Einstein Prediction Builder and Einstein Next Best Action for Lead Curation

Welcome to the X Force Data Summit. Today, we have Sika Bide, who's a team lead at Accenture Accenture Salesforce practice in India. She's gonna talk about Einstein prediction builder and Einstein nest next best action, which I can never say correctly. And she's gonna she's gonna do it to talk about leads. So before I screw anything else up, here's Sika.

Okay. So I agree that next best action is kind of a tongue twister. So yeah. Okay.

Hello, everyone. I am Shikha Bhed. With over five years of experience in Salesforce, I have taste of both products and services. And I also lead the Surat India Women and Tech Group.

Okay. So have you imagined ever how would things be if you could predict them?

Yeah. I agree that it would be exciting. And now you can do that for your business. You can do it with Einstein prediction builder and the next best action to curate an action plan for your leads.

So let's go ahead with the session content. So the key takeaways from this session would be the Einstein prediction builder setup and overview, Einstein next best action setup and overview, a demo walkthrough in both of the features, and last but not the least, the considerations which needs to be kept in mind for prediction builder and the next best action.

Okay. So Einstein prediction builder is basically AI for admins.

Okay. So what this lets you do is predict what happens next in your business by analyzing the past data.

So the first time you enable your Einstein prediction builder, there are two ways you could do it. The first is clicking on get started on the Einstein prediction builder tile on your home page, or you could navigate through setup and then get enable it. You also have an option to toggle the prediction builder on and off as and when needed.

So before going ahead with Einstein prediction builder, we need to be familiar with the couple of terms. I'll start with dataset. So dataset is basically the object for which you are building prediction for.

Out of the object, the segment is the subset of your data, which you would be taking as an example and prediction. So when I talk about example set, it means that the records which you'll give Einstein to analyze and build predictions on. And prediction set is the set of records for which the Einstein will build predictions. So your example set and prediction set together becomes the segment.

Now moving next to the next best action. It is all about displaying the right recommendation to the right people at the right time using the recommendations.

Now before going ahead, what is major with Einstein Best Action is your strategy. That is your business logic that where you decide what recommendation to show for what kind of leads or, in in this case, leads. So that could be another object for you.

So yeah. So the interface has got three tech tabs. One is the elements, manager, and inspector. So for elements tab, anything and everything you could drag to your canvas and build your business logic is in the elements tab. The manager tab allows you to connect and bring data from external data sources or other Salesforce products. And inspector tab is something that you could troubleshoot about what your path the record will take or what your path the recommendation will take.

So this is the strategy builder interface tabs. Now for next best action, we have three important things. One is flows. The second is recommendations, and the third being strategy. So all three are mandatory for this.

So when you create a flow, you basically make a flow make a business flow for your recommendation. So what path does your recommendation takes on acceptance is what your flow could be. It could be about offer it could be about offer. It could be about discounts.

It could be about anything you want to offer to your customers. Now recommendations are something which is used to surface these flows via strategy to the record. So recommendation is something that uses flows as its core. So without a flow, you will not be able to create a recommendation.

Now when you have created recommendations, you'll have to go ahead and set up a business logic, which is called as a strategy in next best action, and where you define everything what happens to your recommendations.

So yeah. This is it. Flows, recommendations, and then strategy, and you are ready to get go for the best action.

Now one thing is that next best action gets invoked for a couple of instances.

Like, you have loaded a page with which has the Einstein next best action component, or you have made the changes to a field which makes the next best action to trigger again. So there could be multiple scenarios where this gets triggered, and your recommendations might change as your record changes.

Okay. So before going ahead to the demo, we have let's understand the use case. What we will be predicting here is the conversion probability of the lead. The preconditions are the Einstein prediction builder and NBA enabled in your org. Okay. So when I say NBA, it's next best action.

You have and you have to make sure that you have enough records to build a prediction because there is a minimum record requirement to build your predictions.

Now where do we see these recommendations and actions and predictions coming into picture? It will be the lead detail page in this case. And how do we do that? We'll see that in walkthrough.

Okay. So now what the user does on the screen during this demo. Is particular to this demo. So user navigates to the lead detail page where already the Einstein Next Best Action component would be residing.

Now here, the prediction builder has run its prediction and populated the prediction on the record.

The next what happens is the next best action, depending on your strategy, shows up the recommendations.

You can do two things with your recommendations. First is accept and next is reject. As soon as you accept it, the flow which you have linked to that particular recommendation would be shown up and it will take that path.

If you have rejected it, then depending on your strategy, either another recommendation comes up or else it will just go blank. So it completely depends on your strategy, how your accept and reject actions would work.

Okay. So let's walk through the demo.

Okay. Starting with Einstein prediction builder. As I said, there are two places you could do it. First is the home page style of Einstein prediction builder where you see the get started button, and you can click on it. Or else, you can go ahead and type Einstein prediction builder and you get the option here.

Okay. So now it opens up a screen which has all your predictions. You have defined till date in your org.

So yeah. Because this is a developer org, so the total number of prediction allowed at ten. And the number of and out of those ten, how much could be enabled is one.

Okay. So this list all the predictions I have built. So it lists your name, what I have tried to predict, what is the scorecard of this prediction, and what is the status. So when we look at the scorecard, we see two, three things here. There are couple of more, but then I'll go ahead with this. So great. It means that you have enough data to enough and quite a diverse data to predict anything out of this particular prediction.

Now when it says not available, it is still running the prediction in the background. And as soon as it is enabled and your prediction is ready, you will see something called ready for review on the status field.

So when it is ready for review, you understand that your scorecard is ready. And after observing your scorecard, you can just go ahead and enable your prediction. So let's see how does a scorecard looks like.

Okay. So, yeah. It gives you a summary of how your prediction looks like. So, it's too high for this one. Now, here is a prediction quality meter where you see.

This is where you see what predictions you have made like which object, what field, when did you last update it. And here are all the predictors. So when we say predictors, they are nothing but the different fields on the object which you have selected. That is the heat here.

So this is the score card, and I have enabled one of the predictions. Let me go ahead and open it. Because it is enabled, I cannot edit it. I'll just go ahead cloning it.

Okay. Here, you give a unique name to your prediction.

Go next. And, yeah, as you remember, we talked about dataset. So lead is the dataset here. You have selected lead as an object here. Now next comes the segment.

So when we talk about segment, here are two options. Either you have no segment. So when you say no segment, it means that all records from that particular object could be used for prediction.

Now there is another option which says focus on a segment.

So, yeah, there could be a scenario in particular to lead that. You might want to predict only for only the leads of a few companies. You might have hundred of companies in your org, but then you wanted want the prediction to be enabled for a few of them. So that is where you could enter your condition, and then the prediction will act likewise.

Okay. So for now, I have selected all the records for prediction. Now clicking on next, what kind of question does your prediction have would be yes, no, or number. So yeah.

So for us, it is we are going to predict if your lead will be converted or not, and this is why it's a yes, no question. We'll go ahead with the yes, no. Now clicking on next, you would be asked if there is a field that could answer if how a prediction happens. So, yes, in this particular scenario, do have we have an is converted field on lead, which, pretty much, you know, suffice to be a field for prediction questions.

If you have no fields, then you can just go ahead and add add filters to your data, which would then be treated for prediction.

Now going on next, as I said that we have a field to predict. That is why we have selected is converted here. And now comes your example set. Okay?

So example set is something which prediction builder will use to analyze and study the data, and it will build prediction on this data. So here I have a condition where I have mentioned it as per the as per my business need. So it could differ from it it must differ from your business to business. So here is your example set.

Now comes next. Okay. So here are all the fields listed for that particular object. That is the lead here.

Now the option to add or remove a field from your prediction is because whatever you include here acts as a predictor for your predictions that are going to be built by the prediction builder.

So there could be fields which could be misguiding, some revenue or something that gets updated after you have converted your lead. So all that could be misguiding. For us, it's none. So it's okay to get all of them selected. So it's just to make sure that you do not include any misleading predictors in the set.

Okay. So now where did where does the prediction stores? Okay. It has to be somewhere on the record. So prediction result. Okay. So it's asking for it's asking for a field label and a name, which Einstein will go ahead and build a number field in the back end that is on your lead object with the same name and the label.

So yeah. Clicking on next.

Now here you can review all what you have entered. Your prediction name, your dataset, your segment what you have selected, what you are trying to predict, what are your number of example records here, and what are the predictors you want to use for that and where would you save the result. So this is the complete configuration of Einstein prediction builder. Now there is an one interesting thing here, is called the data checker.

So data checker basically lets you understand that is your Einstein prediction builder rule ready to be built or not. So I'll click on check data. So as soon as you check data you see the number of records it has for prediction. So, basically, this is the example set that is three four one five records here and one seventy nine records to predict.

So as you see, there are four hundred minuteimum records on which Einstein can build its prediction, and there is mandatory that there has to be one record to predict for Einstein. So this is where data checker allows you to see how good your data is. So it has to has have both the positive and negative kind of data to predict accordingly.

Okay. So let's go ahead. We have already built the predictions, and this is where the where I have just built a list view to show my predictions.

Now here are the prediction scores, which depend on the conditions which I have mentioned on my prediction while building it. So the prediction score changes as you see thirty eight, thirty four, thirty. It will go below, above, beyond. Yeah.

It completely depends on how you have built Now let's move to the Einstein next best action. So it starts with a flow there. So let me open up flow for you. Yeah.

So this is the flow. Basically, here are two tabs, elements and manager. So elements would give you anything and everything you could drag to your canvas and start building a flow. Manager would list all the resources which are already on this canvas.

So it could be variables, it could be templates, it could be some action, or anything, whatever is on the canvas is here.

Okay. So let this is basically a screen flow. So I'll just go ahead and open it.

It shows a header. There is some displayed text and it is asking an input for an email address.

That is what the screen does. Now what does the screen do after getting the email addresses? It will send the offer email.

So we have configured the offer email here, offer email action here. So the body has been set, what template it should use, what should be the subject line, and here is what we get as an input from the first screen. So yeah, this is the action that would get performed after we enter our email address on screen.

Okay. Now we are done with the flow. The next thing what we have to create is the recommendations.

So here are the recommendations. So basically what we have done is we have created one recommendation for each of the flow.

Now let me open a recommendation. Okay. It has the name and description. With it, it has got the most important thing is accept and reject.

So accept is basically the it it lets it allows you to select the flow which you have already built in the step one, and it shows the flow's status here if it is active or not. Okay? So one interesting thing here is you could create a recommendation with an inactive flow also. It does not restricts it from selecting.

But then your recommendation would not be surfaced using strategy to the record if your flow is inactive. So just make sure that your flow is inactive if you want your recommendation to be surfaced.

Okay. Now moving ahead, we are all set to create a strategy. So we'll go ahead and create a strategy.

Okay. So this is the screen where you create strategies and where all your strategies are listed. You can just type next best action here in your quick setup. Yeah. Okay. I'm sorry.

Okay. And you'll be navigated to the screen.

Let me open the strategy.

Okay. Yeah.

As we talked about in the slides, we have three tabs, elements manager and inspector. Elements has got all that you can drag to the canvas. So it is pretty much similar to the flow builder.

So so you can see the strategy builder is like a flow builder. It's just that a few elements are different.

Now next is the manager. You can create a connection here to get data from some external systems, or you can get it imported from other Salesforce products also. Now inspector looks like this. Here, you can pass your lead ID, and you can test how your lead would behave or what path does it take while, you know, entering this particular strategy.

Okay. Now let's see what is on the canvas. We have three recommendations, high, medium, and low. Let me open the high recommendation. Okay. For this particular strategy, if my recommendation description contains fifty percent often conversion, it will be considered as high recommendation.

Similarly, for medium recommendation, the condition is that description should be twenty percent off on conversion. And similarly, for low, it is ten percent.

Now the recommendations have been loaded on this this particular stage.

After load, what should happen to each of them? Because we are getting inputs from three parts. So definitely there there could be a difference in their path how they are taken forward. Right? So that is defined here. So here is the actual logic that resides in your branch.

Okay. So this is high recommendation, medium recommendation, and low recommendations.

So when it comes to high recommendation okay. As we have seen that here, we are using the prediction score. So prediction score is nothing but the field which we had configured to be populated in Einstein prediction builder.

So, yeah, if the prediction score is greater than thirty five, then my strategy con considers it as a high recommendation.

If it is somewhere between twenty five and thirty five, it is considered as a medium recommendation.

And if it is less than thirty five, then it is a low recommendation. So to sum it up, if on a record the prediction score is greater than thirty five, then it will show a fifty percent off because as we have seen that our high recommendation means fifty percent off. Similarly, if it is between twenty five to thirty five, it will show a twenty percent off recommendation, and if it is low, it will show a ten percent off recommendation.

So, yeah, we are ready with our strategy as well.

Okay. So there is some interesting element which is like elementary offers. Okay. How many times should the recommendation appear and on what conditions should a reoffer happen?

So for this particular scenario, I have configured it to appear only when the user has rejected the recommendation. So if a user rejects a recommendation, once then in a period of one day, that is twenty four hours, it could be reoffered. So till twenty four hours, when you once you have already rejected the recommendation, you won't see any recommendation on that particular record. Post twenty four hours, it will be visible there.

Okay. Now we are done with all the setup. We are ready with our price predictions also. We are ready with our strategy as well. So let's see what it does to our records.

Okay. So as you see here, the prediction score is thirty eight. And this is the next best action component, which is already on this particular page, and it is showing a fifty percent con conversion. So my strategy is working perfectly fine and showing me the correct recommendation.

Now next comes the prediction score which is between twenty five and thirty five and as expected it is showing twenty percent off.

And a next one where which is less than twenty five or just no prediction score yet it shows a ten percent off on conversion. So my strategy is working absolutely fine here.

So let's go ahead and accept recommendation.

Okay. As soon as we accept it, it will pop up with the flow because it was a screen flow, so it has popped up with a screen where it is asking for email address.

I'll go ahead and enter my email address.

Now as soon as I enter the email address, I'll have to I'll see that it says accepted here. Also, I'll receive an email.

Okay. Yeah. So this is the email where it says that you have been provided with fifty percent off for next four hours. So you could I mean, you could configure it as needed. I've kept it simple.

Yeah. So my first strat the high recommendation strategy, the recommendations, the acceptance, everything worked fine.

Let us do it for the second one. Let me reject it. Okay. So no action performed because it is rejected.

Now I'll leave the third one as it is, and let me refresh this one.

Okay. Yeah. So now you see that Einstein does not have any recommendations for you right now. This takes us back to the limit re offer.

So as we had configured, if user rejects the recommendations once, then after twenty four hours only, he'll be able to get another recommendation. So my limit re offer is also working absolutely fine.

Okay. Moving back to these slides.

Okay. Now there are a few considerations to be kept in mind for both prediction builder and the next best action.

So prediction builder is available in Enterprise Performance Unlimited and Developer Edition. It is available for all custom object, but it supports it does not support a couple of standard objects. And the list is pretty much available over the help page of prediction builder.

And as I've mentioned earlier also, we have to have enough records to build the build a successful prediction because there is a minimum number of records the prediction builder needs to run. So, yeah, that is one thing we need to take care. As of now, there are limited field for which it could be for which a prediction could be made. That is a checkbox and a few considered formula fields. Numeric is still in beta.

And yeah. One more important thing is prediction created in your production aren't copied to your sandboxes. So so your prediction for production runs absolutely independent of prediction for sandboxes. They have to run separately.

Okay. Now here is Einstein next best action, which is available in almost all the editions. So, yeah, you basically have to have one active flow, a recommendation, and a strategy in place to surface the correct recommendation to a particular record.

And strategy builder is available only in Lightning Experience. Also, there is one interesting thing with recommendations. So while creating recommendations, you might want to include a category field. So the what category field would do is it will give you more control while building your strategy. So you might have the same recommendation names for different categories of your leads. You can basically, you know, segregate them based on and you'll have a perfect strategy to be taken up.

Okay. So before moving to q and a, I have some I'll I'll like to give some information about the licensing and price pricing of both of them. So Einstein prediction builder basically has got two types of licenses, Einstein predictions and the Einstein Analytics Plus license. With Einstein's predictions license, you can have up to twenty predictions, but ten could be enabled. And with the plus license, you can enable thirty five out of forty five predictions which you have built in your org.

And about next best action, it is absolutely usage based. So yeah. As you use it, once you have hit the limit of your free action hits, you'll be contacted by Salesforce depending on your contract.

And, also, you can keep a track of how much you know, how where where is your limits stand for your org. It is available under the company information. Yeah. So once you see the company information, you have something called maximum flow interviews per UI month and then maximum next action request available. So this is all related to flows and next best action. So here you can see what is your allowance, what is your usage, and all of that.

Okay.

Any questions?

We won't sue you. Okay?

Like, everybody watching is just asked to make that promise, then we're fine. Looking at this, it seems like probably it'd be smart to be careful about how you pick your example set. Maybe you'd curate it to make sure that it has good data quality or something like that.

You know, not throw every lead in your whole database at at it, or do you think that's smart or not?

Yeah. So prediction builder is very smart to, you know, understand how your data is. Because if you have actually, when you when if we talk about real data, it would be pretty diverse. Right?

But while creating dummy data, it will show you that it's it is too high or too low. So you I mean, it's I mean, you don't need prediction for them. Right? Because if everything is going so perfect, you really don't need the prediction builder.

And in case it is too low, then there's a lot to be worked upon. So, yeah, prediction builder is smart enough to know if you have correct data for prediction or not.

Oh, okay. So you so I don't have to sit down and mess with my data. I just throw it at prediction builder, and then it it's gonna tell me if it's any good.

Yeah.

The only thing is that we have to have minimum of four hundred records Okay.

Which so that the prediction could be built upon.

Good. Good. I saw in strategy builder that you could bring in outside data.

Is that, like Yeah.

Like, from databases and things like that that you'd match up?

Do you Yeah.

Any any external database external database or any other Salesforce Salesforce products as well, you'll be able to bring in data there.

Oh, okay. So that would that would allow you to add, you know, if you had some information about whether that person has been contacted. Like if a lead has been contacted by via email, you might have that in another database or another system. Yeah. And you could put that into your strategy if you push that in.

That's pretty cool. It's pretty cool.

Have you you seen this used in I mean, is this is all pretty new. Right? This came out last year, roughly?

Yeah. Yeah. Yeah. Around last year, I believe. But yeah.

Have you seen it used in production and had any experience with it, or is it still on?

No. I haven't have worked on a real project using Einstein Billets, but I'm sure that must be a couple of around where yeah. But then it's interesting. Trust me. Just anyone who could go through it would just love it. Yeah.

Yeah. I was just wondering if I mean, from what from what it looks like, it looks like it's a pretty powerful thing and that it'd be worth experimenting with even in a in a production org because, you know Yeah. If you just I assume you can enable and disable the the Einstein prediction builder on a page layout. Right? So if you wanted to only have a certain group of people see it or whatever, could, you know, record Yeah.

That is what you know, it looks very very easy when we just go ahead and enable or disable it, but I'm I'm just I I go into all when I think that how smart it is that once turned off, it will just stop all of all what it is doing.

Okay.

Yeah.

Alright. Well, thank you. Thank you, Sikhi. I think you have another slide to show us some resources here, so I'll let you continue on.

Yep. Thanks.

Okay. So here are a couple of references. I I even I feel that the help pages of the Salesforce are the best place to find anything, and a couple of trailheads which I have been through and would recommend whoever is going to use the prediction builder and next best actions.

Okay. Thank you all. Here is my email address and my Twitter handle where we could stay connected. Thanks, Thanks, x four's data summit for having me here. Thanks, Leonard. Thanks, Casey, for everything.

Thank you. So much. This is a this is an exciting new part of Salesforce, and I'm glad you went and did some research on it and put together a really nice demo to let people know the basics of how to get, predictions and next best action, going in their in their org. Thank you so much, and thank you for staying up late to record this Yeah. Presentation.

‍