THE HUMAN ALGORITHM: WHY THE FUTURE OF AI IN CX MUST STILL FEEL HUMAN

I find there is a strange irony that is emerging in Customer Experience interaction and strategy.

Brands are investing billions in Artificial Intelligence to make customer interactions faster, easier, more personalised and more efficient, while simultaneously running the risk of removing the very thing customers often value most from those very interactions: the feeling that somebody is there and actually cares.

Long before ChatGPT, chatbots, Agentic AI or even the internet existed, psychologists were already interested in what happens when human beings communicate through technology.

In 1976, researchers John Short, Ederyn Williams and Bruce Christie introduced what became known as Social Presence Theory, originally developed while studying telecommunications. Their central idea was deceptively simple: different forms of communication create different degrees of awareness that another human being is present and cares.

A face-to-face conversation is rich in social presence because I can see your body language, hear your tone, watch your facial expressions, monitor eye contact, notice your pauses and immediately adjust my behaviour in response. Nonverbal cues and nuance abound.

The telephone then removed the face but kept the voice. I can still notice your tone, pitch, pacing and inflection, all vocal cues. Then email came and removed all of that. Text messaging stripped communication down further again.

Despite this continuing dilution, we continue trying to put the human back in. We add emojis to our texts, eager to humanise the interaction. We send voice notes, not just because they are faster than texting but to keep in touch, to keep that tonal connection.

All of these behaviours are tiny attempts to restore social presence, adding emotional and human signals back into communication channels that technology has stripped them from.

And now along comes Artificial Intelligence, presenting us with perhaps the most fascinating Social Presence experiment we have ever conducted. For the first time, the thing on the other side of the conversation might not be human at all.

Is Anybody There?

Human beings are profoundly social animals.

For most of our evolutionary history, survival depended on other people. Being accepted into the tribe meant protection, food, shared resources and reproduction, while exclusion would generally mean vulnerability, starvation and probable death.

Our brains therefore became extraordinarily sensitive to social signals. The tone matters, as do the facial expressions, eye contact, pauses, and whether someone responds appropriately to what we have just said.

When you speak with another human being, your brain is processing thousands of tiny signals without you consciously thinking about any of them, constantly asking fundamental questions: Does this person hear me? Are they really with me? Am I truly heard?

That question sits at the heart of Social Presence Theory, and increasingly it now sits at the heart of AI-powered Customer Experience too. Do I feel heard?

Interestingly, our brains appear surprisingly willing to accept social presence from machines. Research into conversational AI has shown that responsiveness, conversational language, personalisation and human-like communication can increase perceptions of social presence, trust and empathy. Give an AI a natural conversational style, allow it to remember context and make its responses relevant to what the customer has actually said, and people begin responding to it socially.

It is why ‘companion AI’ has been so successful. Brands like Character.AI and Replika have 20m and 40m users respectively, facilitating tailored roleplay and relationship companions.

This shouldn’t really surprise us. As humans, we have a tendency to  anthropomorphise everything. We name our cars. We name our Roombas. We thank Alexa. We shout at Google Maps when it sends us down a road we know perfectly well is closed (although admittedly the language used in that particular interaction may not qualify as “social connection”).

The human brain is constantly searching for agency, intention and social cues. And the good news is that AI has suddenly become extraordinarily good at producing them. But there is an important distinction between simulating social presence and creating genuine human connection, particularly when the emotional stakes begin to rise.

Research into service interactions suggests that while customers may be perfectly comfortable interacting with AI for many tasks, when they require genuine social or emotional support, traditionally their preference often shifts towards another human being.

And therein lies one of the most important challenges facing Customer Experience leaders today. The question of course is not “Where can we use AI?” but increasingly, the more valuable focus is becoming “Where should the customer still feel another human being is there?” It doesn’t need to actually be a human. It just needs to really feel like one.

Nobody Wants Brian If The Bot Can Do It

Let’s be fair to AI. If I want to know whether my parcel has shipped, I don’t particularly need Brian from Customer Service to emotionally connect with me. I just want to know where my parcel is.

Similarly, if I want to change my seat on a flight, reset my password, check my bank balance or find out what time your store closes, please don’t make me wait seventeen minutes listening to Vivaldi so I can discuss the matter personally with a human. I just did a mid-term adjustment on my car insurance that took 5 minutes and many ‘press 3’ decision tree moments. It could all have been handled by an AI agent in 30-45 seconds.

This is where AI is magnificent. Technology can remove enormous amounts of low-value friction from customer journeys, providing instant, personalised service at a scale that humans simply could never match.

The mistake is assuming that because AI can successfully handle some customer interactions, it should therefore handle all customer interactions. If I want to make a claim on that same car policy, that will be a far more nuanced conversation and so an AI agent would likely frustrate me.

For example, there is a world of difference between “Where is my parcel?” and “My parcel hasn’t arrived and it contains the dress my daughter needs for her wedding tomorrow.”  From a traditional Customer Service perspective, these might be classified as the same query, but to any human, we can clearly recognise the emotional charge in the second situation.

One customer needs information but the other needs information wrapped in reassurance, urgency and empathy. This is where Social Presence becomes critical, because as the emotional intensity of an interaction increases, our need to feel that somebody is genuinely there increases with it. And this is a significant watchout for Agentic AI in CX deployment.

When Efficiency Becomes Absence

Few companies illustrate this tension better than Klarna.

In 2024, the Swedish fintech company proudly announced that its AI customer assistant was handling around two-thirds of its customer service chats, doing work equivalent to hundreds of full-time agents while dramatically reducing resolution times.

On the efficiency spreadsheet, it all looked magnificent. But in 2025, Klarna began recruiting human customer service workers again, with CEO Sebastian Siemiatkowski acknowledging that an excessive focus on cost had affected service quality. The company did not abandon AI, and indeed continues to use it extensively, but it recalibrated the relationship between technology and people, recognising that customers should still have the ability to reach a human when they needed one.

That distinction is incredibly important as the problem wasn’t Artificial Intelligence. The problem was something I’d call artificial absence.

AI shouldn’t be thought of simply as a replacement for Customer Service employees. It should be thought of as another member of the team, one with extraordinary capabilities but also some significant emotional limitations.

McDonald’s discovered another version of the same problem. The company had been testing AI-powered drive-through ordering with IBM at more than 100 restaurants in the US, an entirely logical use case when you think about it. Drive-through ordering is repetitive, relatively structured and operates at enormous scale, exactly the kind of environment in which automation should thrive.

Unfortunately, customers began sharing some rather entertaining examples of the technology misunderstanding orders, including one system apparently attempting to add hundreds of chicken nuggets to an order while the increasingly desperate customers tried to make it stop. McDonald’s ended that particular trial in 2024, although importantly it remained committed to exploring AI voice ordering.

Again, the lesson isn’t that AI failed. The lesson is that customer-facing AI has to be better than the experience it replaces. Customers don’t care about your AI strategy. What they care about is getting their lunch.

When The Bot Becomes The Problem

Then there is DPD. In early 2024, a customer trying to locate a missing parcel became frustrated when the company’s AI chatbot couldn’t solve his problem, so naturally he started playing with it.

He persuaded the chatbot to swear, criticise DPD and even write a poem about how terrible the company was. The interaction went viral and DPD subsequently disabled the offending part of its chatbot while it updated the system.

The customer’s original frustration wasn’t that he was speaking to AI, it was that the AI couldn’t help him and wouldn’t get out of the way, and that erodes customer satisfaction instantly. Customers are not necessarily anti-technology, but they are certainly anti-frustration.

Frustration appears when technology becomes a barrier between the customer and the solution, particularly when the machine continues pretending to provide social presence while the customer increasingly feels that nobody is actually listening.

The moment a customer starts repeatedly typing HUMAN PLEASE, your digital transformation has probably gone off course.

The Uncanny Valley of Empathy

This brings us to perhaps the most fascinating question of all. Can AI actually deliver genuine empathy? To some extent, the answer appears to be yes, and excitedly, the technology is getting better by the week.

Research into what academics call artificial empathy suggests that AI systems capable of recognising frustration, understanding needs and responding in emotionally appropriate ways can improve customers’ perceptions of an interaction.

That makes sense because, from the customer’s perspective, empathy isn’t only something another person feels. It is something they demonstrate. “I understand why you’re frustrated.” “You shouldn’t have to explain this again.” “I can see you’ve already contacted us twice, so I’m going to take ownership of this now.”

Those sentences mean something to a customer because they acknowledge our emotional reality. AI can absolutely do that. In fact, sometimes AI may do it more consistently than humans.

Anyone who has dealt with a bored Customer Service agent at 4.55pm on a Friday will know that employing a human being doesn’t automatically guarantee humanity or empathy. Quite the opposite sometimes. Call centre talent retention is a major issue in CX.

The danger comes when artificial empathy becomes synthetic theatre, where the machine keeps telling you how terribly sorry it is while simultaneously refusing to solve your problem. There are few things more irritating than a chatbot repeatedly saying, “I completely understand how frustrating this must be,” while demonstrating quite comprehensively that it doesn’t care. Social presence without agency quickly becomes irritating. Empathy without action is really just polite obstruction.

Perhaps We Have Been Asking the Wrong Question

For the last few years, much of the debate around AI and Customer Experience has been framed around the question “Will AI replace humans?”. I’m tired of that question, as I’m sure you are too. We are way past that. The more interesting question is really “Can AI make humans more human?”

One of my favourite uses of Artificial Intelligence is not customer-facing at all. It is technology operating quietly in the background, removing the administrative sludge that stops human beings connecting properly with one another.

Healthcare provides a beautiful example. Anyone who has visited a doctor in recent years knows the experience. You explain what is wrong while the doctor looks sideways at a computer, typing notes, checking boxes and populating your medical record. You are sharing something deeply personal while staring at the side of someone’s head.

Intermountain Health installed an Ambient AI documentation system that is beginning to change this. AI listens to the consultation, creates the clinical documentation and allows the physician to concentrate far more fully on the patient.

Here the technology isn’t reducing social presence but actually restoring it. The doctor can look at you straight in the eyes, listen properly, notice your expression, allow a silence to sit for a moment rather than filling it with the sound of a keyboard. It is emotional, empathetic and intimate.

That, to me, is where AI becomes genuinely exciting. The technology isn’t replacing the human interaction but protecting it. Let the machine take the notes so the doctor can hold the patient’s gaze. Let AI search the database so the employee can listen. Let AI populate the form so the adviser can reassure. Let AI do the process so humans can do the people.

High Tech, High Touch

Bank of America’s virtual assistant Erica offers another interesting model. Rather than pretending every banking interaction requires a human, Erica handles enormous volumes of everyday requests, from transaction searches and account information to personalised financial insights and proactive alerts. That all makes complete sense.

If I want to know how much I spent in restaurants last month, there is no particular emotional benefit in another human being personally delivering that information. But importantly, systems like Erica can also connect customers towards higher-touch service when that becomes necessary, and that should be the model everywhere.

The best AI journey should never feel like a locked room. It should feel like a brilliant concierge. Sometimes the concierge gives you the answer immediately. Sometimes it completes the task for you. Sometimes it anticipates what you need before you’ve even asked. And sometimes it recognises that what you actually need is another human being and takes you directly there.

It doesn’t start you again. It doesn’t ask you to repeat your account number. It doesn’t require you to explain the death of your father for a fourth time because apparently Departments A, B and C don’t talk to each other.

The AI should carry the context while the human delivers the connection. That is the beautiful space where scale and intimacy can finally coexist.

The Emotional Escalation Point

What organisations now need to build into AI systems is something I would call an Emotional Escalation Point. We already route customer interactions based on technical complexity, but perhaps we should start routing them based on emotional complexity too.

AI should become capable, not only of understanding the words customers use, but recognising signals of distress, vulnerability, anger, bereavement, fear, repeated failure or unusual circumstances, and understanding when that moves us from Customer Service into Customer Intimacy.

Think about insurance again. “I need to change the registration number on my policy.” That is all I need to do today. That was perfect for AI, but instead I was on hold for 8 minutes before Declan (who was lovely but took another 5 minutes to simply change a single field of data) completed the amendment. AI would likely have dealt with it in under 2 minutes.

Contrast that to another insurance issue. “My house flooded last night and everything we own is downstairs under two feet of water.” Thankfully not my issue this time but definitely time to speak to a human.

Or take banking. “What is my current account balance?” AI. “My husband died yesterday and I don’t know what happens to our joint accounts.” Not a space for the chatbot.

What about travel?  “What gate is my flight departing from?” AI. “My flight has been cancelled and I’m trying to get home because my father is celebrating his 80^th birthday.” Human.

The technological challenge isn’t simply teaching AI how to answer more questions; it is teaching AI when not to answer them. Knowing when to hand over may eventually become one of the most sophisticated things an AI system can do.

AI Should Remember So Humans Can Connect

There is another enormous opportunity here which I think is often overlooked. AI has memory. Or, more accurately, it has the potential to create organisational memory at a scale we have never had before.

One of the most infuriating aspects of Customer Service is having to repeat yourself. You explain the problem to the chatbot. Then to Customer Service, then to Technical Support and then to Billing.

By the fourth interaction, the customer isn’t simply angry about the original problem anymore. They are angry because the organisation appears to have developed corporate amnesia.

“Can you explain what’s happened from the beginning?”  No. No I can’t. I already did. Three times. If AI does nothing more than eliminate that sentence from Customer Experience, I think I personally welcome our new robot overlords.

Imagine instead that the human agent receives a beautifully concise AI-generated summary before speaking to you.

“Ken, I can see your flight was cancelled yesterday due to the air traffic control issues in London. I see that you’ve already spoken with two of my colleagues and the flight that we rebooked you on does not get you to your event in time. You don’t need to explain any of that again. Let me see what I can do.”

Now AI hasn’t removed humanity. In fact, it has enabled it. The machine remembered and the customer feels heard. Social presence is restored.

Don’t Measure the Wrong Thing

There is, however, a danger waiting for us inside all this efficiency. Businesses love metrics. We love to measure average handling time, cost per contact, call deflection, resolution time and headcount reduction.

AI makes all those dashboards look spectacular. But none of those numbers tell you whether the customer felt cared for. They take no account of social presence.

User Testing have a wonderful new product called ARQ (AI Relationship Quality). Essentially real human consumers testing your AI interface and scoring it across key metrics like trust, control, context and advocacy. This is key to ensure it is hitting the mark on social presence.

Last month, I wrote about my Pain Pyramid, where Customer Service removes the functional pain, Customer Experience reduces the suffering surrounding that problem, while Customer Intimacy creates attachment by demonstrating genuine care during moments of vulnerability.

AI can become extraordinarily effective at the bottom of that pyramid. It can identify the problem, retrieve information, process the refund, rebook the flight and settle the claim. But as we move towards the top of the pyramid, the emotional component becomes increasingly important. AI can restore the cancelled flight, but a human can understand why getting home matters.

AI can process the insurance claim, but a human can recognise the trauma behind it and engage appropriately. The more efficiently AI handles the functional layer of Customer Experience, the more valuable those moments of genuine human connection potentially become.

Humanity At Scale

For years we assumed there was a trade-off between scale and intimacy. You could either deliver personalised, empathetic service to a relatively small number of customers or efficient standardised service to millions. AI destroys that compromise.

For the first time, brands can listen to millions of customers simultaneously, remember individual preferences, understand context, anticipate needs and respond personally at enormous scale.

That is extraordinary, but we need to remember something important. Personalisation isn’t intimacy. Knowing my name isn’t knowing me. Remembering what I bought isn’t caring why I bought it. Predicting what I need isn’t the same as understanding how I feel. And generating a perfectly empathetic sentence isn’t necessarily the same as somebody genuinely caring.

The brands that win this next era of Customer Experience will therefore not be those that deploy the most AI. They will be those that deploy it with the greatest emotional intelligence, understanding which interactions need speed, which need personalisation, which need social presence and which need genuine human presence.

Use AI to remove friction. Use it to create relevance. Use it to remember, predict, personalise, translate, automate and anticipate. Let it work tirelessly in the background doing all the things technology does magnificently.

But when the customer needs reassurance, compassion, creativity, humour, discretion or simply another human being to say, “Leave this with me, I’ll sort it out,” make sure somebody is there.

Social Presence Theory has been telling us the answer for nearly fifty years. Technology, channels, and interfaces change. But the fundamental psychological need underneath them does not.

We still want to feel heard, understood, and that somebody is there. The future of Customer Experience is Artificial Intelligence amplifying Human Intelligence. The machine providing the scale with the human providing the soul.

Customers don’t ultimately build relationships with processes, algorithms or Large Language Models. They build relationships around how interactions make them feel. And regardless of how intelligent our machines become, that fundamental human truth isn’t going anywhere.

Connection is Everything.