right. Well, Einstein is AI for Salesforce. It has a very specific meaning and it means that… Well, obviously it’s AI, but it’s not AI for everything. AI for Salesforce means that it’s constrained in certain ways. You’re not going to build a self-driving car with Einstein. You might do that with some other AI toolkit, that’s not what Einstein is about. Einstein is about adding intelligence to the apps that you build and that you use in Salesforce.
Mike Gerholdt: That’s pretty easy to remember, and we have a character to go with it.
Marco Casalaina: We have a character to go with it. And it’s true, there are literally 45 different Einstein things now, Einstein applications and Einstein platform capabilities that we have now. So over the last three years we’ve put together this bewildering array of Einstein capabilities and that can be a little confusing to folks at times, I understand. But there is an underlying theme and an underlying philosophy to these Einstein things.
And in particular, the philosophy of Einstein is that it should be accessible to a Salesforce administrator. When you look at any Einstein thing, the person who should be able to configure that is a Salesforce administrator, somebody who is familiar with custom objects and fields and flows and the declarative things in Salesforce. You shouldn’t have to know how to code to use Einstein, and you certainly shouldn’t have to know how to use algorithms to use Einstein. So I always say, Einstein is for people who don’t know an algorithm from a logarithm.
Mike Gerholdt: So of the plethora of 45 things to choose from, let me start off with the million dollar question of, where should admins begin with Einstein?
Marco Casalaina: Where should you begin? Well, I would usually start with the pre-built applications. And I say prebuilt and that has a certain meaning too, so we’ll get there in a minute, but every cloud has a set of Einstein capabilities that are pre-built. That is to say, that you don’t have to kind of form what you’re predicting or any of that stuff, and also that serve a certain need in that cloud.
And also, increasingly, we have these capabilities that are included in the base licenses that you have. And so, for example, in Sales Cloud, you can now request opportunity scoring to be turned on, as now thousands of Sales Cloud customers have done. So you can use the opportunity scoring, which gives you an idea of the likelihood of an opportunity to close. The predicted likelihood versus just the reps probability. Rep says 75% opportunity scoring says 20%. Uh-oh, something’s wrong there.
Mike Gerholdt: Oh, well the reps always look, they… I’m going to close this. Don’t worry.
Marco Casalaina: That’s right. Sandbagging prevention, right?
Mike Gerholdt: Seriously. They’re all going for the steak set.
Marco Casalaina: Absolutely, and it can help guide and prioritize the time of sales management, but in Service Cloud, one that I really want to point to here is case classification and case classification in the winter release. So this upcoming release right now is going basically fully part of the Service Cloud-based license. And that means that it will be able to make real time predictions on that. And what does it do? Well, [inaudible] it classifies cases. So let’s say that you have email to case, which incidentally I myself was the product manager of back in the day.
Mike Gerholdt: I have used that.
Marco Casalaina: I certainly have myself too, and whenever I see one of those little ref tags in an email, I’m always like “Uh huh, it’s email to case.” But when you think about email to case… By the way, case classification is not just for email to case, but that’s one of the primary uses of it. When you get a case via an email, you may know who the person is, and you may not. You might be able to do an identification of a contact and the account, but the bulk of what you know about that case is in the subject and the description, and really nothing else in the case is going to be filled out.
So a lot of businesses have people that their whole job is they just sit here and they read the cases and then they route them. They are human routers basically. They read them and they’re like, “This one’s billing, and this one’s shipping, and this one’s whatever” and they route them. These people could actually be providing support, not just sitting here all day and routing cases around.
So case classification uses artificial intelligence to read. It learns to read your cases from your previous history of cases, you need at least a thousand cases to have been opened in the last six months, so it does need data to learn from, but it learns to read those cases then it sets these categories. So it can say this one is shipping and this one is billing and whatever, and then it can route those cases based on its own prediction.
And so for something like email to case, where you might’ve just routed that to a general queue before and somebody had to read it and send it along, or you set up some funky process builder thing that looked for keywords in there and tried to do it in some junky way. Case classification is much, much easier and it’s super simple to set up. It really doesn’t require a lot of mental load to get that set up. It’s super powerful, and when people use it, we have thousands of users now of it, and it’s included in Sales Cloud.
Mike Gerholdt: So I’m bought in, and we’ve demoed Einstein Prediction Builder in the admin keynote before, and we’ve used it ourselves. I get it. It’s super easy to set up. In fact, it’s probably easier to set up than a process or a flow. So if you understand that, you can set up prediction builder. What I want to know is I want to help admins navigate the world outside of the setup screen, who do I start talking to? And what do I need to start telling them, or informing them about the way that Einstein and AI can work for us as an organization? As team admin here, it’s cool, and I know the demo’s got my back, it’s just I’ve got to be in the here and now and convince them before I even show them the screen.
Marco Casalaina: Absolutely. And incidentally, I get a point to Trailhead. We just actually just put up last week a brand new trail on prediction builder that covers some of the new capabilities that you find in there. We can make new, more granular types of predictions and things like that, that are covered in that trail.
So the Trailheads are getting, I would say closer and closer to reality in terms of how you use them in terms of how people really use them in the real world. We’re trying to make them mimic what people are really doing. And so the latest trail focuses on, one of the most common, actual predictions that people are making, which is, are you going to pay your bill late or not? It’s a very common prediction that’s also the subject of the new trail.
That aside though, so prediction builder is interesting, and earlier I talked about case classification, opportunity scoring, and there’s other these kind of out of the box capabilities like engagement scoring from marketing cloud and [inaudible] cloud recommendations. Those things anyway are kind of out of the box, but when it comes to prediction builder, you as an administrator, you got to apply yourself a little bit. You really got to think that through, and also with your partners from business. You got to think through what is it exactly that you’re trying to predict?
It so happens that I was on the phone with a customer just this morning where we were just going through this exact decision tree. I was going through that with them. They were interested in making some predictions, they kind of got that concept. They’re like, “We would like to predict.” And their case, it actually has to do with work and work items, and is this work… and they model their work items and user stories in Salesforce, and it’s kind of like, is this thing going to happen on time or not?
So there’s a number of questions that you want to ask before you even get into prediction builder that are sort of meta questions about the prediction. And some of them we’re going to get to, as you say, what do you do with it after the fact? But it starts with this; what is the thing that we’re trying to predict exactly? And what does it reduce to? If you think about what prediction builder does, and by the way, just about everybody who’s listening to this has prediction builder already because that’s freemium, so everybody gets one prediction for free.
Mike Gerholdt: I know. I have a follow-up question on that.
Marco Casalaina: Okay, we will get there. We will certainly, we’ll get there. But in any case, so the first question you have to ask is, okay, what is the thing really that I’m trying to predict? And for example, the customer I was talking to this morning, they were like, “We would like to predict the date by which this piece of work will be done.” And I sit up, but that’s not exactly how it works. The way it works is you can predict a yes or no question or a number, so you have to reduce it to that. But you can’t ask it in that way. Yes or no. I say this thing is going to happen on October 8th. Yes or no, is it going to be late or not? Which is similar to, are you going to pay your bill late or not?
So you can ask that type of question, or you can ask, how many days is it going to take us to do this work item? Is another kind of a question that you can ask prediction builder. So the first step here is really to think about what you’re predicting and how you might be able to phrase it in the form of that type of question, a yes or no question, or a numeric question.
The next question is, do I have enough data to actually make this prediction? And is that data in Salesforce? With something like prediction builder, the data has to be in a Salesforce object. There is no alternative there. So you have to bring that data into Salesforce. And that’s required because of how prediction builder works. The reason prediction builder doesn’t ask you many questions, the reason you can get through it like that so quickly is because it relies on the fact that the data is already here and that the metadata is already here. So it uses that metadata about the fact that this is not just a, you know, a string of text, this is an email address. This is a postal code. This is a state.
Those things are really necessary for machine learning to work. But fortunately, just by virtue of making custom fields and custom objects, you’re doing that, you are implicitly telling it what they are. So again, the question is, is the data that I need for this, is it in Salesforce? And do I have enough of it? Now, the minimum requirement of data for prediction builder is 400 records. So it needs 400 historical records, things to learn from. So as what this morning with the customer I was talking to, they needed at least 400 work items, historical work items that were done that were closed or whatever, to make that prediction.
But 400 is a really weak lower bound. So you really want, let’s say a couple of thousand and you want to try to do that. And that really applies not just to prediction builder, but to all of the Einstein things. I mentioned it earlier with case classification tool, you need at least a thousand cases in the last six months for lead scoring, opportunity scoring, they need a certain number of opportunities and leads so I don’t remember the exact number right off the top of my head, but this stuff is not magic, right? It learns from your data because that’s what’s happening underneath the scenes is that it’s building predictive models specifically for you, specifically on your data, and your data only. That’s how these things work, but they need your data to work with to make it happen.
And then the question is, okay, let’s say you ask a yes or no question, for example. Let’s say you ask us as the customer I was talking to this morning, is this workout I’m going to close on time? Let’s say that they do have enough data in Salesforce to make this prediction, the response to a yes and no question comes in the form of a probability. So it’s not going to say yes, this thing is going to be late, and no, this thing is not. It’s going to be like, this thing has a 30% chance of being on time. That’s what you’re going to get there. You’re going to get a probability.
And so then it’s up to you, the administrator or your business partner, or whoever, to determine what are your thresholds and what exactly do you do about it? And that may involve rules that you make in flow or in the Next Best Action, which by the way, is another thing that is freemium that you all already have, but you may want to make rules that include this prediction. And a prediction with production builder’s just stored in a field like anything else. You can report on it, you can make let’s view on it, whatever, and you can also make flows and stuff on it. And so you may want to make rules that define your threshold and what specific action you’re going to be taking based on the probability that’s predicted.
Mike Gerholdt: Well, that makes sense. And I could also understand that different businesses might be comfortable with different thresholds, right?
Marco Casalaina: Absolutely.
Mike Gerholdt: All the time.
Marco Casalaina: Yeah.
Mike Gerholdt: So you mentioned freemium and that was actually one of the questions that I had was, I’m thinking as that Salesforce admin, okay. I understand from the business side, I think what we need to predict, I’ve looked at the records, I’ve got a minimum 400 and feeling good about it. What’s one way that I can use that first prediction? Because I feel like if I nail this first prediction, or I give that business value back to the organization, they’re going to turn around and be like, “Mike, we need to talk about Einstein. We need to get more predictions, and we need to really invest in us.”
Marco Casalaina: Yeah, you would not believe the diversity of things that people are using to predict now, or predictions that people have made now. We talk to customers, I feel like we see a lot of different things that people are doing, but some of the most common ones, as I said earlier, one of the more common ones is late to pay. I’m very surprised how many people run their billing out of Salesforce. I didn’t know this going into predictive builder. And so for a lot of folks, including Salesforce itself, by the way, Salesforce actually uses this ourselves, our own accounts receivable is making this very prediction about lateness to pay, and if you’re beyond a certain threshold of predicted lateness, then they might reach out to you and be like, “Hey, is everything okay? Can we arrange a payment plan or something like that?
So they’ll proactively reach out to you to ensure that payment in some fashion happens or that they can work with you about it. That’s a common one. Another really common one is, I talked about, it’s kind of these out of the box applications like opportunity scoring and lead scoring, and stuff like that. And those applications, they rely on the fact that you use a standard process. Like you use a standard opportunity process, You use a standard lead process, but there are lots of businesses that don’t do that. For example, I’ve seen a number of businesses that actually front the lead object with another custom object, which is like a prospect object. And so it actually is a three tier to prospect to lead to opportunity. And they really want to do prospect scoring. But of course, lead scoring out of the box is not going to do that because it’s lead scoring.
But you can use prediction builder to build predictions, and that’s quite a common use case also is that people who use non-standard processes for opportunities or leads, they might use prediction builder to generate that score. Ultimately, it’s the same underlying platform, it’s actually doing basically the same thing, but you’re telling it, “Oh, this is actually my lead object over here. I’m using this prospect object. So tell me, yes or no, is this prospect going to convert or not?” Is the yes and no question that that really is, that’s the fundamental question that lead scoring is asking. Every prediction has an underlying question, whether it’s an out of the box thing or a prediction that you build, there’s always an underlying question there.
So that’s another one. Another one that people sometimes do is variants of what you might call churn or attrition. So that can be, I’m seeing a lot more recently of, let’s say a likelihood of no show, let’s say, so like in the medical field, now that we have a compliance, we’re starting to see how likely is it that you’re not going to show up for your appointment or that the customer’s not going to be home when we go roll our field service truck to their house or whatever like that. So likelihood of no-show, or how likely is it that this student’s going to drop out of our program is actually a real thing that [inaudible] did. So likelihood of something not happening, I guess, is another type of prediction that people mimic.
Mike Gerholdt: It makes me think of just how nimble companies need to be, especially now, because if you think of maybe you’re using likelihood to be home, and you’ve got data from 2018, 2019, all of a sudden 2020 rolls around and there’s a… I can’t think of a day that I haven’t been home to get a delivery. Like all of a sudden now I’m thrown off the curve.
Marco Casalaina: And here’s an interesting example. So if you think about lead scoring, and particularly lead scoring. So there’s an Einstein lead scoring, and at the same time, there are other lead scoring capabilities out there. There’s one in [inaudible] there’s one in Marketo, there’s one an Eloqua, so there’s lots of lead scoring capabilities that people have been traditionally using. And most of those traditionally have been rules-based. So somebody went in there and set up these rules that say, “If you’re in Kentucky, plus one. If you’re in West Virginia, minus one.” Or whatever, and they set this elaborate set of rules.
The problem with that is, well, first of all, it’s hard to guarantee the accuracy of those particular rules because somebody came up with them in some probably data-driven way, but ultimately it’s hard for a human to get their head around all the data. But also the fact that those rules tend not to evolve very fast, unless somebody is dedicated to changing their lead scoring rules every week or whatever, it’s not going to necessarily keep up. And so you think about, there is a company called Reverie, and Reverie makes mattresses and stuff, things to sleep on.
Mike Gerholdt: Sure.
Marco Casalaina: And during the pandemic, well, for one thing, they have traditionally been direct to consumer, so they’ve been selling, basically you go to their website and you can buy a mattress directly from them, but during the pandemic, they actually pivoted the business to some degree. So first of all, the direct consumer business actually grew because of the fact that I guess that everybody’s spending more time at home, but also they created an entirely new division, which is hospitals; selling mattresses also now to hospitals. And that’s a totally different type of buyer, but they’re using Einstein lead scoring, and Einstein lead scoring retrains itself. Like all the Einstein things, they retrain themselves periodically, relearn from the data.
So as the data changes, it will pick up effectively these new kinds of “rules”. And so for Reverie, their lead scoring gets more accurate even as their business changes because of the fact that it’s learning. So as it sees these leads coming in from hospitals and things like that, at first it’s not going to know what to do with that, but within relatively short order, it’s going to figure that out and be like, okay, it starts to be able to discriminate between good hospital leads and bad hospital leads in a way that a rules-based system really wouldn’t be able to do or not quickly.
Mike Gerholdt: Because somebody is not sitting hovering over it, staying up late at night. That’s what I envision Einstein does.
Marco Casalaina: Sort of.
Mike Gerholdt: Constantly burning the midnight oil man.
Marco Casalaina: Einstein literally tries every possible combination of values. That’s really what it’s doing. It’s a giant correlation engine. It’s finding the correlations between, as I said, what state do you live in? And whether you’re a good lead or not. All kinds of things and these can be hundreds or even thousands of different factors. It’s going to find these correlations and balance them against each other. And that’s fundamentally how it works.
Mike Gerholdt: You mentioned Einstein’s Next Best Action. I feel like I’ve had to say that in a keynote.
Marco Casalaina: Next Best Action. Yes.
Mike Gerholdt: Next best… It’s the way you come off the end, it’s the two ends.
Marco Casalaina: A little bit of a contradiction in terms if you’re thinking about it too like, well, what’s the first best action? I want to know about that.
Mike Gerholdt: I know you mentioned that, and I definitely wanted to bring that up because when we demoed that, that was really cool. And if you had any idea is to kind of give me the vision on what’s coming.
Marco Casalaina: I certainly knew. I got some stuff for you on that.
Mike Gerholdt: Forward looking statement here.
Marco Casalaina: This is a forward looking statement, I’ll make a present looking statement, first of all, that next best action is there today. This is something that you can use. It is freemium, it’s in your org already, you definitely should check this out. This is a thing that can be very powerful, it’s effectively a real-time rules engine. You can put these rules in there about what kinds of actions you want to take and you can sprinkle in there predictions also.
And you can make a rule that says if you’re more than 30% likely to be late paying your bill, then we are going to call you up or something like that, there’s place to do that. That kind of thing. But what’s coming for Next Best Action is Einstein Recommendation Builder, which is going into beta in the winter release. And so Recommendation Builder.
Now, you think about, this is the thing that Salesforce has lacked, well, since my first time at Salesforce back in 2005, 2010, which is we really need a way to recommend something. And so what Recommendation Builder is, fundamentally is the ability to recommend object A for object B. And already, we have lots of customers in that pilot already and they’re doing all kinds of stuff.
There are folks that are recommending, for example, like what parts should be on the truck? When we do a truck roll with a field service. There’s other folks that are doing, what colleges should I recommend to students? And what bank products should I recommend to my customers? Is another one that is also happening with one of our pilot customers. And this all integrates to Next Best Action, because the recommendation engine will give you these recommendations, but you still need rules. You need rules to determine things like eligibility, inventory. Of course, you could do product recommendation with this also. Maybe I might recommend to you this volleyball that I have, but before I do that, I have to ensure that is the volleyball actually in stock right now? Or are you in a place that I can ship the volleyball or whatever?
The raw recommendation engine doesn’t necessarily output actionable results. You have to put it through these rules and that’s what Next Best Action is. So what you’re going to see is that the Next Best Action today has a load block, and basically it sort of loads the set of recommendations and that kind of whittles those down with rules, eligibility and those kinds of things. There’s going to be a new type of load block that’s going to show up when you have Recommendation Builder, and that’s the Einstein load block. It has a little Einstein head on it. And in that type of load block, it uses a recommendation model to make these recommendations, but you still pipe them through rules in Next Best Action just as you would otherwise. Because again, those rules always matter.
Mike Gerholdt: That is so cool. And I was just trying to think through like how important recommendation would be, because I used to be in sales. And the good salespeople are the ones that know how to put those rules together in their head on the fly, and the not so good salespeople were the ones that didn’t.
Marco Casalaina: Sure. And think about that too for Service Cloud, right? I mean, think about the ability to make recommendations, make upsell offers from your customer service center. Somebody calls you up, you have the opportunity to make an upsell, for example, or even recommend them the action that the user should take. The restriction with Recommendation Builder is that again, this is recommending object B for object A here.
So these things have to be in Salesforce objects, and you’ve got to have a three object layout here. So you have your recipient that might be, let’s say, contact, you have your item, so that’s the thing that you’re recommending, and then you have your interaction object, and that’s the object that stores historically, these are the people that did take these things, or these are the work items for which I use this part, or these are the students that [crosstalk] this college. But whatever that is, that too like prediction builder, the data has to be in Salesforce.
Mike Gerholdt: Wow. Preaching to the choir. And then even forward, we’re heading into the winter release, but I know before not too long, we’ll be talking spring. Do you have anything interesting coming in spring?
Marco Casalaina: Yeah. And you know, some folks have already been experimenting with our OCR stuff. So optical character recognition is coming. And while a little bit it’s already here, some of our APIs are already exposed and some folks like a Force Panda and [inaudible] saying have already written blog posts about how they’re using this stuff. So you can find that online already.
But the goal of this OCR stuff is not to be an API. The goal is to be baked right into Salesforce. And so you [inaudible] and it sucks it right into an object in Salesforce. And no Blogger JSON, no code, no nothing required there. So for now, we’re exposing just the APIs and we’re doing that bit by bit, those are coming out and people are starting pick those up. But what you’ll see soon enough is a truly declarative interface. Again, the goal is that an administrator with no coding experience, with no algorithm experience can set this thing up and bring it to bear in their business.
Mike Gerholdt: Well, that’s both good and bad news for me. The good news is it sounds really cool and it’s something I’ve always wanted. The bad news is a long time ago in a few previous jobs, I set up an April fool’s joke where I put a fake HTML box on the homepage and told my business users to hold a business card up to it and press a button because Salesforce was pilot testing a new screen scanner to scan data into Salesforce. And when they pressed the button, it took them to the Wikipedia page on April fool’s jokes.
Marco Casalaina: Nice.
Mike Gerholdt: So the good news is that this is actually coming to life. We still don’t have jet packs. We still don’t have flying cars. The bad news is that my April fool’s joke was actually just kind of very forward-looking.
Marco Casalaina: It was science fiction at the time, but it’s funny you should mention that, but we put this on hold just for the time being, because of the pandemic, but we also have in pilot, a business card reader for the mobile app. You can take [crosstalk 00:28:41].
Mike Gerholdt: Oh my God.
Marco Casalaina: … into a contact or a lead record. Unfortunately, we’re not in a place where people are exchanging business cards, but hopefully soon that comes back.
Mike Gerholdt: Soon. Marco, I want to thank you for being on the pod. We got to have you back when it’s optical character recognition on some video form, because I think this is going to be one of the cooler demos.
Marco Casalaina: Oh, absolutely. I’m happy to come back to that, and anytime man, we’re cracking away with Einstein. We’re doing new stuff all the time. So anytime.
Mike Gerholdt: You bet. So it was great having Marco on the podcast. I’m so glad he took time out of his day to educate admins who’s not excited about optical character recognition. Like, does that not sound cool? I’m telling you. It is such a cool time that we live in. All right. So a few things I learned in our discussion with Marco; first, with Einstein, it’s like figure out what you’re trying to predict, and what does it reduce to? So he gave us a very good example, is it yes or no? Or is it a number? So think of it like, will you pay your bill? Or number of days to complete a task.
So I think that’s important because we’re not trying to predict everything, so let’s predict it down to this. And the second thing, is there enough data? Do we have at least 400 records is the minimum, but really like a couple thousand good records that have gone through this process that Einstein can sit and stay up late at night and burn the midnight oil and do a dude’s homework and really understand what’s going on.
And then thinking through the response which comes in the form of a probability of yes or no. So as a company, what’s our threshold going to be of that, yes, right? And those are all fun things to sit down and work through with your leadership team as a Salesforce admin, try things out, I highly recommended it. And the second thing, man, am I excited to learn more and to get my hands on Einstein Recommendation Builder?
So remember, we’ve talked a lot about Einstein Prediction Builder as a Salesforce admin, but now it’s recommendation. So how do we help our reps, or how do we help our customer service people recommend something for something else? Like an upsell or a cross sell. I think Marco used the example of like parts for a truck. That just sounds super cool because I’ve done sales before, and to have that ability of knowing how to recommend or what to recommend, and knowing that this person fits that criteria, that’s just another tool in your tool belt. So very excited for that.
It was just a neat episode, and I hope you sit down, I’m going to link to the Einstein portion of Trailhead so that you can go through, I hope you spend time with this. This stuff, it feels like jet packs and flying cars, but is so within our realm to do and to sit down and have those leadership discussions and really move our business forward and really be leaders within our organization, I’m just excited about.
So I want to help you learn more. And to do that, I want you to go to admin.salesforce.com. So, that’s where we host this podcast. We have amazing blog posts there. We link to things that we’re doing like on Trailhead live, so you can watch essential habits videos, just a plethora of things, lots of resources. And I want to remind that if you love what you hear on this podcast, pop on over to iTunes, give us a review, shoot some stars. What that does is it helps other admins find the podcast. So if they’re searching for Salesforce or if they’re searching for admin, then this podcast pops up and then they can hear amazing stories like Marco talking about how he started at Salesforce and building Service Cloud, and now is helping do Einstein recommendations. This is all really fun stuff, and I want more people know about it. So that’s really cool. And we read them all, I promise you, maybe even tweet them.
You can stay up to date with us, all things social for Salesforce admins. We are @Salesforceadmns on Twitter, no I, you can find our guest, Marco, he is also on Twitter, A-M-R-C-N_werewolf. If you think back, that’s where he got that from. I’m also on Twitter. I am @MikeGerholdt and Gillian who also hosts podcast is @GillianKBruce. So with that, I want you to stay safe. I want you to stay awesome. And of course, I want you to stay tuned for the next episode. We’ll see you in the cloud.
Today on the Salesforce Admins Podcast, we talk to Lizz Hellinga, Consultant and Salesforce MVP. Join us as we chat about why product management principles in Salesforce are crucial if you want to take advantage of new AI tools. You should subscribe for the full episode, but here are a few takeaways from our conversation […]
Today on the Salesforce Admins Podcast, we talk to Tom Leddy, the Product Director of Decision Guides at Salesforce. Join us as we chat about decision making in the age of AI and why cleaning up your data is more important than ever. You should subscribe for the full episode, but here are a few […]
Today on the Salesforce Admins Podcast, we talk to Lisa Tulchin, Senior Curriculum Developer at Salesforce. Join us as we chat about choosing the learning path that fits your learning style and strategies for training your users. You should subscribe for the full episode, but here are a few takeaways from our conversation with Lisa […]
