Still Trying to Try Einstein
If you read my last post you were probably hoping that the title of this one would be My First Einstein Prediction. I was hoping so too.
But we're learning together!
Here's what I've learned so far:
Sandbox - What's the Point?
First, I wanted to test in a sandbox because that's doing things right.
But sandboxes, of course, have no data in them. I spent a chunk of time building out a moderate amount of fake data so that the models would at least let me turn them on. Even the time I spent just resulted in a whole bunch of very boring donations (all the same amount, on people named Contact 1, Contact 2, etc...) I could have spent dozens of hours learning Snowfakery to insert interesting data. But who has the time?
And anyway, even with interesting fake data, how interesting are predictions about it, really?
So working in the sandbox got me proof that it's safe and possible to install, enable, and activate the Einstein Prediction Builder (EPB) models that Salesforce.org has provided.
But if I wanted to really learn anything about the models I was going to have to run them against real data.
New Org - No Magic Bullet
I first turned on Einstein Prediction Builder (EPB) in the org for my client The Modern Classrooms Project.
(Really interesting organization, by the way, and great fun to work with! Super smart staff that are excited to try new technology and have quickly learned how to build in Salesforce and FormAssembly.)
MCP doesn’t have a ton of donation data because they’ve only been on Salesforce for about two years. Plus most of their income is from earned revenue. I was interested to turn on EPB in their org first to see how the NPSP backup prediction models would work when I knew there wouldn't be enough of MCP’s own data to run the machine learning.
Well…I wasn’t too impressed.
Here are the results:
First Time Donor - This one’s kinda’ interesting, if not particularly helpful. As I understand it, the model assumes that if someone has good contact information and the contact record has been created in the last three years, they’re likely to become a donor.
Modern Classrooms is a new Salesforce instance (so all records are created less than 3 years ago.) And we are building out program management on the platform, so we’ve imported well over 6,000 program participants. They all have good contact info. (It would be hard to work with them if you couldn't contact them…) I suspected Einstein was going to say everyone has a good likelihood of becoming a first time donor.
And I was right. The scorecard says “Prediction quality is too good to be true.” Even Einstein is skeptical of its prediction that basically every mentee and mentor in Salesforce is likely to become a donor to Modern Classrooms.
Top Donor and Recurring Donor just straight up failed due to too little data.
Tons of Data - No Luck Yet
Next I turned to the Clean Air Council, a client that has lots of data in their org, both many years of data from within Salesforce and decades of older imported data. I thought for sure I’d learn something from that experience!
Here I’ve been stymied even worse.
It’s been more than three weeks since I installed the predictions and turned on the models. Everything’s still stuck in Pending status.
I even opened a case with Salesforce support. All they’ve been able to do so far is point me toward a Known Issue that has no known workaround.
I Guess I'll Get Back to You
I'm still committed to taking Einstein Prediction Builder out for a spin. Keep your fingers crossed that my next post is actually titled My First Einstein Prediction...