Falling Behind on the Latest CX Tech Stack?

Transcript

[00:00:00] Steve MacDonald: Welcome to the Contact Center Perspectives podcast. I’m Steve MacDonald, your host. Today, we’ve got a really interesting topic and guest, Craig Stoss. Craig, you’ve been in the CX industry for over 25 years. I think you’ve actually traveled to 30+ different countries. You might win the award for the most traveled CX consultant in the industry.

You worked with many different brands. You are uniquely qualified after working in the BPO space and e-commerce space, working with name brands worldwide. We’re going to talk today about falling behind on the latest CX tech stack because we know as CX leaders in just every industry, technology, data, analytics, decision-making, and AI, all these things to come together in a confluence that we need to make sure that we’re in control of and that could be scary sometimes, the idea of a tech stack and managing that and the influence of AI and the analytic tools that are needed on top of all that.

So you’re going to help us demystify all that we can. Before we start, if you wouldn’t mind expanding a little bit on the light introduction that I just gave right now? 

[00:01:14] Craig Stoss: Thank you very much, Steve. I’m really happy to be here. I grew up in a small community in the late 1980s and early 1990s, when the Internet was just becoming a thing and computers weren’t even in classrooms. Certainly, cell phones didn’t exist.

I was really lucky at a young age that my one teacher happened to have a Commodore 64 in the back of his classroom. He allowed people to play with it. So, I got to detect it really early in my life then that translated into some really early customer-facing careers.

Customer experience was really a thing. I was customer-facing. I taught senior citizens how to access government forms on the Internet. I taught summer computer camps for children. Again, it was a very different type of customer experience there. After I got a computer science degree and a software engineering degree, I moved into support.

I loved people, and I wanted to see how I could combine technology and people. It was a fascinating thing, again, in the late 90s and early 2000s, that was pretty new. It was a new concept of bringing in this idea of technical support. Nowadays, we take it for granted, and industries have trillion dollars of TAM out there for this type of stuff. Then, through a series of events, I got into some consultancy and started working with big international brands companies like Honeywell, NCR, and HSBC Bank. I’d got to do a lot of travel with them to meet their different offices and talk about their use cases and different capacities.

[00:02:33] Craig Stoss: This was mainly from a software and implementation standpoint, that’s where I gained my love of tools and processes. When I got the chance to lead my first support team, I really came in with this idea that humans are best when the tools they use are set to augment them and help them become more efficient and that’s always been my philosophy and honestly, I couldn’t have imagined where AI and certain newer technologies machine learning from 5 years ago and now with our LLMs and Gen AI are coming up, which we’re going to talk a lot about in this topic.

That love of tools has done nothing but expand because I think we’re now in an age of customer experience where these tools can start to be a little bit more impactful. The percentage of impact on productivity, customer experience, and customer satisfaction is much greater than it was maybe with earlier types of technology we use. So, I’m just really excited to talk about this topic and see where this future is going to take us. 

[00:03:34] Steve MacDonald: Craig, if you wouldn’t mind, please put us on a level playing field. We work in a very technologically complex world these days. So, how do we understand what we need to know in the CX space, and how do we start to apply it? What are some of the fundamentals that we just have to grasp to be successful today?

[00:03:55] Craig Stoss: That is the fundamental question, and I think if I had the real answer, we could start a company around it. My view on customer experience is where every customer needs to start and where every company needs to start is with the value they provide.

 I think that value is the number one thing we should focus on in customer experience. Wherever the customer wants value from you, you should be able to figure out what that value is and how to deliver that as frictionless as possible. If we take a very complex thing and simplify it, let’s start with the basics.

When someone first comes to your website, why are they there? What are the 12 to 500 reasons that someone comes to your website in e-commerce? It might define your store hours in some software companies that might define the pricing of your product, or maybe a demo of your product.

Maybe it’s defined how to contact you if you’re a large enterprise company or the latest interest rates if you’re in FinTech. There’s something that your customers want from you. How do you get that to them fast and efficiently? Let’s start right there and then say, ‘Okay, now how can technology help me provide that?’

That could be through a conversational AI bot, really useful widgets, or better menu navigation on your website. It’s a technology in and of itself. There are tons of ways to provide that. Maybe it’s video content, so maybe you have to have a video tool or a video library for your product.

With the physical model, maybe that’s AR technology. I know when I worked at Shopify, they released this really cool product where you could scan a product and then, using augmented reality, put it into your house. I want to buy a night table or bedside table. I can scan it. It gives me the dimensions. Then, using this app, I could augment reality and put this bedside table beside my bed and see how tall it was compared to the bed. See how far away it would have to be from the register. If you have an AC vet register or the window or something there. That was really useful. What a neat way. I recently bought glasses online, and they took a scan of my face and put the glasses on my face using augmented reality. What a really cool way to use technology to show what glasses would look like on someone’s face remotely.  That’s the type of thing that I think companies need to start thinking about: What value am I here to provide? What’s the easiest way to do that, and what technology could help me provide?

[00:06:24] Steve MacDonald: What’s interesting is that there are two camps in value. The value we’re talking about here is value for the customer and how we can increase that experience and get them what they want more efficiently in really cool technological ways that they haven’t even thought of yet. 

There’s also value on the analytic side. There’s value internally in the company. So, there’s value to product development and engineering. There’s value in being on the front line to the sales organization, the marketing. So there are technological tools for assessing and providing value as well. That makes it a little bit overwhelming. But when we think about this in terms of demystifying this. Are there some fundamentals or things that we need to be thinking about? What is the problem with today’s tech stack? 

Maybe we should start there in terms of what challenges we need to consider as we build out our tech stack and provide the front- and back-end value that you’re talking about.

[00:07:35] Craig Stoss: Yeah, great question. The good thing about the AI revolution that we’re seeing right now is that these tools make it easier to do a lot of what I’ve just talked about. For example, insights and analytics AI are tools. One of the companies I consult with called the Loops is an AI insights and analytics tool.

What they do is they connect to your entire tech stack. They connect to your CRM and your help desk to your product usage tool. They connect to your log files. They connected your sales calls. They can connect anywhere in your tech stack across the entire life cycle of your customers to understand patterns, trends, and common feedback. If you connect to your social media accounts, things like that, and then be able to assign it information value as far as money. Is this feedback costing you money? Is this type of support to get costing you money? Then, you can start to be very surgical about what I want to do.

This widget costs us $200,000 a year in support costs. If all we did was change that widget to be more user-friendly, we would save that much money. How impacting is that to a business to say, ‘This is costing us here, even though it might be saving us over here, which is greater?’

If it costs less to fix than to support, that’s a win for the business. So, the biggest consideration change in the market today is that we used to go to market for tools that solved some specific problem. I’ll use telephony as an example.

We would go buy a telephone system, a VoIP system, like your ring centrals, your air calls, and your dial pads. It’s very easy to do so. You could buy any of them and still get most of the features you want, around 90 plus percent of the features you want. Obviously, there are differentiations. 

That’s how the market works. That’s how competition works. Pricing is different. The models they do and the UI is different. Those are the valuable advantages of doing an RFP and investigating. The new model is something very different. It’s not about I want to go to market for a conversational AI tool because two tools that could be labeled as conversational AI could solve their problems in very different ways.

The AI model in the background has to be tuned to certain things. You might have a model specifically for e-commerce or FinTech. Each tool is not identical, even though it might fit under the same category of tool. There’s a big shift coming in the way we buy software and I would argue that shift is up to the companies to be able to market their software better. As opposed to being very generic, we will boost your customer experience or we will deflect 70 percent of your tickets. 

You have to be very precise in what tickets are you deflecting or how are you going to boost customer experience because that is what’s actually going to end up providing the benefit, the ROI to the companies that you sell to. Those are the biggest things I think we need to watch out for

[00:10:38] Steve MacDonald: You’re right there, you’ve defined why it’s more complex. Before, it was much more of a hardware decision about what we were buying.

Today, it’s much more about capabilities as product feature hardware has become more of a commodity. I don’t mean anything against any of those companies, but it’s hard to have product feature superiority. What’s defining the tech stack of today is the software and the AI that runs behind it.

[00:11:06] Craig Stoss: I think it also goes to the type of experience you want to provide. When you bought these point tools, telephone systems, help desks, or any tool that runs a particular part of your business, the idea was that you were just putting a tool in place to help you get some metrics out of it.

They weren’t necessarily customer-facing. Customers don’t care what phone system you use as long as the phone number is accessible to them. They email your support desk. As long as the email gets, it doesn’t matter whether you’re using Zendesk, Freshdesk, Salesforce, or whatever. 

With AI technologies, these things are becoming more and more customer-facing. They change the way you search for a knowledge base. They change the way you interact with a brand. So, you need to be much more considerate about what is the experience I want to convey. 

In some parts of Europe, you have to tell people that you’re talking to a bot. Is that something that you’re willing to do? Does that hurt your brand to your demograph? Who is using your brand? If you’re working with generations, millennials or younger, you might want to have a very app-based brand that contains everything you want. It’s all personalized, all that information. 

Whereas if you’re working with millennials and older, there might be a different consideration there depending on who you’re selling to and why you’re selling.

The healthcare market is interesting. I did a lot of work at my last BPO in the healthcare industry, and the empathy side of it—how much can you automate empathy? It’s not much, so you need to consider that. It looks flashy, and it sounds amazing that “Hey, we can deflect stuff. We can save money.” All those things that businesses love, but you have to really think about the impact on who you’re selling to.

[00:12:37] Steve MacDonald: The picture you’re painting for me is that it used to be a very simple puzzle with pieces that you would just put together. What I think you’re saying here today is that you have to define the experience and the value that you want to provide both externally and internally and then design the system to your particular needs.

That’s a very different approach to creating a tech stack, and it sounds a bit more overwhelming because it puts a lot more onus on us versus just getting to what V.O.P. system I need. I’m gonna look at three or four of the top and make a decision to move on because the technology and the AI drive experience and drive value. If you’re consulting with a client, what’s the first step that you would have them do and have them think through in order to be prepared to go out into the market and then search vendors, search partners that could help them put this tech stack together?

[00:13:40] Craig Stoss: I always start by helping them decipher their priorities. For example, I’m working with a smaller company called Truco. Now, that’s a QA tool, not an automated QA tool. I sat down and asked them what their priorities were, and they said, “One of the hardest things we have is to get across our ROI.”

How do we tell people what our return investment is by automating their QA? You start breaking that down into chunks of problems. What’s your value? Your value is that you can QA 100% of tickets versus manually. You can probably get 10% to 20%. Most companies may focus on when it comes to support tickets so there’s a huge value right there.

How do you advertise that in a way that sounds appealing? Even if the cost of your tool might be slightly more than the manual cost, you’re getting so much more out of it. What’s the value of that QA? How can you improve your customer experience? You start to ask a lot of questions about where things break down.

The second thing I go to is, what are your existing processes? How do you operate today? I do a lot of shadowing with my clients. I watch what their agents do. I listen to sales calls. I hear what objections the people they’re trying to sell to have. I hear what their customers are frustrated with.

I watch what their agents struggle with. It’s amazing. A friend of mine used to run the success operations team at a large company that developed software for developers. He always talked about counting clicks and I love that mentality of watching someone and counting the clicks, and you will be surprised at how many clicks something as simple as processing a refund is in the e-commerce space.

You have to click and find the order number, and sometimes that’s multiple clicks because you have to copy the order number to another tool. Then you click search, then find the right order in the list, and click on that. Then maybe there’s a click in the tool because the UI is bad, where you have to go to the payment method they use.

Then you have to go to Stripe, open that window, and click, and automation can start to reduce that again. Another one of the clients I work with is code if that’s what they do, they are able to connect to those systems and say right up front to the agent. Here’s what we think we’re trying to do,  we recognize using AI.

This is what you’re trying to do. You’re going to refund this person. Here’s what we think the refund process should look like. We’re going to email this email that we generated to the customer. We’re going to go and refund this order. We will send out a UPS shipment label so they can return it to you, et cetera. Then, all the agents do is look at it and say, “Yes. That’s what I want to do.” Click, and you’ve reduced six to seven minutes of clicking to one or two clicks in the middle. Those are very powerful changes to your agents. If that’s a priority to make your agents more effective, that may not be super visible to your clients, or your customers.

Your agent’s lives are a lot easier. There are fewer clicks and there’s less manual intervention, which means there’s less chance for copy and paste errors. Those are important things to companies. I sit, watch, shadow, and see where these long processes are that are just constantly repeated day over day, ticket over ticket.

That’s true in sales. I provide a lot of support examples, which is my background, but this is true in sales calls. How many times have you answered the same objections in a sales call? Why not have that information readily available in some meaningful way ahead of the sales call? Educate your customers.

Maybe you should have your 60-second, 90-second demo on your website that answers some of these key questions for your clients before you get on a call. The communication is lopsided, and you’re trying to articulate that as an expert to someone who has no knowledge about anything you’re talking about.

Maybe they don’t even know the terms you’re talking about. Those are things that I think can be approved. That’s where I start when it comes to consulting on the tech stack: where are your priorities, and where are your inefficiencies in the various processes you put in? 

[00:17:30] Steve MacDonald: We talked a lot about AI and how it is fundamentally changing the way that we buy in the tech stack. What are some of the challenges and complications that AI also introduces into the tech stack, and what are you trying to accomplish?

[00:17:47] Craig Stoss: It’s really interesting to understand that what we’re calling AI today is not truly artificial intelligence. 

The idea of artificial intelligence doesn’t exist today. Artificial intelligence really means being creative and generating new ideas. Our current AI does not do that. The most advanced AI models are these large language models that we hear about. Chat GPT, Lama, and Gemini are the Google ones.

These are the large language models we start to hear about. They are really good at predicting what the next thing will be. If I give it a prompt, it can predict what the sentence that answers that prompt looks like. That’s why they’re pretty good at code. I just did this earlier today for a client, I described what a web page should look like that solved a problem for them, and it generated the JavaScript and HTML code because it’s fairly obvious in language that’s logically structured what the next thing is.

With English, it’s not always on because these are not creative bots. These are not creative things. They have to be trained on everything that we want the bot to know so the problems that are introduced are when the models are too generic. We see this with examples recently Chevrolet, someone social engineered the bot to sell him a car for a dollar. 

Air Canada. I’m based in Canada. Air Canada just had a problem where the bot made up a bereavement fair policy that didn’t exist. The courts held air candidates to adhere to that policy. The reason is that those bots were not constrained enough and weren’t trained properly enough in all those cases. These are called hallucinations, where they make something up or do something unexpected.

Some of them are not catchable because these things, again, have a bit of a mind of their own; it’s not really understood how they get to certain answers. So, when it comes to what they’re not good at, they’re not good at answering questions that are broader and less specific unless you have really constrained the model, be like, this is the domain of your knowledge. A broad question can possibly be answered within that domain, and they are really good at answering questions where there is a very clear answer. For example: Do you ship to Texas? Yes or no?

I have worked with companies where, for example, they’ll have a web page that has the shipping policy, and it will say something like we shipped to all U.S. states except Alaska, something like that. The bots are smart enough to read that sentence and understand that means Texas is included. Now, a better way to train it might be to be a very exhaustive list and list every single state, that might be a better way to train the bot, but bots are generally smart enough to get that kind of nuances and your hours, your company values.

I think that in the future, every web page will have a bot that changes how we navigate the web. As opposed to now having menus and drop-downs and trying to find the Contact Us page or trying to find the company history page. There’s just gonna be a bot. You’re gonna have a homepage and a bot, and you can say, ‘Tell me about your company history.’

It will tell you right there that you don’t have to find it on the site—that’s coming soon. It’s really good at things that have definitive answers. It’s not necessarily good at empathy, and it certainly is not good at sarcasm. Those are ways of communicating that humans have evolved.

In fact, I’d argue in my travels that empathy in Asia is very different from empathy in North America. To some degree, it’s even different in Europe, but there are also different nuances between any sort of culture within a subsegment of culture. They’re not good at it because it’s hard to train someone to use sarcasm because it’s a tone thing we do with our voice so that’s where you start to break down. I would argue that you need to take those things into consideration when you’re implementing any of this stuff because you have to know what you want the bot to say and do. You have to know what you shouldn’t say and do. 

One of the first things I do with every bot I encounter when I’m testing for clients is ask how big the moon is. If it’s answered how big is the moon, then it’s not constrained enough. It should tell me, ‘I’m not trained to know this. I don’t understand your question. That’s out of the scope of my knowledge.’ It should tell me that. But if it doesn’t, I know your bot isn’t constrained enough. You haven’t trained it down to the subset you need.

[00:22:14] Steve MacDonald: Interesting. You’ve done a really good job of talking about how you are just approaching. Our tech stack has fundamentally changed, including the impact of the front-end customer and the experience on the back-end analytics, the challenges of AI, and the technology. 

But now, here’s my question. We have to go through this consideration and sales process. We got to put something together. What are the ways to navigate through this for somebody who doesn’t have the experience level that you do?

[00:22:47] Craig Stoss: It’s really interesting. I actually think this is going to change the way we buy software. I really, truly believe that. I was talking to a gentleman earlier this week who runs a company that does nothing but consult on I.T. infrastructure and software, and I told him a little bit about what I do. One of the things that we both picked up on immediately was that there’s a trend here in the A.I. industry. You might need someone like this person for AI software. 

 Big companies go to him and say, ‘I don’t know what the best IT infrastructure looks like. I don’t know what software stack we should have to run this very complex network, configuration, security, and all that stuff. That’s very complex. You go out, and we’re going to give you a set of use cases. You tell us what the tech stack is.’ He gets paid by referral fees and the software company obviously gets paid with a new customer.

I think you’ll start seeing this in the AI world because the world is very complex. I don’t advocate for that necessarily immediately, but step one is not to rely on Google like you used to. Do not go to Google and say the best conversational AI tool, workflow automation tool, and agent assist tool. That’s not going to work. 

What you really need to do is start to think about your use cases very specifically. For example, I want to deflect customer-facing requests for refunds in an e-commerce company. Search that, and that will give you a better sense of what tools might help you there. Number one, be very specific with your use cases. Number two, I think when you’re talking to potential clients and you hear things like, it works out of the box, it’s low code or our models can cover all these use cases; they exaggerate, don’t believe them. 

AI is not magic. It is absolutely not magic. There is absolutely a maintenance cost to this, and that maintenance cost might be hidden. It might be something that your senior people and your team can already do. You might have that skill set, or you might need to hire a full-time AI administrator. It could be anything in that range, but there is a maintenance cost. When having these conversations, you also need to see demos on real data. 

If you’re a company that sells AI listening to this, and you do not allow your customers to test a small subset of their data inside your system, I think you will fail because the data sets are so diverse. You’re mocked-up setting that every company likes to demo when it seems like a hummingbird.  That’s not going to work. You have mocked that up, the customers will learn about it, and when they go and implement it, and it’s more difficult, and it doesn’t work as smoothly, that’s a bad customer experience, and you’re going to see churn, and I have seen that happen. 

I’ve seen vendors come in, and wow, their clients go to the implementation stage, but a month later, it’s no. Now, you’ve wasted the time, money, and effort, and you feel like that person will ever talk to you again as a client.

Returning to your question about how you buy software, use case, and lots of questions about how the model is trained and maintained. I think the third one that I’ll throw in here is that you need to clean your house first because if your data is messy, then the AI will absolutely not be able to make sense of it. So, when you buy it, if you do what I say and test it, do a proof of concept or test your data with a system and ensure your data is clean. For example, the knowledge base, do you ship to Texas? What if I typed in, do you ship to Germany? The only thing in your article says we ship to the United States, except Alaska; there is an ambiguity there as a human since a human might not think that’s ambiguous because it’s explicitly left out. We can assume you don’t ship to Germany, But the AI might interpret that differently depending on how the model is. Maybe one extra sentence: we only ship to North American clients or United American clients.

That’s getting your data in order, and you will have to do a data exercise as part of any implementation. Those are three things I would say when you’re going through the buying cycle. 

[00:26:55] Steve MacDonald: Can you tell us any of the customer stories of people you worked with? I’ve gone through this evaluation process, and the steps that you just went through there are ones that we can learn from.

[00:27:05] Craig Stoss: My favorite one is the refund use case I already used earlier. One of my clients was a food delivery company in California, and they used a process similar to what I just talked about workflow automation tooling. We’re able to reduce their refund processes from 8 minutes to 1 minute by using automation, and this has done it all. It did exactly what I said earlier in the chat. This is what I want to do, and the agent could sit there and say yes or no. The good thing about stuff like that is you can start to make a customer-facing need with some checks and balances, and that’s what they’re working on. The next is how can the refund just be a self-service practice?

I think it’s really interesting to go through and navigate an entire refund process on your own. Uber Eats does that already, which is interesting. This is a text-chat conversation, and I also want to clarify that you don’t always need AI for interesting things. For example, when I was at a company, one of my biggest accomplishments in two different companies had to do with being proactive on things. So, one was at a company where my team helped develop a support tool where a customer would call in and send us a set of details of their instance, and we fed this into a program with a rule set backing it. This was not AI. This was purely pattern recognition. 

The tool would parse all your data related to your server instance of this software product and generate a report categorized by priority—critical high, and items that need attention. It would highlight the top 125 tickets known to exist and discuss proactive support. For instance, when a customer calls with a minor issue, you can not only help solve their immediate problem but also present a set of related issues that may arise or are already occurring but not yet recognized.

Each issue would link to an article on how to resolve it, providing a comprehensive solution. This tool, initially a small feature, has evolved into a full-fledged product within their support organization, a significant achievement that enhances efficiency and delivers an outstanding customer experience.

Similarly, at another company, we utilized simple database queries. We identified scenarios where customers faced challenges and collaborated with the development operations team to monitor the database. When customers performed specific actions in the tool, we sent them emails notifying potential uncertainties, offering them the chance to undo actions before they caused problems. With advancements in AI, such processes are now more effective.

When discussing tech stack strategies, it’s evident there are existing tools that perform these functions and others that can be integrated into your stack to empower both customers and agents. Personally, I’m passionate about anything that enhances value delivery with minimal friction. These customer success stories illustrate the effectiveness of providing rapid value, regardless of whether it involves AI—it’s about accelerating value delivery.

[00:30:08] Steve MacDonald: Craig, we’ve covered a lot here. If there’s one key takeaway you’d like CX leaders listening today to remember from this podcast, what would it be? 

[00:30:24] Craig Stoss: Aside from emphasizing the value aspect multiple times, I believe the most important message is that this trend is here to stay. As a CX leader, it’s crucial to start considering how it integrates with your current systems or how you can adapt your systems to accommodate it. It’s not a matter of whether you should adopt it now or later; it’s about taking the time to assess what aligns with your goals and what internal adjustments might be necessary.

This market is saturated with tools, but not all will survive. I recently spoke with a leader who predicts that some AI tools won’t be around a year from now. The key is to identify the right partners and solutions that will enhance your customer experiences across the entire journey. Begin exploring these options now, because waiting could leave you lagging behind. In our rapidly evolving landscape, where new advancements like models such as Lama, Gemini, and GPT emerge every six months, falling behind could have serious long-term implications for your business.

[00:31:40] Steve MacDonald: Great words of wisdom. I know there will be people who have questions. Would it be appropriate to provide a link to your profile on LinkedIn? 

[00:31:48] Craig Stoss: Yeah, I’m Craig Stoss on LinkedIn. I’m also available at stoss.ca. Please feel free to reach out to me. I love having these conversations. 

[00:31:56] Steve MacDonald: Fantastic. Thank you for coming in and bridging the gap between how we used to think of a tech stack and how we approach buying it now, especially in such a specialized area.

One of my big takeaways was when you talked about not just google X but google specific to your industry, specific to your use case. Find companies that have solved that in particular, evaluate them, and ask questions. You’ll learn a ton. So, thank you for everything you shared with us today. 

[00:32:30] Craig Stoss: Absolutely. Thank you so much for having me, Steve.