Lessons from the front lines of AI-powered service transformation

When we think about AI resolutions or true resolutions, what we’re trying to say is, at the end of the day, did the customer’s issue truly get resolved?

Was the customer walking away satisfied, or were we just closing a ticket? And the industry has come a long way. We have talked about deflection. We’ve talked about containment. We’ve talked about automation. These are very insular-focused metrics. They’re metrics where we’re talking purely about cost saving and efficiency, but what we saw in the market is that over time, bots were getting a really bad reputation because they weren’t leaving the customer satisfied.

And so when you’re looking into AI agent technologies today, you want to understand how the technology is driving resolution and how it is measuring resolution, how the AI agent is understanding that the customer was truly satisfied at the end of the interaction.

The risk with having lots of isolated automation product, projects, or AI projects is that you don’t have a holistic vision for what you’re trying to achieve. How that looks within an organisation is you have lots of different teams running completely separately, maybe using different technologies, and then by the end of the 12 months, yes, you’ve got lots of different results, but you’ve sellotaped 10 different systems together.

That’s a real loss when you think about modern AI systems because the power of these intelligent systems is that the more information they have access to, the smarter they are and the better and more accurate they are and the more that they can do.

And so what you always want to do as a business leader today, start with a long-term vision of what you’re trying to achieve. Break that then down into phases and build a single team or a couple teams that understand that long-term vision, and then when they’re executing on those isolated automation projects, those aren’t isolated anymore.

They’re layering together, and when you’re thinking about the technologies that you’re bringing to the business, you’re thinking holistically about how to solve multiple problems instead of things one by one.

By far, in my opinion, the number one barrier to AI adoption today is cultural. It’s not technological.
There are a lot of powerful tools in the market. There are models being released every single day. There are amazing people, some of the smartest people that I know, all working on this problem. The difference today between organisations that are using AI and organisations that aren’t is completely cultural.

You have to think to yourself, “How am I fostering a culture of innovation and utilisation of AI within my business?” Because the truth is, we’re three years post the arrival of ChatGPT now, and right now in the market when I meet organisations, I’m meeting some who have high AI literacy in their teams and some that don’t, and there is a huge difference between the way high AI literacy businesses are solving problems and making decisions versus those that aren’t.

What we’re going to see in the next couple of years is really outsized impact for the high AI literacy teams. And so changing culture is one of the hardest things that a business leader can take on. It is arguably the goal of a business leader to manage the culture, a CEO to manage the culture of your business.

One thing that you can do today to make a difference is, at the bare minimum, look at how you’re hiring and ask: who am I hiring for this business? How has my hiring profile changed? How am I evaluating within a hiring process AI literacy and whether they’re utilising AI tools to solve their problems?

If I have a challenge within my hiring process, which most roles do (e.g. some sort of presentation, some sort of challenge, some sort of build), then how am I making sure that the challenges that I’m setting will require a candidate to use AI to solve that problem or enable someone using AI to really outperform their peers?

Because then, at least, the people I’m bringing in are right for the future business that I want to be leading. Culture is obviously critical, and likewise, so is having the right people in the business with the right approach to these kind of transformations. And obviously, you will have seen a lot of organisations try to modernise very fragmented service environments.

This is a real risk that I see in the market today. Everybody wants to be bringing in AI, but are you truly redesigning your organisation to be AI forward and adopt the best in breed of these technologies?

The number one red flag that I might see in the industry is a business saying that they want to bring in AI, but ultimately only wanting to build rule-based deterministic systems. What do I mean by that? Traditional AI all the way up to this new generative era, which started a few years ago, was all about systems that were fully deterministic.

Yes, we would detect using AI that an event was happening, but then we would follow a preset list of rules. This was highly predictable for a business. It was the safe space. Nowadays, what you want to be using that’s best in breed is fully agentic systems; meaning AI that’s able to make decisions independently.

It follows instructions or a process that you’ve described in the same way that you would describe them to a human, but ultimately, these systems are making decisions. The reason you’d want to do this is that it fully broadens the number of situations that AI can exist in. It can achieve goals in a whole plethora of real-world situations in a way that a completely deterministic system is not able to react to any outside input input the second it’s already on a set path.

So it’s night and day versus the sort of outcomes you can achieve, but if you’re nervous about the fact that you’re allowing technology to make decisions and you stick to deterministic, I think that’s a red flag. And nervous is a really interesting word, right? Because if you’re empowering these systems, there’s a level of risk and obviously a lot of these service teams really need to demonstrate the ROI they’re getting from these AI systems very quickly, certainly to leadership.

The first thing that I would say to address the perceived risk that we talked about before, is that you shouldn’t avoid the technology entirely and think, “I’m safe sticking to deterministic.”

You should say: “I want these best in breed agentic systems, but I want to understand how we’re going to think about observability? How are we going to be thinking about transparency into AI decision-making? And how is this technology also allowing us to audit why AI took the decisions that it did and alerting me?”

“How are we alerting whenever there is a situation where we might want a human to take over?” – that’s the first thing you should consider when you are making a selection on technology.

So now you’ve selected your technology and you’re thinking, “How can I prove results quickly?”

There’s two sides to this. The first part is that it’s actually really easy to prove value quickly when it comes to automation. So if you want to prove a strong ROI, automation is the way to go. It has the clearest ROI because you get a clear resolution rate. How many conversations am I resolving fully with AI without human intervention, prevention?

That is pure ROI to your business You can launch an AI agent to do this by connecting it to your knowledge sources. Honestly, at the click of a button. It’s really simple. You just point it to the information that you want, whether that’s in your help centre or your website or your product catalogue, and then it can answer any how-to questions that exist in your knowledge. And then on average, our customers get a 30% automation rate with that alone. It really is:

1. pick the right technology that you can trust
2. get started really quickly by connecting it to your knowledge
3. and, putting it live.

What we will see, which is a trend that has been happening for over a decade now, is that human agents should be taking on less and less transactional, highly repetitive work. In retail eCommerce, these are questions that are along the lines of, “Where’s my order? My order hasn’t arrived.”

For subscription businesses, the majority of questions are all about “cancelling my subscription” or “pausing my subscription”. Very mechanical tasks, and we’re going to see human agents not have to do that so much and get elevated to the more sophisticated situations that you might want a person to handle themselves.

As an example, anything to do with sales I tend to see still given to a human agent because, ultimately, those are situations where you might want someone to persuade a customer, empathise, and connect with an individual to take a certain decision, and people are still best to do that.

If you want to think of an industry that’s had this change happen already, you can look at the marketing industry, where we’ve evolved hugely from a Mad Men era of ad men and brand marketing-heavy businesses to a huge part of our marketing teams today being based on digital marketers, performance marketers, people that are looking at marketing operations, people that are looking at the data.

This has ballooned into a huge industry. There are an incredible number of marketing attribution technologies that you can be bringing in, tracking. That was all unlocked through the digital era. You’re going to see the exact same thing happen in service, where you’ll have an increasing number of new roles, where you’re going to be bringing in AI-literate people to own the AI solution, to test it, expand it, build it, optimise it. Start to think about situations where you can identify upsell opportunities and drive revenue from it. That’s the fastest-growing job in CX today.

The advice that I’d give to a leader starting today and looking at their CX organisation really depends on how AI mature and how technologically mature that organisation is. But let’s say we’re starting at ground zero. You’ve come into a business that has not begun their AI journey yet at all.

The most important thing to get started is knowing the basics, which is: understanding what is happening in your customer support, what are the questions that are coming in, who is asking those questions, how frequent are certain questions and how are we handling them?

You need to start to connect case topics with priority, urgency, and results, right? It is really basic. It’s like when you start a new job, and the first thing that you do is you sit quietly and you observe, and you understand what the status quo.

Once you understand, you have the data. The first thing is to start with automation. It is by far the highest ROI enabling thing that you can do. It’s very easy to get started. I would turn on an AI agent and connect it to my knowledge sources, which are things like your help centre and your website, so it can answer those how-to questions that make up about 30% of everything that comes in.

That gets you live with an AI agent quickly. You may learn at this point that your knowledge isn’t healthy, meaning that there was lots of gaps in your knowledge, for example ,if sections were outdated. This is perfectly normal. Once you have the AI agent live, most solutions today will start some level of recommending knowledge health, suggesting what articles need to be updated based on responses your agents are given.

It will start to very quickly improve the health of your knowledge and the quality so you can get to that 30%. That shouldn’t take you more than really a few weeks, depending on how big your organisation is – and then you get into the really interesting stuff. As I said before, you should know what your topics are, and you should one by one, use case by use case, build what we call procedures, or end-to-end automations for those use cases.

You might need to then connect your AI agent to other systems so it can solve those. For example, for a retailer and the “where’s my order?” question, you might want to connect it to your Shopify. so, it’s not just internal tools, it starts to be external tools too. Now it’s solving those questions.

At this point, you need to level up your human agent team too. One interesting phenomenon that’s started to happen in the market is that AI agents are becoming more knowledgeable, more on brand, more conversational, and then you’re handing off to a human agent that doesn’t have access to the right information at the right time. You’ve routed it completely randomly, so it’s not the most skilled agent for the task in front of it, which can lead to typos, mistakes, or not responding in brand tone of voice. It can be a little bit chaotic, especially if you’re using outsourced support.

While increasing your automation rate with your AI agent, you should start bringing AI tools into your human agent team so that they can have access to the same information that AI does too.

This requires a bit of change management because it requires training human beings, which can take a bit of time and is why you should get started now – then you have an end-to-end AI operation. It’s all about optimisation, A/B testing different flows and finding opportunities to differentiate and to drive revenue.

From here on out, your automation managers or resolution managers should be owning, optimising and testing this continually from this point.