00:00:10.260 –> 00:00:14.550
AI is accelerating in collections, but scaling it responsibly requires more than

00:00:14.610 –> 00:00:19.170
technology. In this expert session, Andrew Ducker shares 10 things leaders should

00:00:19.201 –> 00:00:23.760
understand before scaling AI across customer risk, engagement, and recovery

00:00:23.790 –> 00:00:24.300
decisions.

00:00:24.991 –> 00:00:28.440
Hi, I’m Andrew Ducker. I’m the VP and the head of the collections practice for

00:00:28.620 –> 00:00:29.070
Concentrix.

00:00:33.420 –> 00:00:37.260
I think a lot of executives still think that AI is really a panacea for only cost

00:00:37.380 –> 00:00:41.070
out, without necessarily looking at the benefit that it drives from outcomes.

00:00:41.490 –> 00:00:44.820
Outcomes are things that really matter with AI, and doing that in tandem with

00:00:44.850 –> 00:00:47.250
humans is going to be vital going forwards.

00:00:47.640 –> 00:00:50.250
So AI by itself is not considered useful.

00:00:50.370 –> 00:00:53.670
It’s only considered useful when it’s deployed alongside human interactions.

00:00:54.450 –> 00:00:59.011
Why do so many AI and automation initiatives and collections fail to deliver real

00:00:59.071 –> 00:01:02.790
outcomes? Well, the main reason is because they focus only on the technology

00:01:03.090 –> 00:01:06.211
without thinking about how to embed that within the operational model, the

00:01:06.240 –> 00:01:09.420
governance model, and crucially, that integration back into the humans.

00:01:09.480 –> 00:01:12.540
Without any of that, you can’t measure the real value of it in terms of the

00:01:12.630 –> 00:01:15.360
outcomes that AI actually begins to deliver.

00:01:19.410 –> 00:01:23.100
So a lot of the AI work that’s been done so far in proof of concept has been done

00:01:23.130 –> 00:01:27.301
as standalone projects. It hasn’t really been built on the operational environment

00:01:27.360 –> 00:01:31.590
and the data platforms it needs to actually get the rigor built into it to expand

00:01:31.650 –> 00:01:35.790
it. Without that rigor, you risk losing your ability to expand your AI solutions

00:01:35.821 –> 00:01:40.500
properly. Well, good governance will give leaders the confidence to scale AI

00:01:40.500 –> 00:01:44.160
because it gives them the ability to defend and explain what they’re using it for.

00:01:44.310 –> 00:01:48.271
Not embedding that from the outset creates a level of embedded risk that will

00:01:48.301 –> 00:01:51.660
require a huge amount of remediation in the operating model down the line.

00:01:55.650 –> 00:01:59.280
I think if there was one area I’d focus on to make AI work better in collections,

00:01:59.400 –> 00:02:01.891
it has to be around the data structure and architecture.

00:02:02.010 –> 00:02:05.310
AI can only really work its magic when it’s got the right data and the right

00:02:05.430 –> 00:02:09.150
structures to be able to work on. So that’s the bit that needs to get fixed first

00:02:09.211 –> 00:02:11.970
in order to make AI actually actionable product.

00:02:12.510 –> 00:02:15.391
Traditional AI works really well when you’re looking at some backward-facing

00:02:15.450 –> 00:02:18.300
metrics like productivity, efficiency, and output metrics.

00:02:18.421 –> 00:02:22.410
But when you start embedding AI across the collections lifecycle, there’s a

00:02:22.470 –> 00:02:26.430
different level of transparency required, whether that’s around why decisions were

00:02:26.460 –> 00:02:30.630
made by AI, the governance processes by which they were managed, and crucially,

00:02:30.780 –> 00:02:34.590
where they were overridden when a human actually intervened in the process.

00:02:38.430 –> 00:02:41.430
So automation’s been around for many years and allows you to move from A to B

00:02:41.490 –> 00:02:43.830
really quickly in a highly structured environment.

00:02:43.860 –> 00:02:48.870
The cooperation of AI is really a tool to allow you to build more fluidity into the

00:02:48.900 –> 00:02:53.220
decision-making process, which is essential because not all collections work is as

00:02:53.280 –> 00:02:57.810
structured as A to B. And I think AI is a vital component of that, especially if

00:02:57.870 –> 00:03:01.921
you can prove what it’s done to move you from A to B in a more fluid way.

00:03:02.520 –> 00:03:06.270
There’s numerous use cases across the collections journey where it’s really useful,

00:03:06.391 –> 00:03:10.410
starting right from the outset when you’re looking at the AI capability to really

00:03:10.470 –> 00:03:14.490
inform your pre-delinquency work, identifying your risk early.

00:03:14.969 –> 00:03:18.300
But it can equally be used in terms of the engagement points with customers,

00:03:18.390 –> 00:03:20.580
especially at those transactional points.

00:03:21.150 –> 00:03:24.810
But also it can support humans as part of the engagement process as well by giving

00:03:24.840 –> 00:03:28.620
them the knowledge bases and the prompting that they need to really get the right

00:03:28.680 –> 00:03:30.090
outcome from those engagements.

00:03:34.230 –> 00:03:37.230
So a lot of collections work, especially in the early stages, is highly

00:03:37.291 –> 00:03:41.101
transactional, which when you put that in the context of regulated work, means

00:03:41.130 –> 00:03:46.050
you’ve got the opportunity to really structure quick engagements with customers in

00:03:46.110 –> 00:03:47.880
terms of resolving their accounts.

00:03:47.940 –> 00:03:52.291
And you need to identify the opportunity to talk to humans as well, and AI enables

00:03:52.351 –> 00:03:53.520
that right from the outset.

00:03:58.080 –> 00:04:01.531
Well, this shift is really about accessing the additional data points AI can

00:04:01.590 –> 00:04:06.781
provide, especially around determining intent and emotion as really structured data

00:04:06.840 –> 00:04:11.910
points. If we can use those earlier in the process to identify potential risk, we

00:04:11.940 –> 00:04:16.410
can put intervention in place to actually prevent the potential risk of future

00:04:16.440 –> 00:04:22.020
delinquency. Using AI in customer risk profiling really supercharges

00:04:22.110 –> 00:04:27.150
the who, when, and how to contact customers as part of a personalized engagement

00:04:27.210 –> 00:04:31.380
strategy. The crucial bit, though, is to make all of those decision points really

00:04:31.470 –> 00:04:33.930
traceable and auditable as part of the process.

00:04:38.341 –> 00:04:43.380
Well, globally, both customers and regulators are raising the bar when it comes to

00:04:43.440 –> 00:04:49.020
strategies and outcomes. That’s across transparency, fairness, auditability, and

00:04:49.080 –> 00:04:53.490
personalization of approach, especially where it really affects those customers

00:04:53.550 –> 00:04:56.701
with financial difficulties or vulnerable circumstances.

00:04:57.420 –> 00:05:01.620
Well, the collections process needs to demonstrate, and crucially evidence, that

00:05:01.680 –> 00:05:07.020
it’s fully understood and reflected customer circumstances, especially where those

00:05:07.080 –> 00:05:11.281
circumstances are around affordability or vulnerability, as those make the biggest

00:05:11.341 –> 00:05:14.880
difference when it comes to ensuring that you have the right outcome for customers.

00:05:19.770 –> 00:05:21.841
Well, there’s one thing common about leaders in the market.

00:05:21.900 –> 00:05:25.680
They’re doing two things. One, they’re building in the governance from the outset

00:05:25.770 –> 00:05:28.951
to make certain that they have the right controls and auditability over what

00:05:28.980 –> 00:05:32.670
they’re doing. The second thing that those people are doing is thinking about how

00:05:32.730 –> 00:05:34.470
they embed it in their operating model.

00:05:34.830 –> 00:05:37.380
They’re not just doing it as a standalone exercise.

00:05:39.120 –> 00:05:42.630
Well, first thing, map where AI is used within the process.

00:05:42.720 –> 00:05:45.990
You might be surprised where it’s already used, and you might not be aware.

00:05:46.110 –> 00:05:50.701
Second, test that your MI that you currently have is fit for purpose against the

00:05:50.880 –> 00:05:54.420
new level of transparency and auditability that’s required.

00:05:54.930 –> 00:05:58.800
And crucially, most important, make certain that you wrap around the compliance

00:05:58.920 –> 00:06:02.310
framework to avoid future embedded problems.

00:06:07.230 –> 00:06:10.980
Well, there are multiple use cases for AI across the collections environment, but

00:06:11.010 –> 00:06:13.560
none of that matters if you don’t have the right guardrails in place.

00:06:14.370 –> 00:06:18.900
All the decision-making becomes auditable, and that integration back into a human

00:06:18.960 –> 00:06:21.510
involvement is absolutely embedded in the process.

00:06:22.140 –> 00:06:25.740
Well, the collections process will be far more predictive in nature.

00:06:25.831 –> 00:06:29.669
It’ll be hyper-personalized, and it’ll be much more embedded within the wider

00:06:29.730 –> 00:06:33.870
customer experience journey. It’ll deploy AI at multiple touchpoints, and

00:06:33.990 –> 00:06:38.550
crucially, that AI will help agents manage complex cases using the skills that

00:06:38.550 –> 00:06:39.630
they’ve been trained to do.

00:06:44.520 –> 00:06:48.240
Collections should never be seen as a back-end recovery function that just takes

00:06:48.300 –> 00:06:53.970
cost out of the business. It’s a strategic capability that drives real lifetime

00:06:54.061 –> 00:06:56.430
value from customers when it’s managed correctly.

00:06:57.750 –> 00:07:02.311
I’d start with prevention by design, cutting out delinquency from the outset.

00:07:03.030 –> 00:07:04.980
I’d then create a clean data environment.

00:07:05.040 –> 00:07:08.490
Data will drive all the decisioning going forwards, and that will also lead to

00:07:08.550 –> 00:07:12.630
personalized engagement. It’s very much a tech-led approach with humans in the loop

00:07:12.660 –> 00:07:14.730
for those moments that matter for those customers.

00:07:15.780 –> 00:07:18.780
Well, the market leaders will have deployed AI at scale.

00:07:19.170 –> 00:07:23.311
They’ll be taking every opportunity they can to broaden out the use cases across

00:07:23.401 –> 00:07:27.811
areas such as data analytics, customer engagement, and agent support.

00:07:27.870 –> 00:07:32.370
They’ll be deploying tools built on great data that provide great customer services

00:07:32.430 –> 00:07:33.750
that embed governance.