Key Takeaways
The collections landscape has reached a tipping point. Organisations face mounting pressure to maintain recovery rates whilst managing compliance obligations and operational costs. Traditional approaches—whether purely manual or rigidly automated—struggle to deliver the flexibility modern customers expect.
Our recent webinar revealed how forward-thinking collections teams are addressing these challenges through conversational AI agents that combine the empathy of human interaction with the scalability of automation.
Beyond Basic Automation: The Smart Digital Employee Approach
These aren’t your typical chatbots or IVR systems. Modern AI agents function as intelligent digital employees, combining sophisticated playbooks with natural language understanding. They recognise customer intent, adapt their responses accordingly, and determine appropriate next actions without following rigid scripts.
The underlying architecture leverages multiple AI technologies working in concert. Speech-to-text engines convert customer responses into analysable data, whilst AI models identify intent and drive decision logic. Advanced text-to-speech technology from providers like 11 Labs delivers natural-sounding conversations across eight languages, with polyglot support enabling seamless language switching mid-conversation.
Real-World Performance: The Numbers That Matter
The webinar showcased compelling case studies demonstrating tangible operational improvements:
- Customer Containment Success: A major collections organisation achieved 95% customer containment, meaning virtually all AI agent interactions reached structured outcomes without requiring human escalation. This dramatic reduction in agent workload enables teams to focus on complex cases requiring human expertise.
- Payment Plan Adoption: Of customers who engaged with AI agents, 58% successfully established payment plans. The technology’s ability to assess eligibility criteria and offer flexible arrangements in real-time significantly improved customer cooperation rates.
- Vulnerability Detection: 8% of customers indicated financial hardship during AI interactions, with conversations seamlessly transferred to specialist human agents. This proactive approach ensures compliance obligations are met whilst maintaining customer dignity.
The Technical Foundation: Security and Compliance First
These systems meet enterprise-grade security requirements with ISO 27001 and SOC 2 certifications, plus GDPR and HIPAA compliance. Data remains within campaign geography using Google Cloud Platform and Azure infrastructure, with encryption both at rest and in transit.
The architecture includes robust guardrails preventing hallucination or inappropriate responses. Deterministic logic ensures consistent outcomes, whilst fail-safe mechanisms transfer complex situations to human agents. No self-learning occurs in production environments, maintaining predictable behaviour patterns.
Operational Integration: Seamless Workflow Enhancement
AI agents integrate with existing CRM systems and collections platforms through APIs, supporting both standalone operations and CCaaS integration. CSV file imports enable rapid campaign deployment.
Pre-call SMS notifications improve contact rates, whilst identity verification protocols ensure regulatory compliance. The technology supports both inbound and outbound scenarios, from payment plan negotiations to routine payment processing.
Looking Forward: Strategic Implications
Collections leaders who attended the webinar recognised AI agents as more than operational tools—they represent strategic enablers for business transformation. The technology allows organisations to maintain high-volume operations without the need to bloat headcount, whilst maintain high quality customer experience and compliance outcomes.
The ability to conduct conversations at scale without scaling teams creates opportunities for revenue growth that were previously impossible. The technology particularly benefits organisations operating across multiple time zones or requiring 24/7 availability.nts—it’s orchestrating both to deliver exceptional customer experiences whilst achieving operational objectives.
Webinar Transcript
[00:08:00] Kayleigh Tait: Case for AI agents. , and also, it’s an always-on option, so….
[00:08:07] Kayleigh Tait: For those 24-7 collection use cases. It might be that customers do want to be reaching out to you, .
[00:08:17] Kayleigh Tait: Organizing payment plans. Outside of your normal working hours, and with AI agents, you can allow customers to do that.
[00:08:24] Kayleigh Tait: Or it might be that, actually, you’re working across multiple time zones, and you’re just looking for more efficient ways.
[00:08:29] Kayleigh Tait: To have those conversations across the 24. 7 periods. So, , it really offers you that flexibility.
[00:08:38] Kayleigh Tait: You’ve also got the, , I suppose the…. The knowledge that compliance.
[00:08:43] Kayleigh Tait: Is almost ticked off with the consistent scripting, audit trails. You can be sure in the knowledge that your AI agent is following through, , in a consistent way, and following your guidelines as an organization.
[00:09:00] Kayleigh Tait: , and when consumer trust. Is growing in technology. , it really helps when automation is done right. So, proving out, , how AI agents can work.
[00:09:12] Kayleigh Tait: , and, , work for your use case and your organization is a crucial part of this, but….
[00:09:18] Kayleigh Tait: I think most organizations are feeling margins being squeezed at the moment, and at looking at different ways.
[00:09:24] Kayleigh Tait: That you can be more efficient, so this is definitely one of them. So, I’ll hand over.
[00:09:30] Kayleigh Tait: To Matthew Yates, who will talk through AI agents and how they actually work.
[00:09:36] Kayleigh Tait: Great, thank you, Kayleigh Tait. …. Good morning, everybody, and fellows that are watching later on. Good afternoon, or good evening, maybe.
[00:09:44] Kayleigh Tait: , so yeah, really excited to talk to you today. , thanks for the intro.
[00:09:48] Kayleigh Tait: Okay, and yeah, if you could just click through to the next slide for us. , what I want to talk about over the next few slides is.
[00:09:55] Kayleigh Tait: Just providing a little bit more insight into how our AI agents actually work.
[00:10:00] Kayleigh Tait: , without going too deep, technically, into the underlying architecture, I’m going to talk through.
[00:10:05] Kayleigh Tait: The key technologies, and also some of the key considerations as well.
[00:10:09] Kayleigh Tait: Such as, , data privacy, guardrails when it comes to using generative AI.
[00:10:15] Kayleigh Tait: So, hopefully, you’ll find it useful over the coming slides, and then we’ll follow up with a really good set of demos as well, probably in about 5.
[00:10:22] Kayleigh Tait: To 10 minutes’ time. So…. I pulled this phrase together, , as a way of thinking about AI agents, , and I quite it because, , it positions an AI agent as really a very smart digital employee.
[00:10:36] Kayleigh Tait: That combines a playbook made up of rules and logic, with the AI natural language understanding.
[00:10:43] Kayleigh Tait: And so, it doesn’t need to necessarily just follow a script.
[00:10:47] Kayleigh Tait: To the exact wording. It can actually understand the customer intent as you’re progressing through the conversation.
[00:10:54] Kayleigh Tait: Understand that reason why the customer’s calling, and more importantly, what’s the next appropriate action to take following that understanding step.
[00:11:02] Kayleigh Tait: So, I quite that as a starting point for those of you on the call that are thinking, well, , what are these AI agents all about?
[00:11:10] Kayleigh Tait: , next please, Kay. Great. So….
[00:11:16] Kayleigh Tait: The AI agents are made up of a number of different technologies.
[00:11:21] Kayleigh Tait: There’s quite a complicated, complex architecture underpinning the technology. But to simplify it and up-level it a little bit, you can really break it down into what I see as 5 different areas.
[00:11:34] Kayleigh Tait: , very much following the flow of a conversation. On the next slide, I’ll show you a glimpse of the actual.
[00:11:42] Kayleigh Tait: Flow Builder that we have, and the engine that executes the flows, but at a very high level, the first step is really the user input.
[00:11:50] Kayleigh Tait: So, whether it’s an inbound call or an outbound call, the user input is that first step.
[00:11:56] Kayleigh Tait: And so, what we do is we use a number of different leading AI voice models.
[00:12:01] Kayleigh Tait: To start that engagement with the customer. And that’s a scripted engagement to start with.
[00:12:08] Kayleigh Tait: Throughout the conversation. We are using speech-to-text technology.
[00:12:13] Kayleigh Tait: And that’s converting those spoken words from the customer into written text format, so that at various points during the conversation flow.
[00:12:22] Kayleigh Tait: We can analyze what’s being said and determine the next appropriate step. And it’s really important at that stage that we have accurate, highly accurate.
[00:12:30] Kayleigh Tait: Transcription, because ultimately the data underpins everything that follows. So, once we start to transcribe the text of what the customer is actually saying to us, we then identify the intent. And this is where the AI.
[00:12:45] Kayleigh Tait: Elements start to come in. So, we use AI models to understand what the intent is, what it is the customer’s actually asking about.
[00:12:52] Kayleigh Tait: , and actually. Understanding that intent then drives the next step.
[00:12:58] Kayleigh Tait: And we combine the natural language understanding, intent identification, with. A decision logic. So, a number of people on this call will no doubt be familiar with.
[00:13:09] Kayleigh Tait: , conversational flows. Very familiar with IVR builders, for example. It’s quite similar in the background, where you’ve got a.
[00:13:16] Kayleigh Tait: Flow, which is a combination of decision logic, i.e, , rules and structured flows, and we are then using the AI to provide answers or outcomes based upon the customer intent understanding in the previous step.
[00:13:32] Kayleigh Tait: And then the final bit of it is, once we understand what the customer intent is, and we identify the appropriate next step.
[00:13:38] Kayleigh Tait: We use text-to-speech technology in order to communicate back to the customer and keep the conversation going and flowing naturally right the way through.
[00:13:48] Kayleigh Tait: , and what I will say is, , the technologies across all.
[00:13:51] Kayleigh Tait: Of this landscape of really…. Advanced significantly over the last few years, and I’ll talk a little bit more about some of the voice.
[00:13:59] Kayleigh Tait: Options in a couple of slides’ time. Next slide, please.
[00:14:07] Kayleigh Tait: So behind the scenes, , what’s driving these conversation flows, these AI agents, is.
[00:14:13] Kayleigh Tait: A flow builder. So… so we have an orchestration layer where you can actually go in and create your flows.
[00:14:22] Kayleigh Tait: And you’ll see if, , you just look closely at the screen, , it starts off with an introductory trigger event.
[00:14:28] Kayleigh Tait: We do some verification that we’re speaking to the right person.
[00:14:31] Kayleigh Tait: You’ll see this in play later on in the demo. And then we have GPT AI thinking nodes, and this is where the real artificial intelligence comes in.
[00:14:42] Kayleigh Tait: To decipher what it is the customer’s asking about, what the customer need and intent is, and then determine what the most appropriate outcome.
[00:14:49] Kayleigh Tait: To that step in the conversation flow is. So behind the scenes, this is the decision logic and the flow that we build, with the AI nodes in there.
[00:15:01] Kayleigh Tait: Next slide, please. So, voice customizations.
[00:15:07] Kayleigh Tait: , so, . Voice customizations, they’ve been around for a long time, , text-to-speech technology’s been around for a long time.
[00:15:15] Kayleigh Tait: But definitely over the last, probably, 3 to 5 years in particular.
[00:15:20] Kayleigh Tait: There’s been a real increase in not just the. The accuracy of, , text-to-speech, but the amount of options you have when it comes to choosing an appropriate voice.
[00:15:34] Kayleigh Tait: For a specific task. Whether that’s, , outside of work, whether that’s, , doing a podcast, for example, or, , reading a book, whatever it may be, there’s an appropriate voice out there.
[00:15:44] Kayleigh Tait: So, , for your businesses. We offer out-of-the-box a number of different voice model options. , we use industry-leading technologies from 11 labs.
[00:15:54] Kayleigh Tait: And we have a number of options out of the box, but if you go on the 11 Labs website and have a look at the voice options available there, we can incorporate any of those voices into the conversation flow itself.
[00:16:06] Kayleigh Tait: We also use Cartesia, DeepRAM, Azure, and Google. , so there’s a wide range of voices to fit your customer brand.
[00:16:14] Kayleigh Tait: , or you could choose to actually create your own voice.
[00:16:18] Kayleigh Tait: So, with 11 labs, , the partnership we have there. You can actually go in there and clone a voice as well. So you could actually have a clone voice.
[00:16:26] Kayleigh Tait: That purely represents your organization or department within the organization, for example.
[00:16:32] Kayleigh Tait: , we have multi-language support in there, so we currently support 8 languages, which are on screen.
[00:16:38] Kayleigh Tait: And we offer what’s called polyglot support, so some of the voice models.
[00:16:44] Kayleigh Tait: Are intelligent enough to be able to auto-detect the spoken language. So conversation may start off in English.
[00:16:51] Kayleigh Tait: And then switched to French. The model can detect that, and then continue to engage.
[00:16:57] Kayleigh Tait: With the customer in their native language. So, yeah, really big advantages in this area over the last couple of years that we’ve incorporated into this technology through the Curious Thing acquisition.
[00:17:09] Kayleigh Tait: , next step, please. So, guardrails. So, within the Flow Builder, we have the AI nodes, .
[00:17:20] Kayleigh Tait: And what we make sure happens within those nodes is we use as much as possible.
[00:17:26] Kayleigh Tait: Deterministic logic. So, we ensure that we are compliant, and I’ll talk about compliance a little bit more in a slide’s time.
[00:17:34] Kayleigh Tait: , but we have deterministic logic so that the AI steps.
[00:17:39] Kayleigh Tait: Have a number of set outcomes. And always have a fail-safe to drop through to a human agent.
[00:17:45] Kayleigh Tait: So, , if the AI model doesn’t quite understand what the customer is actually saying, they’ll ask them politely to rephrase.
[00:17:55] Kayleigh Tait: What they’ve actually just said, and again, if they can’t understand that, they will drop through to human. So, you won’t find these models hallucinating, making things up.
[00:18:05] Kayleigh Tait: , to actually try and appease the customer. , so we use very well-crafted and very well-tested.
[00:18:12] Kayleigh Tait: Prompts and the parts of the conversation history. In order to build up those GPT nodes.
[00:18:20] Kayleigh Tait: And then, most importantly, I would say, is there’s no self-learning in production.
[00:18:24] Kayleigh Tait: So any of the context that we do use in the AI models, .
[00:18:29] Kayleigh Tait: That doesn’t… is not used to train the AI models. It’s literally just fed in, and the outcome is fed out, and we use the outcome, and we continue down.
[00:18:37] Kayleigh Tait: The whole conversation flow is not held in memory, in context. Each node is….
[00:18:44] Kayleigh Tait: Localized, and it’s just the content within that node that we actually then use to send to the AI model and generate an output.
[00:18:52] Kayleigh Tait: So that provides control, predictability, and auditability as well. Next slide.
[00:19:01] Kayleigh Tait: Integrations. So, we….
[00:19:05] Kayleigh Tait: And you’ll see this in the demo later on. , in terms of importing, , customer lists.
[00:19:10] Kayleigh Tait: Into the solution. You can do it via a CSV file. We do have, , API.
[00:19:17] Kayleigh Tait: Integrations available as well, so you can seamlessly import your customer lists, much in a similar way as many of you do today with our CCAS solution.
[00:19:27] Kayleigh Tait: , we will be integrating, of course, with Max Contact, , and other CCAS vendors, and CRM integrations are something that we do support as well, and.
[00:19:36] Kayleigh Tait: Really, there’s any integration possible. So, , with the APIs available today, we can pretty much integrate with any other solution out there as we need to.
[00:19:48] Kayleigh Tait: And the key point, I think, with. The AI agents that’s worth mentioning is.
[00:19:54] Matthew Yates: They do operate standalone. So, yes, you’ll see that we integrate with Matthew Yates’s Contact CCaaS solutions.
[00:20:01] Matthew Yates: But, if there are departments within your business that aren’t part of the contact center.
[00:20:05] Matthew Yates: They can use these AI agents in standalone mode. So you could build out an AI inbound assistant to handle certain tasks within the departments. Likewise, an outbound campaign.
[00:20:17] Matthew Yates: Could be built out and executed solely within. The AI agent’s solution without the need for the CCaaS solution alongside it.
[00:20:26] Matthew Yates: So, I think that’s kind of really exciting, hopefully, for some of.
[00:20:29] Matthew Yates: The customers out there that use Macs today. To actually think about how they could apply AI agents throughout the rest of the business.
[00:20:37] Matthew Yates: Next slide, please. So, security and data privacy, ….
[00:20:45] Matthew Yates: We are certified, , so Matthew Yates’s contact has been certified for a number of years with ISO 27001, Cyber Essentials Plus.
[00:20:53] Matthew Yates: With the curious thing acquisition, , they too, our ISO certified, they’re also SOC 2.
[00:20:59] Matthew Yates: , type to attestation compliant, and they do. Comply with GDPR and HIPAA.
[00:21:05] Matthew Yates: Laws as well, and regulations as well. , the initial default.
[00:21:11] Matthew Yates: Identity and verification checks are a simple name and phone number.
[00:21:16] Matthew Yates: But we can go beyond that. So, , if you’re interested in using Curious Thing and doing more advanced ID and V checks.
[00:21:25] Matthew Yates: That will be part of the initial scoping phase around building out the conversation flow.
[00:21:30] Matthew Yates: With yourself, and we can do more advanced ID&V checks as we need to.
[00:21:35] Matthew Yates: , all the data remains within. The geography, the campaign country that you operate in.
[00:21:41] Matthew Yates: , and that’ll be within the Google Cloud Platform and Azure as well.
[00:21:47] Matthew Yates: And just you’ve got with the call data today and other data within the MAC CCAS product today.
[00:21:54] Matthew Yates: The data is yours, ultimately. , we are the responsible custodian.
[00:21:58] Matthew Yates: That you can trust, and we will, , purge the data based upon the retention policies agreed with yourself.
[00:22:06] Matthew Yates: , the data is secure, so it’s encrypted both at rest.
[00:22:12] Matthew Yates: And in transit as well. Next slide, please.
[00:22:20] Matthew Yates: Okay, so before we hand over to Todd to take us through the demos.
[00:22:25] Matthew Yates: , I just thought I’d quickly step through. A slightly more in-depth version of that first slide that talked through the different technologies.
[00:22:33] Matthew Yates: As you’ll see in the green box at the top. What we found, , through working with the Curious Thing team.
[00:22:40] Matthew Yates: Is that it’s good practice to preempt. The AI voice conversation, , and one hour before, we recommend sending out an SMS message to inform the customer that the AI assistant will be calling them within the hour.
[00:22:56] Matthew Yates: Then, when they receive the call. , we start to work our way down that conversation, decision logic flow.
[00:23:04] Matthew Yates: If the check is false and it fails, then we can transfer that person to a human agent in the.
[00:23:11] Matthew Yates: Team, , and they can carry on via a human-to-human interaction.
[00:23:17] Matthew Yates: Or, obviously, if the IDMV check is positive. Then we can continue down the flow and offer a number of different options based upon what it is that you want to offer.
[00:23:26] Matthew Yates: And again, you’ll see this later on. Based upon the understanding of the intent.
[00:23:32] Matthew Yates: That the customer provides. We’ll then continue down the flow, and the outcome will be one of the items on the bottom of the list.
[00:23:40] Matthew Yates: I will reiterate that at any point in this flow where AI.
[00:23:45] Matthew Yates: Voice mode, or doesn’t understand what the agent is… sorry, what the human customer is actually saying.
[00:23:51] Matthew Yates: Or the sense frustration. Or identify a vulnerability, for example.
[00:23:56] Matthew Yates: They can drop through to a human agent at any point during the conversation flow.
[00:24:01] Matthew Yates: And Todd will demo that for you shortly. So, that was it for me, keeping it fairly high level, but hopefully you found it useful.
[00:24:10] Matthew Yates: And if you’ve got any questions, please post in the chat and we can cover them towards the end.
[00:24:15] Matthew Yates: Great, thanks, Matthew Yates. Good, yeah, so, Matthew Yates said, , Todd in our team has prepared 3 demo scenarios for you.
[00:24:24] Matthew Yates: Today. , the first one is around making an outbound call.
[00:24:29] Matthew Yates: , so it goes through the process of a customer arranging a payment plan.
[00:24:35] Matthew Yates: , IDMV, what happens when their details aren’t correct when they input those.
[00:24:41] Matthew Yates: , the second journey is around vulnerability detection. Making sure that the right, , the right conversations are getting through to your agents, and they’re not overwhelmed with.
[00:24:53] Matthew Yates: , thousands of conversations that don’t need to be a conversation.
[00:24:57] Matthew Yates: , and then lastly, a more simple routine… automation of routine tasks, which is, , making a payment.
[00:25:05] Matthew Yates: So, the three different scenarios could all be part of your customer journey. We’ve just taken those in a.
[00:25:11] Matthew Yates: Linear fashion to show those to you, and you can see.
[00:25:15] Matthew Yates: , what is available in AI Agents?
[00:25:25] Matthew Yates: Hey, everyone. So today I’m going to be showcasing Curious Thing.
[00:25:29] Matthew Yates: I go to my outbound and inbound calls. All based on a scenario of debt collection.
[00:25:34] Matthew Yates: They’ll result in different outcomes. So I’m going to begin with an outbound call, whereby we’re going to set up a payment plan.
[00:25:42] Matthew Yates: So, to begin, I just need to upload some data. So I will select my outbound campaign.
[00:25:50] Matthew Yates: Click Upload, select Upload List. Here, you can name the list, schedule the time to call it.
[00:25:56] Matthew Yates: I go into begin, and attach your file. So I’ll attach my file….
[00:26:03] Matthew Yates: You can…. Download the sample file here, which will contain all the headers you’ve got pre-configured.
[00:26:11] Matthew Yates: But I will just upload the one I’ve already prepared. So we can see now this is scheduled.
[00:26:19] Matthew Yates: A pre-call SMS is sent. , in this configuration, it’ll send just a moment before the call comes through.
[00:26:25] Matthew Yates: You can configure this to send however long you , an hour, two hours, a day before you call.
[00:26:30] Matthew Yates: It just helps with contact rates. So we’ll just wait for that to come through.
[00:26:38] Matthew Yates: Okay, so I’ve got that text message now, I should receive the call shortly.
[00:26:43] Matthew Yates: Okay, got a call. Hi, Todd. This is a call from Max Contox.
[00:26:51] Matthew Yates: We send you a text message earlier, we were speaking. Scott, can you please confirm if I am speaking with Todd Lellin?
[00:26:59] Matthew Yates: Yes, you are. Okay.
[00:27:05] Matthew Yates: I’m calling from Max Hunter. Regarding your account with Cloud Broadband that we are handling.
[00:27:10] Matthew Yates: This conversation will be recorded for training and quality assurance purposes.
[00:27:16] Matthew Yates: Before I get into the details about your cloud broadband account.
[00:27:18] Matthew Yates: I want to protect your privacy. By testing a few details with you.
[00:27:23] Matthew Yates: Please state your year of birth. 1996.
[00:27:28] Matthew Yates: Okay. It doesn’t match with the record in the system.
[00:27:33] Matthew Yates: Please try again using your keypad. For example, if you were born in 1992.
[00:27:38] Matthew Yates: Please use your keypad to enter 1-992. Thank you.
[00:27:48] Matthew Yates: Okay. We have successfully verified your date of birth. Thank you for that.
[00:27:54] Matthew Yates: Next, can you please state your postcode? For example, M32 JA.
[00:28:00] Matthew Yates: M12ab. Okay, thanks.
[00:28:07] Matthew Yates: We have successfully verified your postcode. We have now completed all necessary verification steps.
[00:28:15] Matthew Yates: The current overdue balance on your account is £175.60. 2 on the 1st of June, 2025.
[00:28:22] Matthew Yates: Are you ready to pay now, or do you want to set up a personalized payment plan?
[00:28:27] Matthew Yates: Can I pay monthly, please? Okay, thanks.
[00:28:33] Matthew Yates: You want to set up a payment plan. We’ll now check your eligibility to do so.
[00:28:38] Matthew Yates: At the moment. You are eligible to set up a payment plan on the phone with us now.
[00:28:43] Matthew Yates: As mentioned earlier, your balance is currently £125.50. Can you please confirm how many monthly installments you would to split this amount over?
[00:28:57] Matthew Yates: Okay. You have chosen to pay your balance over the next 6 months.
[00:29:02] Matthew Yates: Is this correct? Yes. That’s right.
[00:29:08] Matthew Yates: Okay. Thank you for speaking with me today.
[00:29:12] Matthew Yates: To recap, you have agreed to the payment plan covering your balance of £125.50 over next.
[00:29:18] Matthew Yates: 6 months. We have sent an SMS detailing this information to you.
[00:29:23] Matthew Yates: We will be in touch closer for your first payment date.
[00:29:26] Matthew Yates: Is there anything else I can help you with? No thanks.
[00:29:31] Matthew Yates: Okay, thanks. If you need any further support, please call our team directly on 0161.
[00:29:39] Matthew Yates: 142054. That number then was 0161.
[00:29:45] Matthew Yates: 0143054. Enjoy the rest of your day. Goodbye.
[00:29:54] Matthew Yates: Okay, so that concludes the call. As you can see it caught that bound to me, I answered the phone, and it stated the reason for call, and performed an identity check. I confirmed it was speaking to the right person.
[00:30:08] Matthew Yates: And then it forwards some extra checks. It asks my date of birth, or my year of birth, sorry.
[00:30:14] Matthew Yates: , I specifically gave it the incorrect information there. And it asked me to retry using….
[00:30:20] Matthew Yates: And the keypad on my phone. And then I entered the correct information.
[00:30:24] Matthew Yates: It matched, and it’s sort of successful. We’ll need a postcode check, which it matched as well.
[00:30:30] Matthew Yates: And then from there, it stated the balance owed and requested payment.
[00:30:35] Matthew Yates: I requested to pay monthly and set up a payment plan, and in the background there, it did some checks in the file we uploaded.
[00:30:43] Matthew Yates: , we had eligibility for monthly payments or not, and this was….
[00:30:47] Matthew Yates: Am allowed to do so, so I’ll pass that eligibility. And then I asked how long I wanted to pay over, so I gave 6 months.
[00:30:56] Matthew Yates: Confirm that detail, and from there, it summarized the terms, sent me an SMS.
[00:31:02] Matthew Yates: Detailing those terms. And the call ended.
[00:31:11] Matthew Yates: Okay, so the next example will be another outbound call, whereby, once they’re requested, the payment’s requested, the customer’s going to state of vulnerability.
[00:31:21] Matthew Yates: And, , it’s important that that’s picked up on, and it’ll be transferred through to a human who will.
[00:31:27] Matthew Yates: Can help. So, similar to the previous call, outbound, we’ll upload, hit the….
[00:31:33] Matthew Yates: The data…. And schedule that call for now.
[00:31:41] Matthew Yates: And we’ll just wait for the pre-call SMS and the call to come through.
[00:31:47] Matthew Yates: So, again, I’ve received the pre-call SMS, and the call will follow shortly.
[00:31:53] Matthew Yates: Okay, so I’ll answer. Hi, Todd. This is a call from Max Contact.
[00:32:01] Matthew Yates: We sent you a text message earlier letting that we would be calling. Before we start, can you please confirm if I am speaking with Todd Vela?
[00:32:09] Matthew Yates: Yes, you are. Okay, thanks.
[00:32:15] Matthew Yates: I’m calling from Max Contact. Regarding your account with Cloud Broadband that we are handling.
[00:32:20] Matthew Yates: This conversation will be recorded for training and quality assurance purposes.
[00:32:25] Matthew Yates: Before I get into the details about your cloud broadband account.
[00:32:29] Matthew Yates: I want to protect your privacy by checking a few details with you.
[00:32:33] Matthew Yates: Please state your year of birth. 1997.
[00:32:39] Matthew Yates: Thanks. We have successfully verified your date of birth. Thank you for that.
[00:32:45] Matthew Yates: Next, can you please state your postcode? For example, M32JA. M12ab.
[00:32:55] Matthew Yates: Okay, thanks. We have successfully verified your postcode.
[00:33:01] Matthew Yates: We have now completed all necessary verification steps. The current overdue balance on your account is £125.50.
[00:33:09] Matthew Yates: Due on the 1st of June 2025. Are you ready to pay now, or do you want to set up a personalized payment plan?
[00:33:18] Matthew Yates: No, my, , disability at the moment’s really stopped me from working, and I can’t afford it right now.
[00:33:26] Matthew Yates: Thanks. I understand there may be circumstances out of your control that are preventing you from paying your balance.
[00:33:34] Matthew Yates: Please hold the line while we transfer you to an expert who can help.
[00:33:38] Matthew Yates: Thanks for chatting with me. Goodbye. You have reached specialist support.
[00:33:44] Matthew Yates: Please hold the line while we route you to someone who can help.
[00:33:52] Matthew Yates: Okay, so, similar to the previous outbound call, we received the pre-call SMS, and we answered the phone, it stated the reason for the call, checked my identity, and confirmed some details for verification.
[00:34:07] Matthew Yates: And then requested the balance payment from the balance owed, and I then clearly stated a vulnerability. I identified that, and then transferred me through.
[00:34:17] Matthew Yates: Where I could speak to a person.
[00:34:22] Matthew Yates: Okay, so now we’re going to do an inbound call, whereby we’re going to make a payment over the phone.
[00:34:29] Matthew Yates: So, I’ll begin by calling inbound.
[00:34:36] Matthew Yates: Hello. Welcome to Max Contact Debt Management Services.
[00:34:42] Matthew Yates: I’m a digital assistant, and can provide assistance with several aspects of managing your payment balance.
[00:34:48] Matthew Yates: What can I do for you today? Can I make a payment, please?
[00:34:54] Matthew Yates: Okay, thanks. Absolutely. I can help you make a payment now. I’ll send you a text message with a secure link to our online payment portal.
[00:35:04] Matthew Yates: One moment, please. Please click on the link and follow the instructions to select a payment option.
[00:35:10] Matthew Yates: We’ll hold the line while you pay, if you need help, please just ask.
[00:35:14] Matthew Yates: Please let me know once you have completed the payment.
[00:35:20] Matthew Yates: Okay, so I’ve received that SMS, and I’m going to, , click on the link, and make that payment.
[00:35:30] Matthew Yates: So, I’m gonna say now I’ve completed the payment. Okay, I’ve done that now.
[00:35:38] Matthew Yates: Okay, thank you. We appreciate your help sorting this out today.
[00:35:45] Matthew Yates: You’ll receive confirmation via your specified contact preferences shortly. Is there anything else I can help you with today?
[00:35:51] Matthew Yates: No, that was all, thank you. Okay, thanks.
[00:35:57] Matthew Yates: If you have any other questions in the future, feel free to reach out to our team directly on 0161.
[00:36:03] Matthew Yates: 814-2054. That number, again, was 0161-8142054. Thanks so much for speaking with me.
[00:36:15] Matthew Yates: Enjoy the rest of your day. Goodbye. Okay, so that was the inbound call to make a payment. , I called inbound, I stated my intent, I wanted to make a payment.
[00:36:27] Matthew Yates: Checked if I was on a mobile, which I was, and it sent an SMS.
[00:36:32] Matthew Yates: If I was calling from a landline, well, what it would have done is it identified that, and it would have read out the link to the payment portal instead. But as I received that SMS.
[00:36:41] Matthew Yates: I clicked on the payment link, made the payment, and confirmed I did so.
[00:36:46] Matthew Yates: And then, after you need anything else? And I said no, so it ended the call.
[00:36:57] Matthew Yates: Great stuff. So, hopefully that’s given you some good ideas of what… a flavor of what is possible with AI voice agents. Obviously, for those specific use cases.
[00:37:09] Matthew Yates: , I’m just gonna spend a few minutes talking through some real-life examples of customers.
[00:37:16] Matthew Yates: That have had success. , in this space using AI voice agents and how they’ve been using it.
[00:37:22] Matthew Yates: , and the typical results that they’ve been seeing. So, the first one is very similar to the first use case that Todd showed.
[00:37:31] Matthew Yates: , this is an organization that are, , their goal is to use the AI voice agent to expand beyond the routine payment use case.
[00:37:42] Matthew Yates: , really, , to…. Try and, , ensure that most of the conversation that, , your real-life agents would be having is handled first by the AI voice agent, so….
[00:37:55] Matthew Yates: , it can pivot between the taking an immediate payment, , setting up payment plans.
[00:38:01] Matthew Yates: , detecting hardship or vulnerability, , and a result of that is then, , what I’ll share on this next stage, this next page, sorry.
[00:38:12] Matthew Yates: Is, , that 95% of the customers that engaged with the AI agent itself.
[00:38:19] Matthew Yates: Reached a structured outcome without the need for human escalation. So that is an absolutely huge stat.
[00:38:26] Matthew Yates: , where, , you are able to contain quite a lot of the conversation that would typically reach a human.
[00:38:33] Matthew Yates: , all within the AI agent itself. , this specific customer had an 8% engagement rate.
[00:38:42] Matthew Yates: , across all the customers that they were reaching, , with the AI agent itself.
[00:38:47] Matthew Yates: And 58% of customers set up a payment plan, , who engage with the AI agent.
[00:38:53] Matthew Yates: Showing that AI agent, , the strength that it has to enable those flexible.
[00:38:58] Matthew Yates: Payment options, …. It really helped contain that conversation.
[00:39:04] Matthew Yates: , 8% of customers indicated hardship and provided further details. , ensuring that financial distress was proactively addressed and documented.
[00:39:16] Matthew Yates: And at that point, it is then able to be picked up further.
[00:39:19] Matthew Yates: So, it’s not only improved, I suppose, operational efficiency, , and reduced that human work.
[00:39:26] Matthew Yates: But it supports compliance as well. , and in a lean and automated way.
[00:39:31] Matthew Yates: , so a customer, . Another customer, they’re using the AI agent in a slightly different way.
[00:39:39] Matthew Yates: They’re having conversations at scale. , without having to scale their team, so almost acting as a triage for the conversations.
[00:39:47] Matthew Yates: Before they become more complex. , which is great when you’re operating on those smaller margins.
[00:39:55] Matthew Yates: , for organisations. , the typical results that they’re seeing is that.
[00:40:00] Matthew Yates: , 14% of all engaged customers completed a payment, and it was actioned.
[00:40:07] Matthew Yates: , 48% of engaged customers won’t ready to make a payment there and then, and wanted to speak to someone further.
[00:40:14] Matthew Yates: So the AI’s job on this one, , was around filtering and routing.
[00:40:20] Matthew Yates: The conversations to where they needed to be. , it reduced the need for conducting repetitive.
[00:40:28] Matthew Yates: , tasks, and having conversations that aren’t driving revenue for your organization.
[00:40:32] Matthew Yates: And that was over 70% of the work that the agent, , AI agent undertook.
[00:40:38] Matthew Yates: , it’s across Max Contact AI Agents, it engaged in nearly 1 in 6.
[00:40:44] Matthew Yates: Customers, , it resolved 250 payments directly. , and identified over 950 accounts that required that follow-up, , further, , and further investigation from.
[00:40:58] Matthew Yates: The human team. So, it enables this customer to maintain high-volume debt operations at lower costs.
[00:41:07] Matthew Yates: While they still ensured that they had, . Compliant… a compliant escalation process in place for hardship, disputes, and non-paying customers as well.
[00:41:17] Matthew Yates: So that just gives you another flavor of how, , the different types of use cases and the typical results that you might see.
[00:41:24] Matthew Yates: Across this space, , in debt collection. Great, so we’ll go to, , some questions. I think there is a few in the chat.
[00:41:35] Matthew Yates: Now I’ll stop sharing my screen. Let’s see what we have here….
[00:41:44] Matthew Yates: , so we’ve got, how does the payment plan information. Pass to systems, , debt collection systems, ARCA.
[00:41:53] Matthew Yates: , Matthew Yates, is that something that you’re able to provide a bit more context to?
[00:42:00] Matthew Yates: Yeah, sure. So, , I assume Orca is a type of CRM.
[00:42:06] Matthew Yates: And we can integrate with. Pretty much any CRM, , that would be part of the initial scoping phase.
[00:42:15] Matthew Yates: Or, , for any integration that. But yes, in theory, , as you’re going through the conversation flow, , at the point or at the end of the conversation, we would send an update.
[00:42:27] Matthew Yates: Relevant systems as needed with the required information.
[00:42:32] Matthew Yates: Great, thanks, Matthew Yates. , what happens if the AI doesn’t know what to do?
[00:42:38] Matthew Yates: I just don’t understand the customer.
[00:42:40] Matthew Yates: So, it can all be configured, but, , the way we currently configure it is for the AI to ask.
[00:42:48] Matthew Yates: The person to rephrase. What they’ve just said. , the AI model will then attempt to understand.
[00:42:56] Matthew Yates: What has just been said back to it, and if it isn’t able to do so, it will pass through to a human agent.
[00:43:04] Matthew Yates: Thanks, Matthew Yates. . What speech-to-text and text-to-speech models do your.
[00:43:12] Matthew Yates: Agents use….
[00:43:15] Matthew Yates: Okay, so…. Speech-to-text, we use primarily use DeepGrum.
[00:43:21] Matthew Yates: , that’s a product that we’ve been using for quite a while now internally.
[00:43:26] Matthew Yates: Within the speech analytics spoken. Solution that we launched last year.
[00:43:31] Matthew Yates: , very high quality transcription. And then, in terms of the text-to-speech.
[00:43:36] Matthew Yates: We use 11 labs, , that’s probably the most popular, I would say.
[00:43:41] Matthew Yates: , it certainly provides the widest range of voices. Dialects, languages, and….
[00:43:50] Matthew Yates: As well as that, we use Azure…. Google, Cartesia.
[00:43:54] Matthew Yates: , , steep grum as well, actually, so yeah. You find a lot of providers offer different voices, and there is so much variation.
[00:44:02] Matthew Yates: I’d encourage everybody to go out there and have a look, and….
[00:44:05] Matthew Yates: , if you’re interested in taking this further, certainly think about what type of voice would best fit your company and the use case.
[00:44:16] Matthew Yates: , great. Thanks, Matthew Yates. …. Next up is, what is the future of the AI roadmap?
[00:44:25] Matthew Yates: Ooh, good question. So…. We have, yeah, big plans for incorporating AI, , not just continuing with AI development within.
[00:44:35] Matthew Yates: The AI agents, …. But also continuing with our spoken analytics solution, and starting to incorporate more AI into our CCAS solution as well going forward.
[00:44:48] Matthew Yates: But in the near term, we’ll be looking at things , .
[00:44:52] Matthew Yates: A conversational AI chatbot. So, the… the Flow Builder.
[00:44:58] Matthew Yates: We can use to also create chatbot-based workflows, and incorporate AI nodes into there as well.
[00:45:05] Matthew Yates: So that’s something we’re currently scoping out right now, with a view to starting that work and getting something working this year.
[00:45:12] Matthew Yates: , so that’s really exciting. But, , yeah, lots of… lots of improvements, as well as, , the chatbot workflow will be….
[00:45:19] Matthew Yates: Continuing to develop the…. Ai agents, the conversational voice AI agents as well.
[00:45:27] Matthew Yates: Good, thanks, Matthew Yates. , a couple of more questions around. Reporting, , can you create custom reports.
[00:45:36] Matthew Yates: As you can in the Max Contact platform. And linked to that.
[00:45:41] Matthew Yates: , our reports standalone at the moment?
[00:45:47] Matthew Yates: So, reports are within the portal, so if you…. Ask us to configure an AI agent for you today. Once that’s live.
[00:45:58] Matthew Yates: You’ll get access to a portal. So, if you remember in the demo that Todd shared a few minutes ago.
[00:46:04] Matthew Yates: He showed the importing of a CSV file, a customer list.
[00:46:08] Matthew Yates: So that portal enables you to import the customer data. But also, there’s a reporting.
[00:46:14] Matthew Yates: Module in there as well, so you can actually see a breakdown, a dashboard view.
[00:46:18] Matthew Yates: Of, , calls. Over a period of time, the more granular breakdown of call outcomes, for example.
[00:46:26] Matthew Yates: And you can click through and view the individual. Conversations, the full transcripts are in there, the call recording is in there as well, so you get the full.
[00:46:37] Matthew Yates: For you within there. , that isn’t configurable today. , there is an option to export out of there.
[00:46:43] Matthew Yates: But, , yeah, that might be something if people feel it is useful, we can look at incorporating in the future the ability to, , a bit some customers do today.
[00:46:53] Matthew Yates: With the, , , the replication database with our core CCAS product, you can connect to that and pull the data out and crunch it as you wish.
[00:47:01] Matthew Yates: In Power BI, or Tableau, or any other tool. , that might be something we want to do as well in the future with the curious thing AI agents.
[00:47:09] Matthew Yates: Cool, thanks, Matthew Yates. , link to that is, …. Is there a plan for the….
[00:47:16] Matthew Yates: All that information to be in the Max Contact platform. So the reporting information.
[00:47:22] Matthew Yates: So as part of the integration with the CCaaS products, , that we have, there will be updates to.
[00:47:30] Matthew Yates: Some of the database tables that we have in Max Contact.
[00:47:34] Matthew Yates: So, you can see the full interaction history. , including the AI agent interactions. , that’s something we are currently, again, currently scoping out.
[00:47:45] Matthew Yates: With a view to working on that in the second half of this year.
[00:47:50] Matthew Yates: Good. I think that’s it for questions. Usually I say that and then a few more pop in, but….
[00:47:56] Matthew Yates: I’ll just give it a one. Maybe 10 more seconds.
[00:48:01] Matthew Yates: Yeah, , , there is quite a lot to take in, but I think, as Kay touched on.
[00:48:05] Matthew Yates: There’s a lot of opportunity outside of these use cases we’ve talked about today, and if people do have.
[00:48:11] Matthew Yates: Any questions, then obviously contact any of us on the call, or your account manager.
[00:48:16] Matthew Yates: And, , we’ll be happy to follow up.
[00:48:20] Matthew Yates: Cool, okay. Yeah, so it looks , …. That’s it for questions right now, so Matthew Yates said, , we’ll send out the follow-up email with a link to this recording.
[00:48:31] Matthew Yates: So you can share it across your organization if you wish.
[00:48:34] Matthew Yates: , and if you want to know how AI voice agents can help you with this specific use case.
[00:48:40] Matthew Yates: Whether that’s increasing debt collection rates, or control your costs, help with compliance.
[00:48:45] Matthew Yates: Then absolutely reach out to one of the team as well. We can set up a customized demonstration for you.
[00:48:51] Matthew Yates: You can, , be that’s hot on the other side of the phone as well, so you can experience it yourself.
[00:48:57] Matthew Yates: , and one of the team will be more than happy to organize that for you, so….
[00:49:03] Matthew Yates: Thanks very much for your attention today and for your questions. It’s always good to get some at the end, and thanks, Matthew Yates.
[00:49:09] Matthew Yates: , for talking us through how AI agents work. Nice to see everyone. I’ll speak to you soon.
Your Questions Answered
Q: How does payment plan information pass to debt collection systems like ORCA?
A: We can integrate with virtually any CRM or debt collection system through APIs. This would be configured during the initial scoping phase. As customers progress through the conversation flow, we send real-time updates to relevant systems with all required information—whether that’s payment plan details, customer contact outcomes, or vulnerability flags.
Q: What happens if the AI doesn’t understand the customer?
A: The system includes multiple fail-safes. First, the AI agent politely asks the customer to rephrase their response. If understanding still isn’t achieved on the second attempt, the conversation transfers seamlessly to a human agent. This ensures customers never feel frustrated or trapped in an unhelpful loop.
Q: What speech-to-text and text-to-speech models do your agents use?
A: For speech-to-text, we primarily use Deepgram—the same technology powering our speech analytics solution launched last year. It delivers highly accurate transcription essential for understanding customer intent. For text-to-speech, we partner with 11 Labs as our primary provider, offering the widest range of voices, dialects, and languages. We also utilise Azure, Google, Cartesia, and Deepgram to provide voice options that best fit each organisation’s brand and use case.
Q: What does the AI roadmap look like for MaxContact?
A: We have ambitious plans for AI integration across our entire platform. In the near term, we’re developing conversational AI chatbots using the same Flow Builder technology that powers voice agents. This will enable text-based customer interactions with the same intelligent decision-making capabilities. Beyond that, we’re expanding AI features within our core CCaaS solution and continuing to enhance our speech analytics capabilities. The chatbot functionality is currently in development with a target launches this year.
Q: Can you create custom reports as you can in the MaxContact platform? Are reports standalone at the moment?
A: Currently, AI agent reporting exists within a dedicated portal where you can view call breakdowns, conversation outcomes, full transcripts, and call recordings. While this isn’t customisable today, there are export options available. We’re actively scoping integration with the main MaxContact platform to include AI agent interaction history alongside your existing reporting. This work is planned for the second half of this year, enabling unified visibility across all customer touchpoints.
Q: Will AI agent information eventually appear in the MaxContact platform?
A: Absolutely. As part of our CCaaS integration roadmap, we’re updating database tables within MaxContact to capture complete interaction histories, including AI agent conversations. This will provide seamless visibility of the entire customer journey—from AI agent interactions through to human agent escalations—all within your familiar MaxContact reporting environment.
Ready to explore how AI agents could transform your collections strategy?
Contact our team to arrange a customised demonstration where you can experience the technology firsthand.t the best chance for a successful interaction.
