Can AI transform churn risk assessment, or is it an efficiency play? Talking subscription management and CX updates with Oracle's Nate Skinner

Oracle's quarterly CX release keeps analysts on our toes - and this quarter was no exception (Oracle Live: Quarterly Update on Innovations for Advertising and CX.- May 2021)

With Oracle's May 2021 updates, we've got a slew of new functionality to consider, from in-game 3D advertising to subscription management.

But my question is always behind the curtain: Why does this matter to customers? And: what trends are driving this functionality?

Last quarter, those questions led to a provocative chat on why B2B sales needs to change (Can B2B sales be automated, and can bots make sales reps more effective? Oracle CX updates spark the next generation sales debate).

This time around, Oracle's Nate Skinner, Global SVP of Marketing for Oracle Advertising and CX,  drew the short straw hopped on video to talk about a topic guaranteed to generate skeptical questions from yours truly: the role of AI in CX - specifically, in identifying and preventing customer churn.

Subscription management is not just a B2C thing - it's surging in B2B also

Of the new functionality Oracle shipped, it's subscription management that captures my attention. Why? Because the pandemic economy has forced the issue. As Skinner told me:

Over the last year, there has been a dramatic increase in subscription management interest. Companies are looking for ways to manage recurring revenue, to create subscription models from what weren't subscriptions before. This is not just a B2C thing; it's B2B as well.

If you have to change your business model, you do it. Skinner:

We all understand that reaching and finding new customers is difficult under normal circumstances. It's been more difficult under pandemic circumstances, so people turn their attention towards: 'How do we retain the ones we've got?'

Thus the new functionality:

We already had Oracle Subscription Management. We've added predictive dashboards to help customers understand churn probability.

First, says Skinner, you've got to cover the basics:

Simple stuff like: 'How many expiring subscriptions do 'I have? How many active subscribers are we managing? What's my monthly recurring revenue?

These are questions that a year ago, a lot of companies didn't have to think about, or weren't really thinking about. Now they really want to understand where the heart of their customer base is, how they are retaining them, and whether there is any risk of losing them.

But those are churn fundamentals. To get more precise - or more pro-active - you need (ahem) "smarter" data. Bring on the predictive dashboards. Skinner:

What the predictive dashboards do is signal, almost like an early warning system, that you might have subscriber risk. You might have a recurring customer who's not logged in recently, or hasn't paid their bill in the last 60 days, or something like that. That tells you that you might have a problem, and helps you zoom in on that churn probability, and actively renew those customers, as opposed to reacting after the fact.

That's where the AI comes in also: "Our AI that's built-in helps you understand those things. It looks for lookalike instances in the data."

Skinner on the spot - how is AI churn prediction different?

But identifying potential churn is not a new concept. What can "AI" do that is superior to the churn management we've seen before? We have well-understood SaaS metrics around churn. In theory, a human could trawl through the data and locate churn indicators. So is "AI" simply a time saver? Or are we uncovering things the human eye would have missed? Skinner responded:

First of all, it is not about efficiency. It is efficient, of course, but the real benefit is uncovering insights you didn't even know to look for.

For example?

You don't have triggers in normal billing systems that say, 'Customer X has not paid their bill on time for some period in a row.' Yes, your ERP system might trigger that after 90 days, it red flags, and you get your red list, right? But what doesn't happen is: show me any others that are at risk of falling into the same category. I don't know about them, and then it's too late.

Without AI, you're looking only at the trailing indicators, not leading indicators. So what AI does in the context of subscription management is: giving me the insight into customers I didn't even know to be concerned about, before they churn. And so then my customer success team, or my sales rep, or my field tech, can raise the issue with the customer in advance of their churn.

To me, that's classic enterprise AI: do something humans aren't very good at. Exhibit A: identify patterns (or anomalies) at scale. Skinner:

That's what AI is doing for customers in this context. It's the idea that, 'Hey, companies that are of a certain type, in a certain geography and a certain zip code, the machine can start to look for commonality between all those that are churning, and all those that aren't, and see if there's any characteristics.' Human beings can't do that.

To me, where this gets interesting is: could "AI" help define new metrics to identify churn, and help scale a response? Or, if not new metrics, metrics that humans wouldn't be able to keep up with? I asked Skinner if he's seen any new AI churn metrics emerge. One stood out to me: user application activity. In SaaS, adoption is everything. If user activity diminishes, you want to get on top of it quickly. You want to automate/scale some type of useful engagement around that change in behavior. As Skinner put it:

In a SaaS world, everything we do is in a browser, and we log in to these systems, right? We log into our bank; we log into our insurance company's website; we log into our CRM; we log into our marketing automation solution. Everybody logs in. Well, what if they're not logging in? Should that tell us something about the health of that customer?

My take - AI has a role in churn risk, but we must avoid creepy automation

I can see AI as a useful tool for churn risk identification. Here's where the debate gets interesting: will this go beyond an efficiency play? In Skinner's experience, it already has. But I could see it going even further, with AI identifying unexpected risk patterns for churn that haven't even occurred to us at this point.

Certainly, "AI" is far superior to humans, when it comes to comprehensively scouring system usage, or subtle changes in payment timing and processing. Granted, our success here also ties to data plumbing. Without a good data platform, you can forget about using AI for anything useful. Obviously, Oracle thinks they have a good data platform story. Skinner offered his critique of the Customer Data Platform hype, arguing in favor of Oracle's approach - centered around its CX Unity platform.

Whatever you think of that, I do agree with Oracle's contention that you aren't getting anywhere on "customer experience" without access to data traditionally thought of as "ERP" data - including finance data. However, settling the data platform problem is beyond the scope of this article. If that debate hits on your interests, I'd recommend catching the video replay of my CX buzzword deconstruction with analyst Thomas Wieberneit (We also touched on the AI-churn/efficiency debate).

I buy into the argument that pro-active interventions with "at risk" customers reduces churn. I also agree that some combination of AI and automated triggers can help to engage customers that have disengaged, before it's too late. However, there are crucial project design issues to consider. One is the obvious "false positive." Sometimes what AI systems flag as churn risk could simply be a shift in workforce responsibilities. 

Here's the more important concern: avoiding creepy automations. Even if you know a customer hasn't been logging in regularly, an email saying "We see you haven't been logging in recently" is likely to come off as a creeptastic fail. Skinner agrees, and he proposed some alternatives. One example we were kicking around: have a trigger that invites such users to a helpful live product update seminar. Or: send them links that could address their interests.

For your most profitable/VIP customers, a well-designed system might not trigger a direct response at all. It could simply alert the account lead, and they could personally check in. Alternately, I could see AI being used to send triggers to your most loyal and active customers as well. Skinner talked about adding or enhancing loyalty programs for this purpose. B2C companies are perhaps well-versed in such programs; B2B companies new to B2C may not be.

Skinner also walked me through another key component of Oracle's subscription management enhancement: self-service features (example: you want to pause or re-active your water cooler subscription as offices closer or re-open). That makes sense - why burden the service team with account maintenance issues the customer can happily do, if you make it easy for them.

For space reasons, I can't do into all of the new Oracle CX functionality, which you can check out. But I believe this AI churn debate is crucial for illuminating how new tech, data platforms, and customer success  intersect.