Transkrip penuh untuk Why Faster Refunds Can Backfire Without Smarter Fraud Controls

and try to social engineer their way uh around your controls. 00:05:05.600 --> 00:05:16.959 So it's being able to um plug those two gaps really, good decisions about individuals and um consistency in every single channel, the same controls in every single place. 00:05:17.199 --> 00:05:26.800 Um there's a third one though, um uh that's a big gap as well, which is kind of actionable insights, uh being able to react as well. 00:05:26.879 --> 00:05:44.079 Uh so you've got what you you know you think is a really good control in place, you've got the consistency in place, but uh knowing how those controls are performing, uh, how quickly can you identify trends that are happening in your returns data and other business decisions that might have impacted the performance of your returns? 00:05:44.160 --> 00:05:48.879 And how quickly then can you can adapt your controls uh to what you're seeing? 00:05:49.439 --> 00:05:50.079 Absolutely. 00:05:50.240 --> 00:05:57.680 I think um obviously, as visibility becomes more of an important capability, it's also it's becoming a lot easier to use. 00:05:57.759 --> 00:06:03.360 We're noticing customers leaving a lot more fingerprints on the journey more than ever. 00:06:03.600 --> 00:06:11.279 So I think um, yeah, it's less of an expectation and more of a standard at this point because it's so easy for to notice. 00:06:11.920 --> 00:06:23.279 Um, and as policy-based returns processes and high value product categories create uneven risk exposure, fraud is increasingly driven by behaviour rather than single transactions. 00:06:23.439 --> 00:06:31.199 Paul, can you tell us where are the current return policies most vulnerable and how are fraud patterns evolving around them? 00:06:31.519 --> 00:06:32.160 Yeah, sure. 00:06:32.319 --> 00:06:35.279 I think um I think the policy is the key word here, really. 00:06:35.360 --> 00:06:39.120 We often say this policies are essentially a menu or a playbook. 00:06:39.279 --> 00:06:44.720 Um, the worst actors will uh learn them and they'll operate operate inside them then. 00:06:44.879 --> 00:06:50.639 Uh you're up against professionals who who will do their research, you know, that's that's what you're often up against. 00:06:50.800 --> 00:06:56.800 Um, even an opportunistic uh abuser is going to be enabled by a fixed policy. 00:06:56.959 --> 00:07:02.399 Um uh they'll they'll have made decisions on where they're shopping and where they're returning based on that sometimes. 00:07:02.639 --> 00:07:11.600 Um yeah, and um your good customers are going to get caught up in all of these restrictions that you're trying to do um with uh fixed policies. 00:07:11.759 --> 00:07:19.199 So I think that's the key part to take away, really, is that is that policies and rules are a major part of the problem. 00:07:20.000 --> 00:07:35.120 Um you know, you can um uh build a policy uh that would uh that gives your um good customers uh that speed that you're after, that that you know it's gonna let them

through, but that's just gonna have holes in it all over the place. 00:07:35.759 --> 00:07:53.519 Um so that's where retailers are are understanding this, realizing uh that the decision needs to be dynamic, um, that the uh it needs to be pattern-based, it needs to be behavioural based, um, it needs to be you know individual event-based decisions rather than uh policies. 00:07:54.959 --> 00:08:02.959 Absolutely, I think um organizations need to be aware constantly that the their policies are creating unintended vulnerabilities. 00:08:03.120 --> 00:08:06.319 So we need to make sure that they are regularly reviewed. 00:08:06.560 --> 00:08:16.000 And as retailers look to reduce fraud without damaging customer experience, more attention is shifting towards risk-based decisioning at the point of return. 00:08:16.160 --> 00:08:22.160 What practical approaches can help reduce risk controls without slowing down legitimate customers? 00:08:22.480 --> 00:08:31.519 Yeah, and I think if you're if you're already trying to solve this, you've probably already started measuring that um uh cost of friction to good customers. 00:08:31.600 --> 00:08:36.399 And I think um I, you know, it shouldn't be any surprise that there is quite a cost of that. 00:08:36.480 --> 00:08:42.159 Um, it's real, it's often ignored uh when trying to think about how to deal with returns. 00:08:42.240 --> 00:08:43.279 But what is that impact? 00:08:43.360 --> 00:08:54.320 And again, in our benchmark report, we've got something like um 72% of uh shoppers uh said that they went somewhere else when they had a negative return experience. 00:08:54.559 --> 00:08:57.759 So, whatever you do, it needs to be invisible to them. 00:08:57.919 --> 00:09:04.960 That's where your capability needs to focus as much as it does need to focus on preventing uh abuse and fraud. 00:09:05.120 --> 00:09:09.200 Um, it needs to benefit them uh and be a positive experience for them. 00:09:09.360 --> 00:09:15.360 So, yeah, any individual uh uh intervention should match the individual's risk. 00:09:15.440 --> 00:09:20.720 Uh, a flat policy here, you know, it misses the bad stuff and it frustrates the good people. 00:09:21.200 --> 00:09:30.080 Um, so that's the kind of focus that of whatever it is that you're going to do and as a capability uh that you're going to introduce, that's that's where you should be looking. 00:09:30.240 --> 00:09:35.840 Um, you know, as a retailer, you have um so much data on your customers. 00:09:36.000 --> 00:09:39.039 Um, use it to make the right decisions. 00:09:39.279 --> 00:09:53.279 Um the approach that we talk about is constantly build up a picture of your customers, every interaction they do with uh your business, every purchase they make, every return they make, every call to the call center that they've made. 00:09:53.360 --> 00:09:56.000 Um you constantly build up that picture. 00:09:56.240 --> 00:10:00.000 Um and then apply decision intelligence to that. 00:10:00.159 --> 00:10:04.320 Um uh that's where you know machine learning comes into play here. 00:10:04.399 --> 00:10:22.240 It can do the thing that humans can't do around linking um identities together and interactions together, but it also can then use those tailored uh models um to make uh incredibly accurate decisions um uh based on all of this brought together and the event that's happening right now. 00:10:22.320 --> 00:10:23.440 What is it they're returning? 00:10:23.519 --> 00:10:24.799 Where are they returning it? 00:10:24.960 --> 00:10:29.120 Is this normal, is this not normal for that individual person? 00:10:30.080 --> 00:10:30.639 Absolutely. 00:10:30.720 --> 00:10:42.399 I think um yeah, the key here is obviously maintaining that smooth experience, as I kind of mentioned before, that um customers are leaving all the the fingerprints um more often now. 00:10:42.559 --> 00:10:45.039 So creating that smooth experience is becoming easy. 00:10:45.440 --> 00:10:47.600 There's another part as there's another part as well. 00:10:48.080 --> 00:10:52.960 It's really good to take the human out of any friction as well. 00:10:53.120 --> 00:10:54.639 We often talk about this part as well. 00:10:54.799 --> 00:11:13.120 Give your employees the confidence um that uh the controls are correct uh and that the decision is final, almost make it like a credit card decision uh when you pay uh so that the employee is not there in the firing line for potentially uh an abuser or a bad actor. 00:11:13.360 --> 00:11:18.799 Um uh and because that will also uh enable you know ways around. 00:11:18.879 --> 00:11:21.840 They'll use social engineering, they'll use confrontation. 00:11:22.000 --> 00:11:28.639 Um, you've got an individual then trying uh individual employee trying to help them uh to or to resolve the situation. 00:11:28.879 --> 00:11:38.879 So as good as your controls are, you need that uh support from your employees and that understanding that this is the policy, this is this has uh the decision has been made. 00:11:39.039 --> 00:11:40.879 The escalation now is not me. 00:11:40.960 --> 00:11:52.879 Um here's how you can escalate it and have a really clear escalation path for someone to uh you know go and uh contact uh the person that would give them the informed uh information about why. 00:11:53.039 --> 00:11:54.960 Uh all of that needs to be in place as well. 00:11:55.039 --> 00:11:59.759 So it's not just about the control, it's about thinking how does this fit into the overall customer experience. 00:11:59.919 --> 00:12:08.960 Um, you know, you you may well be making a 99% accurate decision with these systems, but there is a chance that um someone needs to be able to escalate it. 00:12:09.039 --> 00:12:13.279 But don't put that in the hands of the employee that potentially that's right there in front of them. 00:12:13.759 --> 00:12:14.480 Yeah, absolutely. 00:12:14.639 --> 00:12:24.720 I wondered where you were going with that point, actually, when you said no

employees, but um absolutely, I think also the key to customer experience is also ensuring the satisfaction of your employees. 00:12:24.799 --> 00:12:29.919 That's a very key point, especially in those um high-risk sensitive situations. 00:12:30.240 --> 00:12:38.799 And as AI and data-driven tools become more embedded in retail operations, return fraud detection is moving toward predictive models. 00:12:39.039 --> 00:12:46.639 How can behavioral signals and machine learning be used effectively to identify suspicious return activity earlier? 00:12:46.960 --> 00:12:52.240 Yeah, I've spoken about it a bit there, and I think um a lot of people talk about scores. 00:12:52.320 --> 00:12:53.279 We don't talk about scores. 00:12:53.360 --> 00:12:58.399 We talk about behavioural signals, we talk about uh lifetime value, very important. 00:12:58.720 --> 00:13:04.639 Um, and then we talk about um that identity uh network, that linking of things together. 00:13:04.720 --> 00:13:15.679 That's something that machine learning is incredibly good at looking through all of your data in real time, um, and building up a uh a constant, accurate picture of risk. 00:13:16.000 --> 00:13:25.840 Um, and the decision isn't made there and then you wouldn't that's what scoring would be, where you would uh constantly be um making decisions about that person. 00:13:26.000 --> 00:13:37.679 This is more about collecting a good picture, an overall profile of every one of your uh customers and every one of their interactions, and then the decision you use machine learning then to uh okay, what's happening right now? 00:13:37.759 --> 00:13:40.159 What is what is the person uh trying to return? 00:13:40.240 --> 00:13:41.759 What is the value of that item? 00:13:41.919 --> 00:13:43.519 Is this something they would normally buy? 00:13:43.679 --> 00:13:47.120 Is this um uh the location they would normally return? 00:13:47.200 --> 00:13:55.120 Yeah, all of those things, or if if if it's online, how many uh different um products uh are involved in this interaction? 00:13:55.279 --> 00:13:58.240 Lots there's lots of velocity, all of those things. 00:13:58.399 --> 00:14:05.279 Um and then yeah, make that extremely accurate decision that machine learning can give you based on all of the different models it's got. 00:14:05.360 --> 00:14:09.279 Um and you have to think about again that good customer scenario here. 00:14:09.440 --> 00:14:19.279 Um if we talk about someone potentially um returning something online and they've already done uh the last two things that they bought, um, they've already uh returned those as well. 00:14:19.440 --> 00:14:24.639 That flat policy approach that we uh spoke about earlier, that potentially would say, nope, you're done. 00:14:24.720 --> 00:14:27.279 Um you've this is your third return now. 00:14:27.519 --> 00:14:32.639 Um uh and it could be based still on uh value of uh of that person. 00:14:32.799 --> 00:14:36.480 We know that retailers do this, but it could be based only on that channel. 00:14:36.559 --> 00:14:43.279 So the focus really needs to be on what else do you know about this consumer outside of just their purchases and returns on your website? 00:14:43.519 --> 00:14:50.399 Can you leverage uh if you're a multi-channel retailer, can you leverage everything you know about them in store? 00:14:50.720 --> 00:15:04.799 Has your machine learning picked up on this is the same person as the uh someone that shopped in store, maybe looking at their payment card usage, their loyalty card numbers that they've used in store and saying these are the same people, this person's actually high value here. 00:15:04.960 --> 00:15:13.679 Yes, they've returned the three things that they've uh bought online, but um actually uh that they're very low risk and you should still proceed with this, and then vice versa. 00:15:13.759 --> 00:15:26.240 You know, they could be um you could you could have that very clear now informed uh uh perspective of that person that um this is the first time they've returned something online, but actually we know they've been returning a lot of things in store. 00:15:26.399 --> 00:15:39.600 Um uh and they've been potentially wardrobing in store, buying things, returning them the next day, probably probably uh after um wearing them or bracketing, um, those kinds of uh behavioural things that you might not see in the channel that you're focused on. 00:15:39.759 --> 00:15:42.159 Machine learning picks up on this very, very quickly. 00:15:42.320 --> 00:15:47.759 It connects all those dots that uh a human can't do and gives you a really, really good risk profile, then. 00:15:48.960 --> 00:15:49.440 That's great. 00:15:49.519 --> 00:15:55.519 That's a really uh fascinating example of how AI um can move from reactive detention to that proactive prevention. 00:15:55.679 --> 00:15:56.720 So thank you very much. 00:15:56.960 --> 00:16:10.879 And before we end our discussion today, as returns management becomes more central to profitability and operational operational strategy, organizations are rethinking how data fraud teams and logistic functions work together. 00:16:11.120 --> 00:16:19.279 So looking ahead, what changes in structure or capability are needed to make returns management more proactive and strategic? 00:16:20.320 --> 00:16:24.879 Yeah, I think um returns basically needs a seat at the table. 00:16:24.960 --> 00:16:27.120 I think that's uh that's the way I would say this. 00:16:27.360 --> 00:16:44.240 Um whenever there's a policy being written, whenever promotions are being planned, whenever a new channel uh or a new partnership is happening, uh, all of those things, there needs to be uh decisions-based and discussions around how does this impact returns? 00:16:44.399 --> 00:16:50.000 Uh, how can we predict what the uh what that impact is going to be from a uh returns perspective? 00:16:50.159 --> 00:16:53.279 How does it impact the controls that we've got in place, those kinds of things. 00:16:53.440 --> 00:17:12.000 Uh retailers that treat returns as a feedback loop, you know, on product quality, on customer intent, on abuse vectors, uh they perform differently to businesses um that treat returns as a cleanup operation, as a as a as a as a byproduct. 00:17:12.160 --> 00:17:17.200 There's a big distinction between treating returns as uh as a signal, essentially. 00:17:17.680 --> 00:17:19.519 Um, you know, the data's already there. 00:17:19.920 --> 00:17:24.480 Um the question is um, are you set up to use it? 00:17:26.160 --> 00:17:26.880 Absolutely. 00:17:27.039 --> 00:17:36.079 I think um it's it'll be great to see how more organizations use returns management um as a more integrated part of the overall retail strategy. 00:17:36.319 --> 00:17:41.519 Um now, unfortunately, that is all we have time for today, but I would like to thank Paul for being here. 00:17:41.680 --> 00:17:49.839 It's been really insightful to get an idea of how retailers can balance faster refund times with effective fraud prevention strategies. 00:17:49.920 --> 00:17:51.759 So, thank you so much for joining me, Paul. 00:17:52.000 --> 00:17:52.720 Thank you for having me. 00:17:53.359 --> 00:17:53.839 Appreciate it. 00:17:54.079 --> 00:17:56.960 And from all of us at CX Today, thanks for watching. 00:17:57.039 --> 00:17:58.160 Goodbye.