Author transcript
Nicole Willing: Welcome to CX Today. I’m Nicole Willing. When a major disruption hits, customer service can quickly become mission critical. Whether it’s a severe weather event, a power or systems outage, or a public safety incident, you know, contact volumes can be sent soaring in a matter of minutes. AI has a role to play here, helping organizations get useful information to customers sooner and give frontline teams more support, but the real test is how well automation works alongside human expertise. That’s the focus of today’s conversation with Rhys Harris, AI product director at Content Guru. Thank you for joining me, Ree.
Rhys Harris: Thank you very much for having me. It’s lovely to speak to you again, Nicole.
Nicole Willing: Great. So, you let’s start with the pressure that organizations might face when something unexpected happens. you know, many AI programs might be judged on a normal day, but then what happens, you know, when a when a major event or disruption causes demand to surge?
Rhys Harris: Yeah, it’s it’s a great question and something that’s absolutely front and center in our mind. So, at Content Guru, we provide the AI [snorts] behind critical moments um campaign that’s now launched across London and and I’m sure will expand across the rest of the globe as well. Um and it’s always been in our DNA to look at not only um what happens during a business asusual day and how you provide great CX in a business as usual day but how ultimately you can cater to whether events, outages, public safety, incidents, disruptions, surges in demand. I mean the key thing is really nothing should change. Um I always describe it as um you know your service needs to be like a glorious white swan. It needs to be serene above the water. uh whilst the legs, you know, your infrastructure layer, your application layer, your service layer, the the design are all doing the hard work and and paddling away under underneath the water. That’s the the feeling that your customers expect and that your customers, you know, expect to experience 24 hours a day, 7 days a week, 365 days a year. [snorts] Um ultimately during these spikes in demand, the dials go up. If you’ll be utilizing more compute, there’ll be more tokens, but the customer experience absolutely has to to stay the same. So I suppose from a a technical and service design aspect, there’s a lot that needs to change to be able to cater for these events. From a uh actually how does it look to the outside world? Um really nothing nothing should change on that.
Um the only way that it it might do is if potentially you’ve um had a public safety incident and you’re getting a lot more critical or complex um or emotional inquiries, you need to be able to handle those um in in an appropriate manner. And where you have the human um capital to to be able to support those inquiries, get them handed off very quickly. where you don’t you need to utilize things like AI to be able to really capture and and adapt to those demands, capture important and and necessary information such as potentially contact details, etc. If if somebody gets cut off, you need to be able to proactively uh be able to reach out to them again and to make sure that that information is is well handed off.
So uh ideally nothing should change but ultimately if if it’s a you know an event that actually changes the natures of some of your inquiry rather than just surging demand um they do need to be handled differently. Um and ultimately as said it’s going to depend on uh a lot of influences such as the data that’s available your humans that are available to handle them. Um, and your AI needs to be able to adapt to that and and ultimately um provide value still in those circumstances to capture critical information and make sure ultimately it’s it’s handed off well when somebody does route through to a human adviser.
Nicole Willing: Exactly. So where is the balance here? You know, where can AI kind of safely absorb routine demand during those moments and where should organizations be careful not to automate?
Rhys Harris: Yeah. So again, this this is absolutely something that’s baked into our methodology when performing any sort of full AI automation be that um for for digital channels or or for for voice channels is we follow a what we call the Q model. Um, and I mean Paul Q, so CUE. Um, and that stands for is something complex, is it too complex to to be handled by AI? Um, is it urgent? You know, something that’s really time dependent or potentially mission critical, or is it something that’s particularly emotive um where actually AI might not be the correct correct way to handle it? And you need to make sure that for for any interactions that fall into that small subset that you’ve got the correct guard rails or or escape hatches to be able to make sure that those interactions get to the correct human quickly. And if you haven’t got humans available at that point in time that you capture the the really critical information to if you can automate them but ultimately that that you can still pass them over to a human with a good set of context that that’s sent across.
I mean ultimately um where it can safely absorb routine demand during those moments depends on the service and depends on the nature of your organization. Um you know if if if potentially you’re working at a police force, there will be emergency services inquiries that you really wouldn’t want to automate. [snorts] Um but maybe you know some of the non-emergency services certainly could u you know take AI triage or could even you know utilize agentic AI to to log cases etc on on on behalf and soak up some of that sort of routine demand to focus your your humans on dealing with the the important urgent and emotional topics as well. Um ultimately it boils down to AI is is inherently probabilistic. it’s getting a lot lot better at um you know and continuing to get better at being able to achieve consistent um out outcomes particularly for complex tasks nowadays. But if you say actually I can’t afford 0.1% of the time for for um this to um be automated by AI and not necessarily get the correct outcome, you will want to either automate it via a deterministic pathway or do some very basic information capture and make sure you get it to a human who can handle it correctly.
Nicole Willing: Definitely. Um, and so then on that, you know, what does a poor kind of AI to human handover look like, you know, when someone’s already under pressure? Um, so then what information should the human agent receive instead?
Rhys Harris: Yeah, I mean, it it’s going to again very much vary by industry. Um, but typically the rule of thumb is you never want people repeating information. That just makes an individual feel like they’ve not been heard, that they’ve not been listened to, and that they’ve actually just wasted their time. um if the human agent goes back through everything that’s that’s been covered, it it’s just a waste. It’s a waste of your tokens. It’s a waste of um your human agents breath. It’s a waste of your customer’s time. It it it’s not efficient and not particularly fun for anybody. Um so really making sure that you know you’ve got the capability as we do to provide a full transcript of of a conversation plus a clear and accurate summary written straight into a CRM or customer data platform is is is key. Uh if you’re not doing that, if you’re not logging the fact that somebody’s spoken to an AI and you know, maybe it’s even automated their whole inquiry, but haven’t written that to to their customer record, people are still going to have to repeat that information if if they do speak to you next time, which is is going to be a poor experience as well.
Um you want your AI to be able to recommend um next steps in the conversation. um that can ultimately, you know, give the agent the capability to to easily ask those next steps, get the the conversation going with the caller whilst they brush up on some of the other items in the customer data platform for for that individual. Um the information ultimately, you know, that that’s required to be captured depends on use case, but in a lot of cases, um capturing at least name, contact details, and locations are typically very helpful, especially for emergency services. Um but ultimately you you want to make sure that this portion of the journey is is designed well is unique to to your use case but also takes into account um the context of of the individual that’s called as well. Um if they’re distressed or vulnerable, you actually don’t want to spend a lot of time in the AI. A better handover is actually not capturing that much and and and being able to pass them across to a human as well, particularly if if that resource is there and available to cater for their interaction.
Um, so I think that for me really sums up, you know, poor AI to human handover is is where information is lost. You’re making people repeat themselves. What good looks like does depend on the use case, but making sure that the conversation’s logged, that you’ve got the next steps ready for your human, that they’ve got a concise summary that they can read that’s not just distracting them from, you know, greeting and and welcoming the caller, um, giving out compliance statements, etc. is, you know, massively key.
Nicole Willing: Absolutely. Um and there’s an opportunity there you know to the reduce the number of people who might need to contact the organization in the first place. Um so how can you know proactive communications digital channels machine generated signals really reduce unnecessary inbound contact you know before those contact center cues start building?
Rhys Harris: Yeah I mean look for for me um you know obviously locked to have a conversation with somebody but I I really want it to be my last resort. I’d much rather get information proactively and I think you know wide array of of the populace feels feels very similar. Um you know research has shown that particularly for younger generations they want to be a little bit more self-sufficient. They don’t want to be guided through everything. So if um you know you can do a bit of deterministic automation to understand their inquiry, detect that actually you’ve got a web form or a web guide that might they may not have been able to find that could help them performing a proactive communication via an SMS via an email if you’ve captured their email address or or have have that on record is is really key or you know if they come in via a digital channel having a chatbot that signposts people towards those um can can be massively important and you know IoT or or as we call it here the the dig digital customer um again is is becoming and continues to be more and more important.
Um you know there’s going to be hundreds of billions of of IoT devices by 2030. That’s a massive array of data that can be captured and ultimately all of that data can be utilized as as context to best best serve an individual. So, you know, for example, um you know, you could utilize a smart meter in the utility space to detect and on its last gasp a postcodewide power outage and instead of as it was in the early 2000s where you’d scramble around with a torch and um find your bill and find your telephone number to call in, sit in a queue for 4 hours whilst you waited alongside everybody else. actually an IoT signal coming through a platform like ours giving its last gasp to a proactive SMS messaging system that can then blink into um you know an automated fault system or something similar ultimately means that by the time my power’s gone off and I’ve thought about contacting somebody actually I probably got an SMS message already you know within 10 to 20 seconds of of my power going off to say we’re aware of an outage and we’re on it and that’s a much much better experience as well it prevents me needing to call in as as the consumer, but it also means that um you know still if if there’s um customers that certainly do need to contact those uh organizations where you know maybe you’ve you’ve got life support equipment in your house etc that they’re actually not sat in a queue and and are able to get the correct support that they need very very quickly. So it’s it’s a win-win on all fronts to be able to proactively communicate through digital channels IoT or or otherwise um to improving your customer’s experience.
Nicole Willing: Exactly. And then you know the the other question then about the role of AI is how it helps the people who are handling the conversations you know because AI is often framed as a a way to reduce agent workload. Um so you know how in in these mission critical environments can it help make humans um you know more effective in their job rather than simply reducing the number of agents.
Rhys Harris: Yeah I think avoiding sort of reduction of agents and full automation and focusing on the points that can help. I mean number one you know platforms like ours that can provide um you know agent assist solutions that are bespoke to to an organization are massively key and there’s a couple of of really important aspects for for that as well. So number one, you know, having an accurate transcription, having a a tailored summarization for the inquiry at the end of an interaction is is really key because um ultimately you don’t have to try to multitask. Then as a human agent, you don’t need to sit there and scribble a lot of notes, put somebody on hold whilst you you take notes down or ask them to repeat themselves. you’ve got that fall back and you can really focus on the conversation that that you’re having at hand and knowing you’ve got the safety net of of AI in quite a basic way in that use case to to support them.
Um the second part is you know ultimately if it’s a mission critical environment typically you will be dealing with you know very important conversations that that might be urgent or emotional um and actually you know humans will need a break from that. there’s only so many um you know call after call or interaction after interaction you can handle that might be very stressful um that that you’re able to do in an effective manner. So having something like a sentiment analysis as we do that can proactively notify supervisors where an individual has had multiple backtoback very stressful or difficult conversations can make the human agents more effective as well because it gives them that chance to maybe go on over to a lower stress contact channel quickly. Um you know take a couple of hours to to reset before then coming back and providing great service on on some of the more difficult conversations again.
um you know other areas that are are massively key um you know in Britain obviously we and a lot of countries now very multicultural with lots of different languages that can be spoken particularly if you’re working within a public services organization um you you know you might have to contact a human translator that can take sometimes quite a long time to to be able to reach those services can be very costly so utilizing AI not only to translate digital interactions but now also voice interactions for us we’ve we’ve found that’s a great way to to support people because you’re not just keeping a citizen on hold for a significant period of time whilst you get a human adviser on to to be able to do that. But you can also utilize it to number one coach them across the whole portfolio of of interactions and conversations they’re having and look at their whole workload and how best to support them. That’s going to reduce friction in in the workplace. You’re going to have less um you know discussions between an agent and supervisor. Well, yeah, I didn’t perform that well on this call, but there were 99 other calls that I did perform very well on. You’ve picked the wrong one. It reduces all of that friction. Um, but ultimately, you can then utilize AI based off um the full view of of an individual to help generate, you know, tailored coaching plans for some of their areas they’re weaker on. The other item there is, you know, AI doesn’t hold any human biases. So, um it can’t be influenced the fact that you might have had a falling out with your your supervisor a couple of days ago or something similar. those sorts of bits don’t creep in. I mean, ultimately, they shouldn’t creep in in a human environment, but it is natural that they do. So, AI not only can really help um individuals and human agents during the interaction with with how they manage the case, but also actually in in the grand scheme of how they handle all of their interactions by analyzing and coaching them after after each as well.
Nicole Willing: So, you know, we’ve talked about keeping services available, you know, handling demand, maintaining continuity and and giving frontline teams better support. So then the the question from there is how organization can turn those ideas in into something they can put into practice. So you know if if a CX leader wants to prepare for that next critical moment um what’s the first kind of practical step they should take?
Rhys Harris: Yeah, I I think again um so we utilize um we utilize a pyramid for categorizing our interactions for customers and at the bottom there’s going to be a lot of um you know very repeat low urgency high frequency interactions that you know really you can can sort of discount from some of the the critical moments piece particularly if you’re then able to automate them you can focus on the more important topics that that come at the top of the pyramid. Um the key bit then is ultimately if you’re utilizing something like um uh you know interactive voice AI or or or chat bots etc is making sure that you’ve built those to a framework that supports um critical moments in in the best way. And what I mean by that is for us uh we have an automated framework where actually handling any of the critical scenarios the the complex the urgent or emotional scenarios and and failing on those scenarios counts as as a fail for the implementation. You need to make sure that those are are handled correctly and your services need to be built in um you know built with with that in mind and built in a way that you’ve got the correct guard rails to uh pass somebody off to to a human agent if they’re available and if um the AI detects that an interaction is is urgent or emotional.
Um, so I think a lot of it goes into to the service design and understanding your customer inquiries, having the correct data to do that and thinking about, you know, what are the fringe cases that may come into us um that we really need to handle with with care. So great planning and working with a team of of four deployed engineers like ours. We will help coach customers through those particularly um you know we will know from our experience working with a lot of customers in similar verticals what they may have seen that you as an organization haven’t yet or maybe haven’t considered. So, I think that’s number one is is making sure that you’re working with a a service provider that really understands the industry that you operate, what the critical moments could be and how best to support those, be those, as I said, spikes in demand or something that’s critical to an individual, you know, could be critical to me if if I’m on holiday and I’ve lost my credit card, you know, that that’s that’s a critical moment. Regardless of whether it’s a critical moment for the whole populace or or just for me, still critical that that gets handled correctly. And as I said, building those in by design is is massively important.
The other area is there always will be bu business continuity events. Um we like to think that the worst won’t happen, but we absolutely have to plan for for the worst happening. So making sure similar um you know AI is very similar to to any other technology in this case that you’ve got your disaster recovery and business plans um business continuity plans in place is is really key. So again, for us, it’s it’s not just about how do we handle the context of a customer interaction for for something that might be a critical moment. It could also be what if there’s a massive surge in demand? What if there’s a an outage with a vendor that we orchestrate with? And for us, providing an orchestration layer is is great for that. Allows us to ultimately flex with the demand of our customers. Having that orchestration layer and working with a a vendor or service provider that provides that massively important as well. So, as in all of these cases, it goes back to great planning up front. Um, but also you need to be continuing to keep an eye on those disaster recovery, the changing circumstances of the services you provide. Um, and then running, you know, similar type of break glass exercises on a routine basis as well. And, you know, for us and our customers, we’re very much here to support that for all of our expertise as well.
Nicole Willing: I think that that’s a a great takeaway for our audience and a great place to finish. So, thank you, Ree, for joining me.
Rhys Harris: My pleasure. Thank you very much for your time. It’s great to be on again.
Nicole Willing: Great. And thanks to our audience for watching. To learn more about Content Guru, you can visit their website at contentguru.com. And for more interviews, news, and analysis on customer experience technology, you can find us on cxtoday.com as well as LinkedIn. Thanks again for watching, and I’ll see you next time.
