REZA SADIQ & CRAIG FARLEY - Agentic AI: Redefining Human Roles

With AI evolving beyond Large Language Models, can businesses progress beyond chatbots and towards agentic AI to transform how customer experience is delivered? What opportunities does this create for improved service, and how does this impact human roles?

We’re excited to be discussing these points and more with Reza and Craig, who each bring their own perspectives. Reza as a CTO of a large telco, where he leads initiatives that harness AI, CX UC and Cloud technologies to pioneer data-driven business models. And Craig as a strategist who empowers businesses to harness cutting-edge technology and supports contact centres to deliver exceptional customer experiences.

From automation to autonomy?

We began our conversation with a discussion about where agentic AI is atand the expectations versus reality moments some businesses may facewhen turning to it.

From Verizon’s vantage point, Reza tells us that much of what is currently labelled as agentic AI is an evolution of AI and experimentation with automation and orchestration, rather than new capability.

But, he tells us, “What is new this time is the context that we use. We have more confidence that we want to do this in an artificial intelligence approach, so how do we add the tools to help humans do their job better?”

They’ve already done a lot of work on their data fabric, which is no small job, “Being a big telco, we come across lots of data from our customers. From that, we can use AI to analyse and make historical and real-time decisions. That allows you to dig a bit deeper before you make decisions and action on them. The next phase is using agentic-type systems which work alongside humans, but humans are very much in control.”

This will mean AI will perform some of the repetitive tasks that humans used to do. And following on from that, they’ll move into prediction, as Reza calls it, “The exciting stuff.

“Can we, from what we've known before and what we're seeing in real time, predict what's coming? So, it's not just a tool, it's being built across the architecture, it's closing decisions between data and then execution. It's more of an advanced set of tools that help us in our job, that's what we’re seeing at the moment.”

From a delivery perspective, Craig at IP Integration thinks that ‘true’ agentic is being achieved more in back-office processes than front office, customer-facing ones. He also notes that most organisations are still evolving from rule-based automations and chatbots towards systems that can make very bounded and guardrail decisions.

Craig notices a nervousness about agentic AI in some organisations. And what they expect to immediately achieve with it may also differ from what’s practical, “I think when people talk about agentic AI, they often imagine systems acting completely independently across the customer journey. But in reality, most organisations are still evolving from those rule-based automations and chatbots towards systems that can make very bounded and guardrail-defined decisions.

“The leap isn't from IVR or a single chat bot to full autonomy. It's a progression. And that progression at this stage is as much about organisations and consumers building trust as it is developing skills or dedicating time to the build itself.”

For Craig, the move towards agentic AI typically starts by looking at the foundations and the readiness of the organisation for it – because AI agents need access to systems and resources, just as human advisors do.

He remarks, “It's easy to think, if everyone's talking about AI agents doing amazing things, ‘I just need to turn one on and it will go away and do all this’. But if it doesn't have the tools to do the job – integrations, knowledge, data – then it can't do its job. It's not going to bemuch use.”

For him, it’s often about getting these building blocks in place first before introducing agentic AI, “Then it's building up and adding the LLM capabilities to improve understanding, improve the conversational aspects and join things together.”

We wondered if there’s sometimes a resistance to doing this foundational work if organisations are expecting an easier route to switching on the potential power of agentic AI? He acknowledges that a proportion would like to jump straight to it, “But that's not really our style at IP Integration. We'll always try and bring it back to making sure you're ready operationally, consultatively approaching it and looking for a reason for adopting the AI: what challenges have you got? Where would AI add value? And then coming up with the cool AI stuff.”

Combining human and artificial intelligence

Craig is in a position to see how copilot and human-in-the-loop models are delivering value as he sees how service teams actually work day-to-day.

He tells us, “We're seeing quite a bit of traction in the agent assist and copilot-type technologies for advisors, supervisors and quality analysts – so surfacing knowledge, summarising interactions, guiding next best actions.

“Going back to what Reza was saying, actually these models work because they respect the complexity of the human interactions. The human is there to give empathy, context, govern the risk – that all is still very important.

“It's not AI replacing agents, it's AI reshaping where the human judgment adds most value. The copilot is reducing cognitive load, it's improving consistency and it's freeing up agents to focus on exceptions and customer relationship building.”

Craig considers this important because it starts to build trust for generative AI internally before rolling it out externally, where the perceived risk is higher.

Redefining human roles

From what Craig is saying, agentic AI has the potential to make human agent roles almost more human – with a focus on an increased use of the skills that are key to us and less of the repetitive work.

This is also something that Reza expects to see at Verizon as they think about redefining the role of human agents when AI is able to take on more of the reasoning, recommendation and orchestration tasks.

Reza explains, “What we're seeing, it's not about replacing humans in any way. It's about redefining what humans actually do and where we add value. So, it's more about getting the laborious stuff done by AI, supervised by humans. The real decisions are still tested and quality checked by humans – there’s frequent checking, there's approval, there are escalation paths to get to a human being on that conversation.

“And it's about a little bit of experimentation but building that trust and safety for governance. Over time, as the models mature and we can see that confidence in AI decision making, we can see that they do more and more of the automated routine decisions and then humans step in only when there's something of question, or risk, or exception.”

Reza points out that this is happening across all industries and so, for Verizon, there is a question of economics. In their wireline operations, after an order is submitted they currently spend 74 cents in the dollar just on tracking, troubleshooting and coordination.

He continues, “Being a big telco, we've got lots of silos, lots of human hand offs. Agentic workflows help to streamline, help to automate and reduce that cost. It's something we can’t stop, and we need to work with it and help the economic process by adopting these technologies.”

Risks, governance and deployment

While there’s potential for economic advantage and more rewarding roles, front-office agentic AI in customer-facing environments is a different challenge from back-office deployment.

When we talk about this, Reza points out that the risks depends on the industry sector, with some, such as banking, having regulatory rules on what an agent can answer back to a customer, “Craig mentioned guardrails earlier on. Those safety mechanisms need to be stopping AI from saying the wrong thing in a very strict industry.

“On the other hand, you might have other industries where you can have a little bit more leeway; I'll just say retail, but they also have their governance. So, it is a bit about testing out the system. We've been doing this with automated Q and A's for many years where the exact answer given out by the agent is protected.

“It's to do with industry and then working with that industry and the rules that they have to navigate.”

Reza reflects that the difference between back office and front is that back office is more about getting tasks done through the automation, whereas the front office is where there are there are legal and regulatory rules to navigate, “Both have their own issues. You've got HR process in the back office, in the front side you've got the government framework, so you’ve got to go case-by-case and work with the industry specialists in developing your AI facing technologies.”

Craig picks up on the governance theme, sharing his viewpoint that this is not only critical for brand reputation and trust, but also for building trust internally within organisations. He tells us, “That's often just as important if you're about to embark on a big cultural change within the business to adopt AI. In many ways, consumers are actually quite used to using AI, even if past experiences might not always have been great, whereas employees are probably a bit more sceptical because they see the news stories about layoffs, they might be fearful for their own jobs.

“It’s easy for me to come and talk to a senior leader about how AI might make their business more efficient, but without starting with why you're adopting AI and what it's going to mean for them, they can be resistant to change. This isn't new for AI, this is just change management 101 for a technology project.”

In Craig’s experience, before approaching AI implementation, organisations need to be clear about why they’re doing this, the outcomes they want, as well as what challenges they’re trying to solve or benefits they’re aiming for, “It should tie into where your company is heading strategically, with that kind of senior sponsorship. This will help prioritise time and investment.

“It's going to avoid lots of unknown science experiments going on all over the place, so you maintain control, and it will help you understand whether something's working later down the line, and whether you've been successful faster, with less waste.”

Alongside this, he sees a need for an internal programme to increase general agnostic AI literacy – covering what it is, how it works, how it can help – to try and remove some of the fear and preconceptions and open up new development paths to use the tools.

Reza agrees that employee training on AI is important, “We have to take a certain level of training and readjustment to this new world. We should take that as a positive thing. AI is there to help, rather than maybe people are fearful that their jobs will change or go.”

Craig comments that while there are often stories about job losses due to AI, there will also be new opportunities, “It's never a case of just turning AI on and letting it do its job. You've got to have people to tune it, tweak it, improve it, monitor it.”

When it does come to deployment, building in guardrails is vital to determine what AI can and can’t do, “You're ensuring that compliance is met, you're secure, it's acting within its brand guidelines.”

As is carrying out a pilot on a small scale so you can iterate from it, “You'll normally get good findings from that which will help improve the process.”

Craig continues, “When you do roll it out, make sure that it's someone's responsibility to check performance, check the outcomes, because for things like virtual agents, we often look for a high containment rate – if they're simple interactions, we want the agent to be containing customers within that process and not passing them on to our more expensive human resource.

“But if all that bot is doing is looping the customer around, or it's not providing the outcome they want, then that customer's just going to call in anyway or, worse, they're going to call your competitor. So, it's not just the surface level metrics, someone needs to pay close attention to it.”

Creating a smoother customer experience

If organisations aren’t ready for agentic AI in live chat or voice, we wondered where they could still improve customer experience using AI-driven assistance and orchestration?

For the team at IP Integration, AI isn’t always the answer to a better customer experience, but where it is, they often recommend a process of Observe, Assist, Optimise, Act. This helps build trust and improve the experience iteratively.

Craig talks us through this, staring with the Observe part of the process, “You can turn on things like speech and text analytics to observe what's going on in your contact centre. Once you've got voice transcription or you've got web messenger’s interactions, you've got a lot of data that you can start analysing with AI in a very low-risk manner to understand what your call types are about, what your chats are about, where there's friction, where there's process failure further up the chain – which is normally more expensive than the £5 or whatever it is to actually take the call. And then you can use that as a starting point.

“You've also got ‘Assist’, which is any kind of agent-assistive technology, and other things like automated quality as well, so that's a big area of benefit.

“Then you can start Optimising your more external-facing processes, using AI to orchestrate workflows, predictively routing interactions through the best path based on their history or what it is that they're contacting about, and triggering discrete processes that can be automated.

“And then the last one, ‘Act’, is where things start to get a bit more agentic, so putting together all of that toolkit that you've built and everything you've learned and building the more agentic autonomous processes as well.

Reza sees an opportunity for more autonomous AI emerging in proactive service. He tells us, “Near-term, autonomy delivers high value in proactive services, predicting things and resolving them before customers are impacted. That protects the brand, it builds trust, it reduces the inbound core volume.

“If you fix problems before people have noticed, hopefully they're not going to call in and complain. It allows agents to focus on the more high-value interactions, rather than dealing with routine troubleshooting and problem solving.

“The biggest value comes from moving from fix it when it breaks to fix it before they notice.”

“The autonomy delivers this customer an economic impact. There's a change in the customer satisfaction as they're more likely to stay with you because they're happy and it's a reliable service.”

Reza talks us through how this could work, “You’ve got to first of all Predict. And this isn't easy because this is about building those models and learning from the data before something happens. We've done this in things like predictive maintenance in other industries for many years. We’ve now got some new tools in AI – a different approach because we're using data to guide us.

“Then we've got to Prevent these things from happening by saying ‘Where are those pain points?’ In a telco, for example, there are lots and lots of systems that work together to provide a global service. AI helps us to get to the root cause a lot faster because of that data and the analytics. Then it removes the likelihood of an outage because you're dealing with it as a proactive service.”

Reza continues the process, coming to Personalised, “We talked earlier about augmentation where, if a fault or an outage does occur, you've got the right information available to the agent in real time while they're dealing with the customer, they're not having to give updates further along the line. It reduces frustration, improves customer service and helps a better understanding with the customer.

Talking operational transformation and ROI

For businesses wondering which processes are the strongest candidates for more autonomous AI, Craig will quite often start on internal initiatives to build trust and ROI, “That then helps fund that self-fulfilling cycle of innovation. You make those efficiency gains, you gain the insights that help you improve processes, and that can be done fairly safely behind the scenes.”

He tells us there will be many organisations that haven't adopted AI yet or that are struggling to get traction, “They might hit a stumbling block that they haven't considered, and usually that's something around operational readiness, such as the data, the technology or the integrations, or involving compliance teams, where there's a lot of nervousness.

“These systems aren't always stuff that you can just turn on and leave running. It's best to have someone getting the most out of it for you.

He advises, “The more you can consider those elements early on, the faster you'll get to that ROI, the transformation and the success later. Then make sure that you've got some good use cases and good reasons for adopting the AI that's going to give you the benefits and the outcomes you want.”

This may seem daunting to some businesses who are doing this for the first time and who don’t have people in their team with experience of having deployed AI before. That’s where partners come in, as Craig references, “I would recommend working with a partner because they're going to have the experience of having done this multiple times.”

Looking at it from an executive level, we wanted to know what signals that AI, whether agentic or not, is actually transforming service operations as opposed to simply shifting cost or complexity elsewhere.

For Reza at Verizon, the key signal in back-end systems is reducing structural complexity, “We deal with cloud. They've got many accumulated layers of complexity over the years and those systems are needed to keep our global customer services running. They're dependent on lots of subsystems, hand off between teams and systems, and complexity means it's also expensive to run.

“Where AI changes it is because you're transforming those service operations and then making them hopefully more automated and simpler.”

Reza refers back to when he talked about the post-order cost reducing. If AI can cut troubleshooting and coordinating between teams and silos through automation, this reduces friction and reduces costs in the business.

As well as back-office systems, they also look to see better customer service through augmentation helping the agent, “So from a top level all the way down, you've got front end and back end improvements using AI.

“And then of course the third part of that is improving the service that we give to our customers, providing the right answer, keeping customers updated and keeping them interacting with the business in a more sort of efficient way.”

Reza concludes, “These improvements are very important to modernise and restructure and become a future-facing business.

“AI is not just simplifying and reducing costs, it's improving outcomes, it's improving customer service – it’s a complete renovation of the whole business.”

Craig agrees with Reza on this, “Efficiency is table stakes, that's what you expect with AI. But it's also vital to use it for increasing service, improving consistency, reducing risk.”

Reza reflects, “We're both in technology, but we're seeing that AI is being applied throughout the business. There's some improvement that will affect and hopefully positively change every part of the business, from the front to the back, customer service to HR to legal.

“Everyone's getting benefit out of this because there's a new tool to help us to do our jobs better. It is important in keeping our customers happy. This is actually modernising and improving our workplace, and it's better to see it in a positive light in my personal point of view, rather than be fearful of it.”