In July we hosted our annual community event, Board Masters.

As well as showcasing our roadmap and gathering feedback, we also hosted an insightful fireside chat to discuss strategies for implementing AI tooling and how to use additional capacity for proactive support.

We were joined by Daniel Bunton, Head of Customer Support at Cleo AI, and Shamas Aziz, Director of Partnerships, SentiSum (and ex-Ocado, Selfridges, Matches Fashion, Net-a-Porter)

Industry experts Dan and Sham shared valuable insights on the implementation of AI in customer support and its impact on proactive working. As AI continues to reshape the landscape of customer service, their experiences offer a glimpse into the future of support teams and the opportunities that lie ahead.

How are you using AI and how has it benefited you?

Dan:

Our chatbot is AI at the forefront right? So we were there before LLMs became big LLMs definitely improved our product. So what we did was we integrated with the LLMs and so where before. Our AI was very simple and you might say hey, I’m struggling to add my card to my account to pay my subscription. We would hear the word ‘card’ and we would go, ‘you would like to sign up for our credit card. Here is the sign up flow’, so that was our AI before. LLMs radically changed that. So what we did, and what we found immediate success with, was just monitoring intent like understanding what a user was thinking.

I think right now, for AI in it’s current implementations, that is what it can do very, very well. So what we’re doing right now is ai’s so elements are listening to the intent from the user says and then also on the back end it’s saying, ‘Hey, these are all the buttons that you can put in front of this user. What button do you think we should put in front of this user?’, and it is that now successfully picking which buttons to put in front of the users. So that is a very high engineer heavy implantation of AI where we’ve done but yeah so the first one low-hanging fruit would be the imagination

Sham:

Intent is something that you can use very quickly and you can automate that overnight. It’s very low-hanging fruit and you can respond almost immediately.

Don’t listen to the providers that tell you it’s like 70% automation. The sweet spot is 30%, because really, you don’t want to throw your customers into that death loop of sending an email and getting a response that then just go running in circles. So I think that practical application is really good at not contact reduction, but it helps your team to focus elsewhere.

I think the second one which picks up on the internet that you’re hearing about is the implementation of things like auto-tagging your contacts. Using AI to do that is a much better way. AI doesn’t get tired. It’s more consistent using LLMs as it can actually pick up the intent that you can’t do. So there’s research out there that in the contact centre the average accuracy for a tag on a contact added by a human is about 50%. With AI, you’re looking at 80-90%, so actually, if you’re walking to the exec boardroom and you need to tell your execs what your customers are saying, you could do with the confidence knowing that they’re not going to read the ticket and find out that you are completely wrong. So there’s an element of using tags and using them in an easier way.

So I think intent is a very good use case for AI. It’s very low hanging fruit available today and so you could turn that on quickly and get to the root cause of things without using so much of your gut. It’s good to combine the two so they’re too practical. Use cases that you can start with medium.

Dan:

One thing that we’re expanding quite rapidly is annotations, so we have annotation teams.

Essentially in AI there’s there’s a phrase called ‘human in the loop’. Have people heard this term, in the loop? Yeah great, so basically humans are going through and looking at conversations and going like, ‘okay, was this good? Was this terrible?’. So there are a whole companies set up now to do that for you, and so we are outsourcing to one company and suddenly they have got you know, I think 100 people annotating our conversations, and so they are going through our conversation saying, ‘yes this is good, no this was bad’.

We we try and get a very specific tone of voice in Cleo, we try and call her like our edgy big sister, right? Like, she’s going to give you attitude, but she does love you and it is coming from a good space, and so we’ve got humans going through all that.

And so you can also use your team’s product expertise right? So, what we found really useful is putting our agents downtime onto annotations and so what we’re actually doing is going who’ve got the these external annotators who are pumping through hundreds of thousands of annotations reading what Cleo was saying to our users and then we’ve got our customer service agents who are then like, in their downtime going, through and going, ‘yeah, this is good, not as good, not good’. They are seeing completely different things, and so like, I’ve said this a million times to my CEO, I was like, I will put money on my agents knowing your product better than you do, because a customer support agent understands the product so much better than the the exec teams, right? They understand it so much better than anyone else, even product managers as they usually work in a very specific area of a product. They do not understand what’s happening in the rest of the product. But a well trained agent can understand your product better than 9/10 of your company, so we are putting another kind of work in their schedules to help our AI get better and better.

Sham:

SaaS have matured slightly, it’s become more accessible. So previously you’d need an engineer to get involved to build something, now with the advent of SaaS businesses, and they’re not all evil by the way, you can access something really quickly. With cloud-based, two-click integrations you can get live in one afternoon, really quickly, with a pilot or proof concept.

Also the accessibility and pricing models have moved on a lot. It’s not the same way, you know the beginning of the year you pay a year’s worth of contract up front. You don’t have to do those things anymore, so I think it’s become accessible through removing the barriers.

All the types of barriers you’d see in a typical contact centre around with me returns the SaaS industry actually took that away. They solved it. There were payment issues, they’re removing them. They saw there were challenges with signing like three contracts, so they’re removed.

So I think the accessibility, and also the single-click integrations and barriers have been removed now. If you’re not using AI, it’s almost more fool you, because the benefits are really quickly starting to outweigh the horror stories.

Dan:

It’s like when ChatGPT was launched. I’m sure you all remember what what your first time conversation was with GPT? No, I don’t actually remember either. But like, the conversations you can have with GPT are way, way better than what AI was doing before. You know before before LLMs.

So yeah, this ain’t going anywhere. There is money to be made. Capitalism will make it work. So yeah, I don’t think it’s going anywhere. Mainly because it’s actually impressive.

Do you have any advice for how to introduce AI into your team?

Dan

We have a monthly customer support all-hands call that we do. and when LLMs first launched, people panicked and thought, oh my god, I’m going to lose my job. And I think that panic has now subsided because it’s like, okay, your job’s still in place six months later.

But yeah initially we’re just like, let’s over communicate. Let’s answer all the questions they have. Our CEO is really good about it, he was like, we’re not going to fire anyone because of AI.

But then with those performance conversations with the team and it’s like cool, your your jobs never again going to be the same again, it’s going to become more complicated. You’re going to work on a higher complexity and level of tickets in general, and the other thing is moving into this annotation space and opening up opportunities to go after as the team is growing.

Sham

We’re also giving them more difficult things to do, and so I think there’s an element of how can we be smarter about this? What we’re finding is, the contact centre is going back to the customer journey, whether that’s to your supply chain, product or the ones in marketing, and saying, ‘Hey, sort that campaign code out, or hey that supplier is always late all the time’.

It’s not just about what you pay for the cost of delivering a package and it’s sort of holding different parts of the business accountable with your information you have from your support team.

The agents can actually be the voice of your company back out to the business and to customers and then they can become the profit centre that we’re sort of trying to get to, so that I think we should make it easier for them.

So, in terms of transition-land, I think there’s easier use cases that you can bring in, which I discussed earlier around automation and the agents holding the business accountable with real information that you’re picking up from contacts. There’s so many sort of ways to transition into it.

I think that the red flag that I picked up on is that when you take it to like higher ups it’s going to be much harder for you to discuss the hard things like metrics. And I’m still trying to get my head around it too.

For the C-suite it was like, ‘hey our cost per conversation is going to go up. Our resolve rate may go way down because suddenly we’re dealing with really complex issues and they’re going to take longer to solve.’

The first time I sat down and I had to explain to the execs why NPS had dropped by two points because we’ve introduced automation. I don’t know if there’s a reason why, because at some point it will get better and go up again and he was just like, ‘okay thanks’. And it was one of those ones where they still didn’t get it. and it was at that point where we just had to leave it with support and trust the system that in the end is it all for the right reasons.

Sham

There are two main skills you need to sell anything. One is stats and the second one is storytelling. And I think storytelling’s probably an underutilised skill that anyone and everyone in any industry can learn. I spend a lot of time telling my son, who’s 11, about getting the storytelling right to get ahead in life.

I think if you can get into a boardroom and you can add a few stats and then provide a story, and I always like to name my customers, so when I go in I’ll talk about Anita, I’ll talk back Jack and I’ll say Jack did this and do you know and what happens then? It’s the stuff where the execs can relate that to the customer and empathise with very quickly. The CFO is usually the last one left that you will need to turn around and usually that’s where the stats come in where you can say, ‘hey, here’s the ROI’.

We’re in the middle of economic climate where people want ROI immediately. Sometimes it’s not always visible in 12 months. So how do you show it over 12 months, 24 months, maybe 36, but not just the low-hanging ROI which is X pounds equals Y benefit. It’s actually going back to the customer story and saying it’s that old journey that caused a problem.

Could you imagine yourself, your partner, your children, and somebody that you know going through this? That right here is actually part of the ROI and we’re going to eliminate that over time and soon. So, it’s not looking for everything you need it in terms of application, but building towards the future and just reminding everyone, including the CCO, to unlock the money for us and our customers.

Dan:

Right and if you can connect it back to the company that’s always going to work. Like, what is tanking conversion? What is tanking retention? So, if you can connect a major issue that you’re seeing in customer support to retention and or conversion you’ll always get buy in.

How would you recommend leaders use additional capacity for proactive, profit driving activities?

Sham:

It’s the WIsMOs (where is my order) which we can’t get away from. We’re in 2024 and we’re still talking about it, tt drives me insane but I’ll solve that on another day.

A quick easy one that we used to do at Ocado, for example, is we have peaks and troughs in every calll centre with availability, so we used to get advisors to ring customers and tell them, ‘hey your order’s going to be a bit delayed today’, so that was a nice sort of personal touch and it was using your existing head count, so no additional cost to stop doing that.

So that was one sort of implementation of getting to the customer before they got to us. So it was this one thing that actually drove up CSAT because it was a nice surprise for the customer to get a phone call from a grocery provider.

The other example, though, which is more interesting is a Selfridges example, again dealing with WIsMOs. So we found a way to automate WIsMO queries. They’re using AI, so there was extra time that we then got back and rather than make redundancies, which is point you touched on earlier, we then redirected those staff members to do outbound contacts, and so, ‘let me help you dress you for that party that you’re going to. Did you consider this purchase? I see you can make this purchase before’. And this, with came with some sort of rules around opt-in etc, but it was just making the job more interesting for those advisors who were perhaps interested in women’s wear or into accessories. And so these are sort of two examples of using your resource in an outbound way to add value, not just to the business but to their day and they’re working ways and I guess it’s a win win.

Dan:

So there’s the annotation thing that I was referencing earlier, and the other area like talking about the proactivity thing and I love the your ‘might be delayed’ thing.

One area that we’re really noticing is that so many people cannot be bothered to reach out to customer support to solve their issue, right? Like, their intent needs to be really high to use your product, to go through the hoops of reaching out to customer support. Either they’ve got money on the line and they want a refund, or they really really love your products, right? And so what we were seeing with some of our major issues was that so, for example, we had this problem in the conversion funnel, where users couldn’t set up their account properly and we were hearing from roughly, you know a couple thousand people a month on this. But then, when we did some research we found that there’s actually three other people who experience that error that did not reach out to us. And so now we reach out to them next day and say, ‘hey did you get this error? I’ve gone into the back end and fixed that for you’. And so in that way, we’re actually moving into like proactive customer support.

Sham:

I think the other use case for me is how can you sort of jump into that profit centre? So for as long as we’ve been talking about user flows and we’re all saying talking about being a profit centre and in our call centres it’s not always obvious, but in today’s economic climate it becomes even more obvious around, you know, well, how can we help people sign up?

That’s probably the onboarding journey, so you find that people join the journey and then drop off. If so, how could you help them sign up, so you can do things that don’t scale? Because with AI, you’ve probably now got that availability within your team to do some outbound work.

So I think it’s identifying how you can move that needle. If you’re then helping generate additional sales or additional conversion or additional attention, like I was mentioning earlier, the perception of your contact centre starts to change. Then when you go forward and ask for more investment and more tech and other things that you want, you’ve already built that relationship. Now, it becomes easy because you’ve built trust back into business that you’re not just a cost centre.

How has your relationship with your BPOs (Business Process Outsourcers) changed?

Dan:

We have restructured a number of times post-LLMs, but we haven’t had a situation where we need to radically downscale. We’ve built a really great relationship with our BPO.

Sham:

There was a period where we would say you wouldn’t want to outsource certain things. But I think now you need to think of it differently.

If you are a brand that specialises in, let’s go with sneakers. If you do sneakers then that should be your point of difference as a brand. Why don’t you be the best ever at designing shoes and selling sneakers, and then therefore if you go to a BPO and you assume at the beginning that actually, they are really good at supporting customers and providing care, then your relationship would be as good as possible with your BPO.

So let them be the experts of doing care and you can be the experts at sneakers or whatever your product is. It’s that simple. So for me, if anything having a great relationship with your BPOs who are dealing with your customers who are doing the thousand WIsMos and the thousands of tasks having that great retention, then it’s helpful. And then with the implementation of AI, whether you do that yourself as a brand or you ask your BPO to do it, you’re making everybody’s job easier.

So, I think there’s another point of creating a partnership, and moving away from this sort of supplier relationship, because in the end the customer’s interacting with the brand and the BPO so it’s the same to them.

If you have a problem in the custom journey you’re unsure of how to solve it, go and speak to your BPO as they are the experts. So when you’re sitting there and you’ve got problem and you’re scratching your heads saying, how do I solve this for my customers? They can give you 10 use cases immediately from the other clients that they have. So you’re not having to sit there and figure out and start from scratch. You can use the industry knowledge of your BPOs to set the direction.

Dan:

We’re getting our BPO to do annotations as well.

They’re just like our internal team, we interview each of them and train them. So in Surfboard we’ll create their schedule and say, ‘hey you’re on annotations for these 2 hours. You’re on chat for these 2 hours’, just like we do with our internal team. And so yeah, we don’t really see a massive difference except it’s much less of a kind of operational lift.

What kind of deflection rate have you seen from your AI?

Sham:

It’s not about deflection, it’s all about customer experience right? So without getting on my high horse, deflecting the customer is not an experience we want, so we shouldn’t be deflecting any customers anyway. And that’s why I prefer resolution. So if something has been contained and resolved, let’s call it that. It’s been resolved in an automated way, and then if it hasn’t then it can go through to the team.

Do you have any advice for how to identify proactive opportunities?

Sham:

So look at the customer journey all the way from onboarding, pre-purchase, order has been placed, post-purchase; understand those points. And to have good analytics, you’ll understand what’s happening with your customer journey and then you can figure out how you can add value quickly.

And so if you know for example that X thousand customers are getting in touch about a post-purchase exchange issue, then how could you as a team address this in a scalable way? So things like making outbound or also sending emails. Or, for example, at Selfridges, we used to send three people from the call centre into the distribution centre to go and get the returns and then manually push them through faster so we can push refunds through faster.

Now, I could have said at that time we’re in customer care, why am I using my capacity in the distribution centre? Surely, they’re benefiting from that additional three head count that I am spending on. But the end result was that refunds will go through faster, so I was getting less calls.

So always go back to the customer journey. Look at the issues, use your analytics and then think about how you can fix this stuff with people in your contact centre.

What do you think the future of support is?

Sham:

I envision a world where there’s an LLM for WIsMO, returns, exchanges, getting all of that knowledge base type situation from inside the team.

And then also adding more context. So as the customer speaking to you or they’re sending an email in or there’s a chat session. Your CRM, like a Dixa, would be able to see that and match the right article, but not 26 pages of an article, the exact paragraph that you need to answer that response to that moment in time, so you’ll drastically reduce your AHT and I think that would tie back to forecasting and planning. So in any contact centre your biggest bang for your buck is getting the right people into the right place at the right time and that’s something that I would use Surfboard for, to make sure that my adherence is correct. So this isn’t about reducing people, but it’s about having the right people the right.

Dan:

Yeah that’s massive, and I think that the obvious next one for me and where my brain goes next is training and quality assurance – highlighting areas where you can improve. Like, I want it to read everything, I want it to understand who’s doing the right thing? Who’s doing something completely new and innovative, that’s actually amazing, who should we elevate? Who should we like paying attention? Yeah there’s the sky’s the limit.

What would you do if you had £100m investment for your support team?

Dan:

Firstly, I would give everyone a pay rise. They all deserve it. I would love to give everyone access to our Amazon account and let them give out gift vouchers, so if you think we’ve messed up and we owe them something, then send a gift.

What I do is I really start to kind of like squad-ify each agent. So can I put can an agent, an analyst and an engineer and they they all work together to solve users problems. Like one of them is talking to the user and understand the problem we need to fix for that user. The other two are going okay is this just them or does it affect everyone else. How can we also help them? What can we do here? So yeah there’s that and then I’d take the whole team to Spain, obviously.

Sham

So definitely, we should pay more. There’s a big thing around real living wage and London wage and actually, that’s the better minimum of my view. So there’s a massive retention and attrition issue around stopping recruiting within contact centres in the world. No surprises right, we’re paying at the bottom of the market or just above where we’re supposed to, and so that’s not the most exciting career. Most of us have built a career through it, but we’re in quite a small room with small number of people.

So actually, I’d reposition that role and time and pay more like you said and actually go out and just recruit anybody that wanted to help customers and be in that sort of coordination, troubleshooting space. Not necessarily something that needs customer care experience, because ultimately that’s some of the stuff you can train, I think including product.

I would use some of that warchest to identify the real issues, through customer care and use that money to walk into a supply chain or into the marketing team and say on Wednesdays there’s a problem, here’s a brief and some investment. This is how you’re going to fix it.

And I think I would build this mad LLM sort of model. I imagine Demolition Man, the movie, the original right? So this is, but there’s a bit in it where Wesley Snipes walks up to computer on the street and he asks it a question. Then he downloads Kung Fu, he essentially downloads knowledge and then he immediately uses that knowledge.

You could get the answer to any question that you wanted and somebody knew how to ask it. The right question to do this. I think you should empower agents got the best possible tools ever with the right salary and power to go and do it.

Key takeaways

The discussion highlighted that AI is not just about automation or cost-cutting; it’s about enhancing the customer experience, empowering support agents, and driving business growth. As AI continues to evolve, it offers exciting opportunities for support teams to become more proactive, efficient, and value-driven.

The key takeaway is clear: embracing AI in customer support is no longer optional. It’s a transformative force that, when implemented thoughtfully, can lead to happier customers, more engaged support teams, and stronger business outcomes.

As support leaders navigate this AI-driven future, the focus should be on leveraging technology to augment human capabilities, fostering a culture of continuous learning, and always keeping the customer experience at the forefront of innovation.