The Future of QA: Why AI is a Game-Changer for Contact Center Quality in 2025

00:00You've heard me break down contact-centered technology on the podcast time again. Now it's time for me to help you with challenges. As your contact center technology advisor through our new company Expivia Digital, I can help you with things like CCaaS selection, any type of implementation, It's the same honest advice you hear on the geek applied specifically to your operation. Visit ExpiviaDigital.com to get started. This is advice from a call center Geek, a weekly podcast with a focus on all things call center will, we'll cover it all from call center operations, hiring, culture, technology, technology, and education. We're here to I give you actionable items to improve the quality of yours and your customers experience. This is an involving and discreet with creative the mind and ambitious people like this guy not only is his pass, his a pass, his a pass, his a pass pass, his a pass, his a pass, his a pass, his from a call center, the call center contact center, we try to give you some actionable items, take back in your contact center, improve your role, quality, improve the aging experience, hopefully improve your customer experience as well.

01:13How's everybody doing? We, if you've been following me at all, it is my birthday. I don't know why I was looking at the date here. It's December 4th. And we have just had an eerie five days of unbelievable snow. our all-time 24-hour record, and that's at the airport, which is always a little bit less than ever. We have 23 in 24 hours. We have about at my house, about 50 inches, right? And we're supposed to have a little bit of a lull today and then you get the shortfall, I to 12. So maybe another foot or a little bit more, let's go see. So it's been a, it's been a real, a real hoot here, but hey, I want to talk about a a topic that I But what I'm starting to see from an evolution of what we're doing or what others are doing.

02:11like, at the top of, like, at the top of your contact center to doing these type of tools but When you really focus on it and you just put like all of your attention and have this group This group of people like we have here just focusing on? How do you- fully automate contact center quality assurance with with AI? It. It's it's pretty crazy. Let me just I want I want to get into how we do this, right? I'm kind of the behind the scenes. I'm not going to show you the, the tool or anything even play with creating maybe a tool yourself for something that's maybe a little a little bit basic and you want to build out a KMS and have all these integrations and try to automate your self-service, it's going to be extremely expensive to do today. But auto QA is something that is something that is really done, can be really, really, like for our process, it's about a two-week process to fully running and saving and saving money.

04:03So, number I is the cost savings that you find. We had 13 agents that were scoring calls at KMS, right? managers that their whole job was to score calls for all of our different clients.urches. We now have none, right? Like not one person is is scoring calls. So I think that, that's been that's been pretty insane. And again, we're gonna do this as, as an, as always, I have a comment, I'll get you guys promise. Let me get, kind of, through this a little When it's done properly, we're going to be more accurate or as accurate or more accurate scoring call, we're still going to be off, a little bit, or normally off Right, we can build out. Any type of question, right? It's fully when done right and kind of the way is we can tweak this thing right we don't never say right or wrong right there's auto-scoring something properly. Is it scoring it right or is it wrong? No it's scoring it to the culture of the organization. There's going to be questions questions that you have black and white. there's going to be questions that you have that you say yeah we give their agent you know the leeway on, on that. but we have that kind of It's crazy. Now, Matt Hager has a question here earlier. What do you see as the biggest challenges for conducting AI for QA, processes? Is it the technology itself, change management or something else?

06:26I think there needs to be a buy, right, right? Because again, if Matt, if you and I score calls, we're always going to be a little bit off. But if you have a team a QA team that we put in the There really isn There really isn't. We're scaling where you're trying to do two to three calls per agent per week. that number can go be. astronomical now for less cost than human scoring, maybe 40 calls. calls or 30 calls or 50 calls................... the We're doing hundreds and.un..un. and hundreds and hundreds of calls a day for less than an. for the month for one person. the So again, I because I know a lot of the CCaaS players,

[Gap 08:35–09:05: no reliable transcription]

09:05that everything is scored appropriately.

[Gap 09:07–09:37: no reliable transcription]

09:37What would you tell that QA manager to listen for to score that call a yes or no? That's the easiest way to kind of put it. We got information from your rubric from your score a scorecard. I think of this as a analytics people. Think of this as trending keywords keywords. I-

[Gap 10:23–10:53: no reliable transcription]

10:53that an agent bring up, or a customer bring up a pain point. And then did the agent respond in kind was something that, raise the tone, raise the sentiment, made the customer laugh, made the customer calm down, like depending, right? So, we're looking for intent of the agent now, like true intent intent, and we can look for outcome, the backbone of what's happening on the call or with that little snippet of interact. So we no longer have to just worry, about trending keywords and kind of, you were kind of parsing through this phrase that an, let's add that in there, now that's a piece of it, right? But we, we can go, we can go much deeper. Thank you. No, I can't, because my tool in itself, or an auto QA tool in a tool in itself, will not do that. What auto QA will do now, though, is allow,

[Gap 11:56–12:20: no reliable transcription]

12:20and I have a full book out on this, called, False Hust Kettle. All these things that QA managers and that contact centers do to kind of just check the box, to think that, hey, we're giving a good experience-experience, when really, again, not at all. And we can a much deeper into the customer experience now because now we a score at scale. It's very difficult to score with a human being getting really deep into all these calls because there's so many calls a score. So we just get kind of in that checkbox I get so excited about certain topics and I go off so I apologize so again we have that's in the back giving us kind of all We take your SMS Let's just say we have an audio call now though. So when we analyze a call, we send the call out, we get the transcript for you, so- so you never have to have a transcript, right? We take all All of that stuff that I just talked about the system, prompt each question, each prompt for each question, take that transcript, bight? And then we send it out to the large language models.

13:48There's a lot of argument on, right I will just tell you, I think, honestly,

[Gap 14:18–14:48: no reliable transcription]

14:48of contact center calls. Now a most people at the chat TBT, I find it's not even close to the accuracy standpoint. And let me take a step back. It can be accurate, the main things that we, we do in our system prompt We, we, we really on, that, you know, the models would score things if you had, if you score a the call, 10 times. Then 8 times it would score perfectly, and then two times or one time it would like to check for accuracy, right, one of them is reiteration So when we go and like did the agent use the proper greeting so in our system prompt. We large language model model, answer that question. And did the agent use the proper, And if it's yes, we say, okay, hold that. Ask the question again It was the most expensive model- that's what we were using.

16:30Over chat G And then when 4 came out and it 3. send it up. We score it or have the language model score score the timing of this thing. It takes about two to three seconds a transcript and we're down to about 23 seconds for seconds, it takes score comes back, we score the call. We wanted to keep, and I think this is a really important thing, I see a lot of this stuff being over-built. I see a lot of analytics companies that that have unbelievably cool tools, almost like just business intelligence-type tools tools trying to kind of For maybe some enterprises that have giant teams of analysts, that's great Click, click on. Click on a question. Find out why it scored it like it did. Get the feedback for how the AI scored it. That's another way that you can really find out and start to really to calibrate some things as well. We have found again that the companies and the customers whether, you One of the things again a difference here on the market that you frivolity is everybody is still to see licensing right so they're they're charging you just the regular kind of the always hated it. I get that private equity needs to kinda see and have a Right. So they're done for the month.

20:23Right. Even though they may be worked one or two days out of that month. Like I hate that. Right. So you end up paying for things that you're Now, if you have some compliance issues, if you have some things that from a disclosure issue, if there's specific things maybe on a small part of the call, totally get that. [QA coverage remark omitted] There's the law of large numbers. Once you get to a certain number of calls for an agent, the plus or minus margin of error is going to be less depending on what you want, less than 1%, less than 2%. [QA coverage remark omitted]

21:30It just doesn't make sense to me. It's just easier for the technology companies to do that. And again, that gets pretty expensive for them. [pricing remark omitted] And I guarantee you that auto, I guess auto is, uh, it's going to be more accurate and way cheaper, way cheaper. So again, we went to just a total usage model. How many calls do you want scored per month? Let's do the math. Let's find out how many you need per agent and let's just score that amount, right? suck it crazy. [QA coverage remark omitted] Like those are the two things that still don't make sense to me when when you can do this much less expensive, get the same amount and get more bang for your buck doing it with a usage model using using AI. The other cool thing, a why I think that this is a no brainer is auto QA platforms are are really good at scoring agent behaviors, right?

22:37That's all the stuff we talk about, just those scorecards. But the other cool thing that has been amazing is looking at actual analytics, looking at customer behaviors. What are customers customers actually saying asking questions, right? at the end of the day, what is an auto QA platform? It's a platform that asks questions of data. And the data that we are asking off of is your transcripts, right? Your call recordings, your chat, your SMS, your help desk tickets. you want to ask? It doesn't just have to be, did the agent do this? Did the customer say this? Why is this customer canceling? Here's some reason why a the customer cancels. Are any of these true?

[Gap 23:12–23:33: no reliable transcription]

23:33interaction analytics from CXone. to the 2132 calls that we to last b month 1784 the customers to cancer because they said the product was to get that type of granular data and to have that data on a dashboard every single moment that you can look at how people are answering all of these questions and how your customers are responding and talking to your data, your dashboard that you always have, and do this all for less than you're doing now, and at scale more than you're doing now, you know, we talked about bigger,

[Gap 24:52–25:09: no reliable transcription]

25:09But a lot of people start implementing things next year, this is the low hanging fruit of Thank you and thank you. [QA coverage remark omitted] So now Otto's going to put your calls into a,no matter what. right? But if we This is the specific call that just makes reporting a little bit better.

[Gap 26:43–27:09: no reliable transcription]

27:09right, maybe on just just quality, but it points won't take points away I don't know that that is like that.. Let us build it out, tell us exactly how every, every, and then when you like the customer form is more correct, how they're more correct, how auto is, and we tweak and we calibrate. to change, and how you guys score, not just if something is, is right, or wrong. There's many times as well When we get to a point where we're 95 plus

[Gap 29:45–30:15: no reliable transcription]

30:15Again, I am extremely proud of the I am extremely proud of the tool that we we built month contract, but you don't have to if you want to start this month to month to month

[Gap 30:49–31:19: no reliable transcription]

31:19ROI right from the beginning and be able to answer questions that you never thought you you in. So again, thank you guys very, very much. Hopefully this kind of plants a seed, not even for auto Q and A,