Busting through the retail AI hype with Salesforce's Rob Garf - at Shop.org '17

How does this sound for an enterprise gig? What if your job was to go around to various teams and tell them what you've gleaned from trend crunching - and, if necessary, challenge their direction?

Granted, that's probably not how Rob Garf, Salesforce's Commerce Cloud VP of Industry Insights, would describe his job exactly. I think he'd agree with this much though: to deliver for his internal teams, Garf needs to drill past retail clichés like "storefronts are dead" and "AI will solve the omni-channel," and share what's really working.

How does he do it? By analyzing a unique set of data, compiled by blending consumer shopping across the Salesforce platform and his team's primary research.

At Shop.org 2017, Garf told me what he's learned, and helped me to pierce through the AI fog with some practical views of Salesforce Einstein in action on the Commerce Cloud.

Siloed retail intelligence doesn't work

Garf wasn't the only one I interviewed who noticed the surge of AI marketblasting since the NRF's "Big Show" just eight months ago:

You can't really walk two feet without seeing the two letters A-I. Artificial Intelligence is certainly a buzzword.

The hype is out of control:

We made a joke at the Shop.org digital council meeting that all you need to do now as a retailer is retitle all your business analysts to "data scientists," because it sounds much sexier and will help retain them and attract new talent.

When it comes to turning "AI" into results, Garf sees two key hurdles. First up: not all companies have data science teams to build-their-own algorithms. But even if you have figured out how to generate useful analytics, there's another disconnect:

If you do have the intelligence baked into your systems and processes, you need a way to actually execute it.

Siloed intelligence won't work in retail, where real-time actions are needed:

It's no longer about looking at one interface and gaining the intelligence and then going to another interface to execute it. It's really baking it into the day-to-day, the day-in-the-life if you will, of the merchant - and even the consumer as they're shopping.

So what does Garf recommend for retail customers who want to capitalize on retail intelligence, but don't have a deep data science team?

We say start small and tangible, and base it around specific problems you're trying to solve on behalf of the customer.

It's easy to say "start small," but how? Garf says you might want to start with your known and unknown visitors. That tripped me up a bit. Because, let's face it, even when companies have the advantage of dealing with a "known" visitor or customer, they tend to botch things up.

How many times have we returned to the same hotel and they miss the golden chance to leave our favorite items on the dresser in our room, either for a welcome gift or obvious purchase? What say you, Mr. Garf?

I like the hotel example... I think for the known, it's first of all being able to organize and distill all the different signals and profile and preference information across the various systems.

Welcome to the omni-channel blues:

Part of the issue is retailers have, by no fault of any individual, invested in specific systems to solve specific problems for specific channels, right?

With that setup, connectivity is a beast:

There's disparate data. There's tough integration. There's no real-time integration happening.

Intelligence can't be automated without a context

Then you throw in new - and highly relevant - external data such as social media channels, and perhaps product sensor data, and you're adding fuel to the data fire.

I like that analogy. It's very complex, for retailers to get their handle on all of that data and, as we often talk about, action it. To be able to, in real-time, in volume, be able to get that exposed to the point of interaction. Whether that's the store associate or the consumer themselves.

"Real-time" must also have these three traits:

• Device-aware - "Understanding what device the consumer is using, so it can be formatted in a way that makes most sense and is easy and intuitive."
• Location-aware - "It also needs to be location-aware, in terms of where they are in the context of physical brick and mortar locations. It also needs to be context-aware, understanding what the consumer has done in the past, and what they're likely to do in the future. It's all this awareness that helps artificial intelligence get smarter around the data that already exists."
• Context aware - "Understanding what the consumer has done in the past, and what they're likely to do in the future. It's all this awareness that helps artificial intelligence get smarter around the data that already exists."

But wait - aren't we now looking to automate this data, in an intelligent/personal way that doesn't feel like machinery to the customer?

Garf makes the retail AI case for Salesforce Einstein

The answer, of course, is yes. Naturally, Garf thinks Salesforce has come up with a great way to handle that, via Einstein AI, and the Salesforce platform. Garf gave a customer example:

They were really intrigued by ratings and reviews, mostly reviews, for user-generated content. They put a system in place through Salesforce Marketing Cloud to trigger an email to basically say, "Can you provide a review?" In some cases, they include an incentive to do that.

But now Einstein takes it up a notch:

Through Marketing Cloud Einstein, they realize these sets of consumers provide reviews on Saturdays - generally ten days after the purchase. So don't send them an email on Tuesday, because it's going to get lost in the inbox. Let's send it on, say, Friday night/Saturday morning. By the way, that's not a digital marketing manager making that decision. Although they might have conceived that through the data. It's the engine. It's the artificial intelligence that is actually making it smarter.

This is personalization beyond brute force segmentation:

In this case, it's the ability to do it on a one-on-one basis. That's part of the beauty of it. It's not segmented, where you're trying to group people together based on, typically, one or two dimensions.

Makes sense - but what about those "unknowns" that have befuddled retailers, with only basics like location or IP address to go on? What is our goal with them? Are we trying to get them to purchase something? Are we trying to get them to make baby steps? Salesforce says they can lure unknowns into shopping actions:

With our Commerce Cloud Einstein, what we're doing is we're constantly listening to the digital footprint on the website in terms of what they're clicking, where they're clicking, duration of time on pages, what they put into the cart.

That profile is then acted upon in real-time:

On the fly, that's rendering different experiences based on who that individual is, and what their preferences and profile might be.

Garf gave a common retail example: a value shopper versus a fashion shopper:

We get to know ["the unknowns"] pretty quickly as to what their preferences are, and start to really curate their experience based on that.

My take

Diginomica readers are notorious for actually expecting results, not just cool tech. So are we getting more conversions as a result of this "retail intelligence," which also includes predictive recommendations? Short answer: yes, more engagement, and ultimately, more conversions. To prove the point, Salesforce is about to release fresh personalization research in the next couple of weeks.

I did not succeed in pulling any teaser data out of Garf from the report. But he did say this much:

[Our data] just shows unbelievable conversion-rate uplift for those websites that have recommendations driven by Einstein on their website.

And yes, that includes both the known and unknown visitors. Garf did concede that so far, the impact isn't on clicks. But the impact on revenue is substantial:

It's some mind-blowing numbers in terms of improvement. It's still relatively low in terms of click, but those that do click, you're seeing a disproportionately high percent of revenue overall from that.

Luring in the most-likely-to-purchase consumers via real relevance - and not spray/pray - is a pretty good step. I like the relentlessly practical approach here, but I won't deny - I'm looking forward to more sophisticated examples as this unfolds. Yeah, warm macadamia nuts waiting in my hotel room is all I want. Then, and only then, I'll pronounce retail intelligence a success.