University of Cambridge Professor Mohamed Zaki answers your questions about using agentic AI to enhance customer experience – from what use cases you should prioritise to how you can avoid losing trust.

“Agentic AI agents have the potential to orchestrate customer-facing workflows and connect what are typically segregated functions in established firms,” shared Mohamed Zaki, Deputy Director of the Cambridge Service Alliance and Principal Researcher at the Institute for Manufacturing, University of Cambridge, in a recent webinar.

Agentic AI has become an essential tool for organisations that want to stay competitive. It doesn’t just present opportunities for greater efficiency and more strategic insights internally, it’s also affecting customer decision-making and choice.

“Gone are the days when you engage with customers through your product prices and services. Now everything is experience-led. Your content online has to be spot-on for agentic AI to recommend you,” said Professor Mohamed Zaki.

“This will be the future of how customers interact and make decisions. So organisations need to learn how this technology can work for them, not against them.”

So, where do you start? Watch this webinar with Professor Mohamed Zaki to learn how you can use AI agents to enhance customer experiences:

Here are some of the answers Professor Mohamed Zaki shared in the webinar to live questions from participants.

For organisations at an early stage of adoption, which agentic AI capabilities should be prioritised first, and why?

Professor Mohamed Zaki: My advice is to ask yourself what the best use case will be to resolve a problem that you or your customer has. The answer doesn't have to always be agentic AI. There are simple ways to leverage AI more broadly when you have a pain point or friction.

It’d be the wrong advice for me to say you should go jump on the wagon and start to leverage ChatGPT or Claude for example. You have to start with the business problem.

You also need to assess the maturity of your company in terms of data foundations, because if you don’t have the data you need, the agentic models aren’t going to work – they won’t provide you with the right insights to make strategic decisions.

As a first step, you should identify the business problem. Then define which would be the right technology to use. The check the maturity of the company and its data foundation. To what extent you need new infrastructure, capabilities and integration capabilities to make everything ready for the customer.

That’s when you’re ready to pilot and measure the impact. Then, once you are really satisfied with the impact and have the right configuration, you can launch that service to your customers and measure how it works over time.

What are the differences between agentic AI and generative AI?

Professor Mohamed Zaki: Let's go back to the evolutions of machine learning and AI. So, we used to have different fields here. Some of them using quantitative data, which we leverage to understand patterns from data, like sales figures, using traditional machine learning exercises, like classification models etc.

And then we have another discipline which we call natural language processing, which deals with textual data. And this has seen a huge evolution over time. We used to use it for requests like ‘go capture specific information from a corpus of text’, like customer service data or qualitative reviews.

This is where the new generative AI is starting to come into play. And the generative AI is basically the evolution of natural language processing that changes the way we can interact with, analyse or generate text.

Now, where agentic AI came in was when these architectures we trained with billions of parameters, coming from public data like Wikipedia, social media, etc.

Now we're starting to observe that this technology does things beyond generating text. There’s no science yet as to how. But basically, if you input large amounts of data into these models, with the right cloud infrastructures, it can now break down tasks and build reasoning and logic into its thinking.

The difference between this and generative AI, in my view, is that generative AI is more limited. At the end of the day, it's a predictor of the next word, so sometimes the answer will be different, based on the context you input.

How do you choose which one to use?

Professor Mohamed Zaki: It needs to be based on your objective. If the objective is, for example, generating an answer and response, this is where generative AI should be used, because it does well when it’s generating data.

When it comes to agentic, it can work on a more complex tasks, where breaking down the task, reasoning and understanding the order of actions are important. For example, retrieving information from different systems.

Bear in mind, when it comes to agentic, you can have an agent with machine learning capabilities communicating with another one that has generative capabilities. It's more about how you design architectures to achieve your objectives.

Based on your research at the University of Cambridge, are there any governance frameworks or design principles that you see as essential to ensure agentic AI enhances rather than erodes customer trust in service experiences?

Professor Mohamed Zaki: Governance is very important activity. Executive teams have to put a strategy in place to decide which technology should be used, which technology should be invested in, and which technology has the right return on investment. You also have the ethical and regulatory considerations.

Obviously, it's not a bulletproof technology. It could fail, and that could cause problems for firms. That's why governance structures – including which technologies we can use in specific cases, which use cases we should discuss and apply, which ones don’t work for us – are really important. You need to build the right frameworks to enable this to happen.

You always need to start by qualifying the right use case, planning where to focus your effort, choosing the right technology for your goal, designing, experimenting, then piloting and testing your capabilities until you reach a maturity where you can institutionalise this AI and introduce it to your customers.

How can different departments use agentic AI?

Professor Mohamed Zaki: I’ve seen that a lot of the AI applications in companies are around helping employees use the assistant AI capability, to increase productivity inside or outside their organisation.

Another example is technology teams – there are huge use cases for documenting architectures and writing the business requirements to help programming and data science teams structure a product.

There are massive improvements in productivity for that team as well, to come up with new digital services to your customers and design and pilot and experiment and build the digital app out of it.

In different functions, like HR, activities like onboarding new employees take a lot of time in big established firms. Writing standard operating procedures is another big use case, where a lot of firms generating a lot of different processes for customers.

On the operation side, obviously there’s supply chain visibility. Accessing data points like news, orders or supplier information, understanding capabilities and networks of people based on data. So, for example, if you have an issue with one supplier, or there is a disruption in the market, you can spot an opportunity for other suppliers to intervene.

This webinar is part of a regular series where we discuss current industry insights and trends with expert University of Cambridge academics. We encourage you to register for upcoming webinars to take part in these discussions live.

Want to build AI capabilities in your organisation but not sure where to start?

Gain essential insights into AI’s role in seamless customer journeys in Professor Mohamed Zaki’s course Data-Driven Design for Customer Experience (CX). Or take a look at our full range of technology and data courses.