My First Einstein Prediction
[This turns out to be the third in a series of posts. Not that I meant to have a series in the first place. Or if I did, I hoped it would be about the cool things I learned from AI predictions, not a deep dive into installing and licensing...
Anyway, this post should stand alone, but if you want the background, read “Cheap (or free?) AI for Nonprofits?, and Still Trying to Try Einstein.” I’ll have more detail and some updates here, but I’m not going to rehash everything I wrote before!]
I finally got Einstein Prediction Builder (EPB)/Einstein for Nonprofits to work in a real client production Salesforce instance!
Actually, true confession: I pulled that off at the end of November, three months ago. Then I promptly didn’t do anything with it (or even write this blog post) because I was lazy. But that, I think, is part of the learning journey we’re on together!
How to Install Einstein for Nonprofits
Though there is help documentation, I think it’s worth listing out the steps to install while we’re at it.
1. Go to Setup>Einstein Prediction Builder, review the terms and conditions, and turn it on.
Eventually, you'll have this:
You need to already have Einstein Prediction Builder (EPB) turned on before you can install Einstein for Nonprofits (EFN).
2. Install Einstein for Nonprofits via Salesforce.org Metadeploy (https://install.salesforce.org/products/efn/latest)
More on this down below, but it’s possible you won’t be able to install if you don’t have Nonprofit Cloud.
3. Navigate to the Einstein for Nonprofits app (see this help page) and click “Setup and Train Models.”
4. Navigate back to Setup>Einstein Prediction Builder. You should see three predictions installed now. Now’s your chance to [try to] activate one of them.
Quick Sidebar into My Experience Activating Predictions
As I noted in Still Trying to Try Einstein, the installed predictions refused to activate for me in one of my client orgs (the one that has interesting and plentiful data). They were stuck in Pending status. Support eventually suggested that I clone them and activate the clones. I procrastinated on that for a while because it seemed like it would be a pain, but it actually turned out to be easier than I expected.
Einstein takes care of creating a target field for holding your prediction results. You don’t have to create a field before you start the prediction builder wizard. Nice!
It was a bit of a "gotcha" that the Description field on a prediction can only hold 128 characters. (You don't find that out until you try to save.)
Of course, if you have to clone a prediction that failed to activate, you’re going to have the target field for the failed one cluttering up your org. Not a big deal, but this kind of thing annoys me.
Support tells me that my cloned predictions will use the Einstein for Nonprofits backup models if there isn't enough data in my org. (I have absolutely no way to evaluate whether this is true.)
A Working Prediction!
With my cloned prediction active, let's look at the scorecard EPB provides:
It appears that EPB thinks that having a value in Preferred Email (home, work, alternate email selection) is the single most relevant field for predicting if someone is likely to become a first time donor. I have...opinions...on that. (But what do I know? Maybe the machine learning has spotted something interesting that we humans wouldn't assume. That's the whole point of this exercise, so I'll give it the benefit of the doubt.)
I then enabled the prediction and went away for the weekend while it churned through the data. On Monday, a contact report showing my target field had results!
In the image, I have grouped the report by the result values of the target field, so the Record Count is how many contacts have that number in the result field.
What threw me at first was that a Yes/No prediction ("Is this contact likely to become a first time donor?") results in numbers, rather than true/false. The numbers are the percentage likelihood (of becoming a first time donor.) That totally makes sense now, but it had been a while since I completed the Einstein Trailhead modules, so I needed a moment to understand it.
The real problem is that I’m not sure what to do with this now that I have a distribution. I suppose we could target outreach and fundraising efforts to the 29,000 people that are at least 31% likely to become first time donors, but that’s a lot of people. Far too many for individual outreach efforts...
This is the point at which I found myself really stuck. I'm still not sure it's worth showing this to the client in whose org I tried it out because I don't think they have the staff bandwidth to use the information. And I'm not sure how I feel about the quality of the prediction.
With Try Einstein, the freemium trial, you can have only one prediction activated at a time. My understanding is that you could deactivate this one and activate another one and the values in the target field (what's showing on my report) will persist. So you can switch back and forth to learn from any or all of the predictions you have in your org (those installed by Einstein for Nonprofits, those you create by cloning, or those you build from scratch). However, the inactive predictions won't update as new data comes in. That's probably OK depending on what you might do with the predictions.
I haven't actually tried out activating the other two predictions that EFN installs. Partly that's because I would have to go through the rigamarole of cloning them first, due to the activation issue I experienced. But also, in that org I am unsure of to evaluate the quality of—or what we might do with—predictions about who is likely to become a Top Donor or a Recurring Donor.
Now Let's Get Into Licensing
Yes, we're going to have to get into the weeds of Salesforce licensing and pricing.
It turns out that Einstein for Nonprofits isn't exactly "free" and isn't exactly available to every nonprofit. Shocking, I know.
First of all, according to the person I spoke to at Salesforce.org, Einstein for Nonprofits should not install into an org that does not have the Nonprofit Cloud SKU. What constitutes that SKU is hard to pin down (and is probably going to change multiple times). But the quick-and-dirty is that organizations with just the P10 donated licenses do not have Nonprofit Cloud. A Nonprofit Cloud license is more expensive than a regular "11th" Sales Cloud license for a nonprofit, though I haven't been able to learn how much more. The Nonprofit Cloud SKU comes with Einstein for Nonprofits, Accounting Subledger, and Insights Data Platform Integrity, so doing the math from the crowdsourced pricing table, I think it's a good chunk of change.
One strategy might be to buy CRM Analytics+ (CRMA+) [formerly known as "Wave Analytics" and then "Tableau CRM"]. That bundle is reasonably priced, is discounted for nonprofits, and would give you Einstein Predictions and Einstein Discovery. (Einstein Predictions would allow you to build and activate more EPBs at a time.) But CRMA+ should not actually allow you to install the EFN predictions. I'm told that Discovery (which is to say, "Tableau") may actually be more insightful for nonprofits than EPB because it allows them to see more of the "why" behind information. Moving more toward selling CRMA+ and helping nonprofits do more with visualizing their data is likely part of Salesforce.org's future direction.
Your Installation Mileage May Vary (Mine did.)
Finally, let me note that I was able to install EFN into two orgs that absolutely do not have the Nonprofit Cloud SKU. You might be able to as well... but Salesforce's intention is to fix that. (Maybe they already have.) Consider grabbing EFN while you can! You also might want to clone the predictions in case they disappear if/when Salesforce fixes the licensing tie-in.