We’ve all experienced Inference Engines
When you go into the doctor’s office, and you tell them you have a headache, smart doctors won’t tell you it’s a tumor.
You’ll describe your symptoms and you’ll give them more info, and more info, and as they ask questions, collecting more information, they’ll be able to make a stronger and stronger inference as to what you might be suffering from.
This approach is additive. It layers knowledge on top of knowledge.
If they find Symptom A, and also Symptom B, they can guess you’ll have Symptom C, and if you confirm it, they know even more and start guessing about Condition D.
This basic artificial intelligence (AI) approach to inferences (which is additive in nature) is called Forward Chaining. It’s a common way people build expert systems.
Thru rules and data collection (via questions), we build up knowledge.
Backward Chaining is altogether different, yet you experience it at the doctor’s office as well.
You walk in and say you have Wilson’s Disease.
If a doctor can quickly determine that you have no copper accumulation in your liver or brain, they rule it out.
They haven’t figured out what you have. But they know it’s likely not Wilson’s.
That’s because backward chaining starts from the end – the goal – and works backwards. Almost as if saying, “For this to be true, so must this.”
These are two common methods or approaches that I’ve used in the past to build rule engines and expert systems.
But what if we could see this level of intelligence grow from data collection that doesn’t require humans to create rules for the machines?
I know, you’re thinking I’m talking about SkyNet. But I’m not. (If you missed the reference, here.)
I’m talking about something called machine learning.
Machine learning is a fancy way of saying that algorithms (or rules) can be constructed by software rather than people by looking for correlations in data collected.
What if we saw that happen in eCommerce?
Let’s say we were talking about a regular eCommerce store. What would happen?
If it were good, it would look at my abandoned cart and send me an email. It would say something like, “Hey, did you forget to finish your transaction? Your toothpaste is missing you.”
If it were great, it would send me an even better email.
“Hey, we see you didn’t have time to finish checking out today. Here’s a coupon for 15% off on that toothpaste if you come back in the next 48 hours.”
Most people who suggest doing this kind of stuff tell you not to worry about the coupon or the 15% off. They’ll highlight that this was an abandoned cart and that you were likely to lose all of it.
But what if we had super smart, amazingly intelligent data bots that were moving thru our data, finding correlations and creating rules to evaluate things – all for us?
And what if you powered your eCommerce site with something like that?
Then you might get an email like this:
“Hey I don’t know if you noticed when you were last here but we’re doing a deal on tooth brushes – where if you buy one in the next 48 hours, you’ll get it for 30% off.”
Wait! What just happened?
Initially the email looks similar, doesn’t it? But it’s not.
Because it wasn’t offering you a coupon on the product you’d put in your cart. It was offering you a discount on a product you didn’t put in your cart.
So now the product you might come back to buy will still be full price, no discount. And the discount is for something you hadn’t even looked at yet.
But it did it because it was smart. Because it had looked at hundreds or thousands of carts and noticed a high correlation between tooth paste and tooth brushes.
This is but a silly example.
This would be powerful. Wouldn’t it?
Because it doesn’t require you to create tons of rules. It doesn’t force you to determine what the right upsell or cross sell products are.
Wouldn’t this be crazy? Amazing? And ridiculous?
What if this was already built?
What if this was already working?
What if you could run it on your small store for free?
What if you’d never heard of GraphFlow?
Good thing is…now you have.