How Agentic AI Is Replacing Scripted Retail Chatbots

Retail teams are moving beyond scripted chatbots and into agentic AI systems that can understand context, take action using tools, and coordinate across functions.

In this CX Today interview, Nicole Willing is joined by Gilbert Gooijers, Managing Director and Chief Operating Officer at CM.com, to explain what retail leaders often misunderstand about agentic AI, why multi-agent setups beat a single do-it-all bot, and how the right channel choice, from WhatsApp to RCS and voice, shapes customer experience across generations. Gooijers also shares real-world examples, including event-scale customer service and guided purchase journeys, before highlighting the biggest emerging risk: security.

The message to retailers is clear: treat cybersecurity as the starting point, not an afterthought, as AI agents become more capable and more connected.

For more Customer Experience tech news visit https://www.cxtoday.com

Transcript

SPEAKER_02 0:08

Hello and welcome to CX Today. Agenda KI is reshaping customer experience in retail from decision-making systems to cross-functional coordination. And the implications are significant for how brands engage, serve, and retain their customers. I'm Nicole Willing, and today I'm joined by Gilbert Gears, who is Managing Director and Chief Operating Officer at CM.com. And we're going to unpack what this means in practice. Thank you, Gilbert, for being here.

SPEAKER_00 0:32

Yeah, great to be here. Thank you very much.

SPEAKER_02 0:35

So to start, obviously, you know, there's been no shortage of innovation in retail over the past decade. But what makes this current generation of AI meaningfully different for retailers?

SPEAKER_00 0:46

Yeah, I think I think the way how we use AI now in CX for retailers is really a big step up compared to uh maybe six months ago or three months ago, even, or even a month ago. The developments are going that blazingly fast. Um, let's say a year ago we were using chatbots. If this, then that uh pre-scripted uh chatbots. That that's what the world now still understands as when they interact with an uh a retailer, they they are using chatbots. But actually, yeah, the current uh generation of AI agents, you cannot compare it anymore with these chatbots. That's that's really a big difference.

SPEAKER_02 1:24

So then how should um leaders really think about that relative to you know like earlier automation or investments in analytics?

SPEAKER_00 1:33

Well, yeah, well, if you compare it to chatbots, that all these prescripted uh flows, oh, it's in some cases it works very well, but as soon as a consumer starts to deviate a little bit in a conversation, that was difficult uh to solve with chatbots. Now with agent agenticai, um it's all based on a knowledge base, uh it combined with an LLM and um combined with specific style style settings um and based on analytics of all the previous conversations. So um now a virtual human or a multi-agent system um can perfectly understand the whole content and the whole context of the whole conversation, um, and all the intents, and that that really brings a human-like experience right now. And um yeah, we are now at this point in time that a lot of uh contact centers and and also the the marketing side of of retailers are exploring this. Uh, everybody has their own experience now with with ChatGPT, with Cloud, uh, with the with the LLMs. And now the next step is okay, how do we bring this in the in the professional environment uh where there's a lot of uh customer interactions? Yeah, and yeah, I think we have a very nice solution for that, yeah.

SPEAKER_02 2:46

Yeah, sure, because you know obviously there's a lot of buzz around the term agentic AI, and you mentioned you know multi-agent systems and various things coming in, and so you know it can mean different things to different people. So when uh retail leaders hear this term agentic AI, what do they most um you know misunderstand or underestimate about that?

SPEAKER_00 3:05

Yeah, so agentic AI is is there where a LLM gets hands and feet, so it has tools to actually do stuff. Um misconceptions can be that people think that you should build one agent that knows it all. Uh, and that's not always the best case to start. Um, where we've learned that it's smart to really think in multi-agent systems. So think of it as a virtual team that does uh various jobs, various tasks, and um try to prompt these specific agents as good as you can. Um, they will work together, uh, and together they will solve uh tasks at hand. So, for example, if we uh create for uh uh our customers a multi-agent system, we just start with the most asked question. Um, let's see what kind of uh knowledge base and what kind of tool sets, uh database connections there are needed. Let's build that first, and then you know you add extra agents uh next to that. For example, we we do um the haul for the Formula One in the Netherlands, that it's a very big festival, and we process all incoming questions for um for them. And uh there used to be every year there are about 13,000 to 15,000 questions just a few days before um the the show starts. Uh 300,000 people visit this uh this big event. So you can imagine that people have a lot of questions. Can I bring my umbrella to can I bring my dog? Can I bring my own food? All these questions. So we used to do that with uh 15 people uh every year. Now, last year with Agente KI it was just uh with two uh to do quality control, and the rest was done by a multi-agent system. And I give this example because on Friday we noticed uh the first day of the race that there were a lot of questions coming in uh regarding the weather. Okay, interesting. So people asked uh Dutch Grand Prix about the weather, apparently. So what we did is we just added an weather agent into the system, and yeah, automatically all the weather questions were solved by the weather agent. So there's an example of how we can introduce in the Halo system uh where you have all these agents alive, it's very easily a new agent to uh to build out. So the misconception could be it's it's quite difficult to start, or maybe very expensive. Well, the answer is well, actually, no, you can start uh within within 48 hours, you will have your first uh agent in the system. Yeah.

SPEAKER_02 5:35

Yeah, that's really interesting because then yeah, this idea of it, I guess makes businesses more nimble in the sense that if if there is a gap somewhere, they can introduce an agent with the knowledge base to be able to handle that.

SPEAKER_00 5:48

Yeah, yeah, yeah. That it's it's as simple as that. Um it's not only the knowledge base, so uh it's also the channel. Uh, for example, for Revolution Beauty, uh and a customer of ours in the in the UK, we uh looked okay which channel is the best to use for doing all these conversations with consumers. Uh, and there uh WhatsApp is a very good channel. But for example, for Blackhawk, uh agency that uh that processes a lot of uh vouchers uh for games, for example. We are now looking next to WhatsApp also to RCS, a new uh version of SMS brought to the market by Google and the telecom operators. Um the combination of picking the right channel, web chat, voice, RCS, WhatsApp, really brings the right experience uh to the consumer. Uh, and you and you can address multiple generations because talking to an agent or multi-agent system is not only for Gen Z or Gen Y, it is also for for uh other generations if you pick the right channel. Yeah, and so Voice is a very interesting new add-on um where everybody that still likes to call, well, they can call. And it's not only during office hours, it's 24-7. Um, yeah, and if you're a bit lonely, you can just keep on talking. Yeah, I mean it's it can be very entertaining conversations with with AI.

SPEAKER_02 7:11

Yeah, so that that that's true, actually. Um, so you know, bringing this, you mentioned the the different generations of customers. So from the customer stamp standpoint and you know the the CX experience, um, where are you seeing the most immediate value today from AI agents? Is it service, is it the personalization, fulfillment, or something else?

SPEAKER_00 7:31

Yeah, so so um in in service, it's where it started, uh I would say. So um, of course, AI is really very accelerating at the point where there's a lot of repetitive work. Uh, at the service desks, every hour the same questions come in. Of course, AI can uh can help uh a lot there. Um, but also in the in the marketing space or in discovery, uh uh AI is very helpful. So agents can help to help you pick the right products, uh guide you through a buying process. That's what we do with Blackhawk, for example. People uh say, Hey, I'm I'm looking for a gift for a for a friend of mine. Uh okay, how much uh do you want to spend? Uh I want to spend uh 50 euro or uh a pound, for example. Um, and then you can have a conversation about what type of friend it is, what type kind of games or vouchers it should be. So it's it's a whole um well buying journey guided by a an agent. So it feels like like an agent, but uh behind the screens it it it maybe will be four, five, six agents that are helping you. Uh but doing this in in, for example, in WhatsApp uh gives a very smooth experience. Yeah, yeah, 24-7.

SPEAKER_02 8:41

Sure, yeah. And so then how does Agentic AI kind of change these functions? Because you know, talking to people, you know, that there's a lot of talk about how traditionally um organizations across the board, but certainly in retail, have operated in these functional silos. So then when it comes to you know bringing together customer service, inventory, marketing, how does agentic AI change that?

SPEAKER_00 9:05

Yeah, in the end, uh these agents are talking with each other, so it will blend a bit. Um I think within not so long periods we will view companies as a very big database where a lot of data is stored, where a lot of touch points are stored. Uh, we do this in our uh customer data platform uh that has multiple dimensions uh in itself in a database to make sure that everything can be cross-related with each other. If you do that, actually these silos break down. I mean, it a piece of data can be a service question, a piece of data can be this person likes to use this channel, a piece of data can be uh somebody uh bought this product uh or somebody uh sent the product back to uh to the company. So there's uh a lot of data, no matter if it's now if it's marketing or service, uh it's it's there in the database. Um if you have the right agentic setup, it will be able to pick whatever it needs from the database and and bring that to a meaningful interaction, uh, whether it be uh marketing or service. So I I think the silos will disappear, are disappearing. I I see it happening, yeah.

SPEAKER_02 10:18

It's interesting though you mentioned the right agentic setup. So uh that also suggests there's a wrong agentic setup. So, how should companies approach that to make sure they get the right foundation to be able to make use of that data?

SPEAKER_00 10:30

Yeah, we help a lot of companies with this process. So let's say that the software is there, the LMs are there, the the brain power is there, uh, the tools are there. Um that's a big checkbox. So the step that we now need to make is to to uh to guide our customers uh implementing Halo and and making sure that okay, what are the steps to get there? I mean, we talk about a very big transition for uh the workforce in the world. It's the AI transition that we are all going through. Um that means a lot of uh change management has to be done. Um what if you have 10 people on your service desk and actually you can do it with two? Yeah. Well, it it w what are you going to tell those people? I mean, most probably uh some of them will will already find find another job, okay, and then but some of them they should re-yeah, re-revamp their work flow. Um the good thing is there are also new jobs coming uh to the work floor. Uh, for example, from these 10 people that were there for the at the service desk, uh some of them they might want to become uh quarantine controllers. Um so instead of doing the conversation yourself, you can log into the system and have a look at all the chats. And then per answer, you can say thumbs up, thumbs down, uh, or hey, next time, dear agent, you should uh alter your your answer a little bit like this. Um and by doing so, it it gives also a very nice new new um yeah uh meaning to the work that people used to do. Um so quality control is a new job. I I think uh prompting is a new job. Um and a part of these 10 people can spend more time with actually the customer, with people, uh focusing more on edge cases, uh focusing more on proactive support. So uh I think work and jobs become even more fun for the people that like to do them. And uh in the end, it should be a good symbiosis between uh mental human and machine. Yeah.

SPEAKER_02 12:44

Yeah, yeah, interesting. Um, you mentioned um you know a couple of customers. Could you share an example of how you know a retail workflow might look before and after a gentic AI is introduced?

SPEAKER_00 12:57

Uh workflows uh change very rapidly right now. Um yeah, an old-fashioned workflow is that uh somebody asks uh a Wismo question, where is my order? And there's a human uh sitting behind a computer uh looking uh up all the data in four or five different systems in the distributor system, in the in the selling database structure, a lot of lookups that need to be done, and then somebody needs to type an email back to the to the consumer. So it's very cumbersome and it's very time consuming to for just a one simple question: where is my order? Um, of course, this is very repetitive. There's a lot of these questions in the world nowadays with all these orders and uh and uh home shopping. Um yeah, this workflow completely changed. I I think it it it the workflow at at itself disappeared. It's just one question to uh to a multi-agent system. Several lookups that uh are being uh programmed uh in the system uh to be tools for the multi-agent system, and uh they will fetch the information within seconds, and that's it in the channel of choice of the consumer. So it this is not only an example of of change of a workflow, it's the total disappearance of a workflow, maybe. Yeah, it's just an instant answer, right?

SPEAKER_02 14:13

Yeah, absolutely. And then with that, kind of greater autonomy for these agents, it kind of can also introduce new complexities. So, what um kind of risks emerge when um to customer experience when you've got you know AI systems that are coordinating, making decisions, you know, operating across multiple touch points channels, and then how should retailers deal with that?

SPEAKER_00 14:37

Yeah, often people see it as a scary thing that uh that these virtual agents are doing things, but yeah, now in the real world, humans did it and also made mistakes, of course.

SPEAKER_01 14:48

Right.

SPEAKER_00 14:49

Uh the good thing is that everything that all these agents do is 100% controllable and checkable, and uh everything is locked in uh in a very nice overview. Uh you can drill down an analytics, so it's quite easy to find uh hiccups in the in the in the system. That that's not my worry, actually. What I worry about uh and what retailers should worry about is if they start using all kinds of tools without a very good strategy or uh even cybersecurity strategy, then that poses risks. Um, I mean, some agents are very proactive if they are uh uh put alive without control to open up doors. Maybe you'll open a back door or a door on the side of your company, and criminals can come in. Uh and they will come in because these criminals also use very powerful agenc tools uh to the to do their work for them. Um, so this this this cyber security battle will uh will be a very heavy battle that's uh that's coming up. I I uh I predict. And uh yeah, that that that that's something that people retailers should be very, very aware about. Yeah, yeah. And of course, there are companies that can can help them with that to make sure that uh the tools that are being implemented are safe tools. Yeah, for us, it's uh that's part of the uh implementation process, it's a very important part of it. Yeah.

SPEAKER_02 16:17

Yeah, absolutely. Um I talk to people about cybersecurity a lot, especially when it comes to AI agents. Um, so what kind of approach should retailers take with this? You know, you mentioned that there are companies that can help them. What kind of advice would you give to leaders who are kind of considering how to deal with this?

SPEAKER_00 16:35

Yeah, well, the the best advice is to make sure that you have a security department or a security officer. If that's not the case, then yeah, start uh to do that uh to hire this person uh tomorrow. Um, and then uh when you are designing uh agentic AI and multi-agent systems uh uh with your supplier, uh start with cybersecurity, uh cybersecurity in mind. It's it's it's the beginning and the end. Uh if that's not not in under control in the right place, then that poses a risks. Um but that also depends on the tool. There are tools that are very open, uh, and there are tools that are more more closed and have their guardrails uh built in. Um yeah, that the the the last solution that that's what what we chose to build uh and and provide to our customers that that gives uh a bit more safety. Yeah.

SPEAKER_02 17:24

Great. Well, that that sounds like uh good advice and a good place to start with ensuring the security is is in place tomorrow. Um yeah, uh, but thank you for your clarity and perspective.

SPEAKER_00 17:35

Thank you very much. Always uh fun to talk with a human about uh about AI these days. And uh yeah, you know, uh about the subject we we can have such an interview again in a week or in two weeks or in a month. It will be uh again a little bit different. This AI transition is going so fast. Uh so uh yeah, enough uh content that we uh that we can make and uh and knowledge we can we can share definitely.

SPEAKER_02 18:00

Absolutely, yeah. As you say, it's changing. It feels like almost every day there's something new happening. So 100%. I would pick you up on having another conversation.

SPEAKER_00 18:09

Thank you very much.

SPEAKER_02 18:10

Thank you very much. And for our audience, you'll find more stories, videos, and resources on our website. Make sure that you are subscribed to our newsletter and continue the conversation with our LinkedIn community. Thanks for watching, and I'll see you next time.