Can AI-First Support Deliver - or Is Human Still Essential?

Everywhere online, you see the same headlines: AI is taking over the world. Every new model is supposedly smarter than the last (well, maybe not that much smarter, despite the marketing hype - but still, progress is undeniable). AI is framed either as a magic pill that will cure humanity’s problems or a curse that will doom us all (I’m still not sure if it’s the blue pill or the red pill). And yes, in the coming years, AI will likely replace a lot of professions. It will also push science forward in ways we can’t yet imagine.

That’s hard to argue against. Personally, I’d call myself an AI optimist. I believe the future will probably be better than what we have today - different, yes, maybe not what we expect, but more positive than not. Of course, I also fully trust humanity’s unique ability to ruin everything it touches… but let’s not dwell on that for now.

Still, even as an optimist, I have to admit: the speed at which my own job (I am a Customer Success gal) could be replaced by AI feels much faster than the arrival of some shiny utopia with universal basic income. That’s why I wanted to reflect on my expectations about AI in the customer communication space and share them with you. And I’d love to hear your thoughts, too.

I do love Intercom - their community, their blog, and of course the product itself. I use it, I learn from it, and I genuinely appreciate their perspective. So when I came across one of their latest posts, this statement caught my eye:

“AI Agents are reshaping how businesses deliver service, earn loyalty, and create measurable value.Very soon, we believe AI will have the potential to handle all customer service. And eventually, every customer interaction. Humans will be involved, but they will move to primarily work on analyzing and optimizing the AI Agent system.Customer experience is the goal. Customer service is the first frontier. It’s where this transformation is happening fastest, and where the impact of AI is already clear.”

Sounds bold, right?

But here’s the thing: I can’t quite agree with it. In fact, it left me a little disappointed. Especially because Intercom positions itself as a tool that promises “a better support experience for customers, agents, and leaders” and aims to “deliver faster support that customers love.”

And that raises the big question: can people actually love AI support?

That’s what I want to dig into here - with you.

Customer-Facing AI Agents Today

Before diving into the pros and cons, let’s map out what kinds of AI we actually see in customer support right now. Broadly, they can be classified by how much data and context they have access to.

Humans you may meet in Support Chats

Of course, it’s not all AI out there. On the other side of the chat window, you’ll often meet humans - and they also come in different “types,” depending on their role, training, and how much context they have.

Other customer-facing roles - like community moderators, account managers, technical consultants, escalation managers, and trainers - play important but occasional roles in support. While they guide discussions, maintain satisfaction, resolve crises, or onboard customers, most day-to-day support is handled by fewer people wearing multiple hats. For this article, we’ll acknowledge them but keep the focus on core support interactions.

What are our support expectations?

Customer expectations

When you ask customers directly, their answer is usually simple:

• Speed: Yep, faster is always better. Nobody enjoys waiting.

  

• Correct answer: Speed doesn’t matter if the answer is wrong or useless. A quick “bullshit” answer will frustrate the customer more than a slightly slower but accurate one.

  That’s basically it - at least on paper. But we all know that there’s a layer beneath the surface. Customers may not always admit it, but they often want something more subtle.

• Empathy: This one’s tricky. It’s hard to define, even harder to measure, and deeply tied to human social needs. We’ll dive into this later, because it’s one of the areas where AI and humans behave very differently.

  

💅 I don’t think AI would be able to add such wonderful memes to an article, to be honest. I hope that a sense of humor will be irreplaceable longer than hard skills.

There are also few points I thought of, but they are rather about long-term relations, so let’s keep them in this chapter, but a bit “separated”:

• Convenience: Easy access to support and self-service options are increasingly important. Customers appreciate when they can resolve issues quickly and efficiently, whether through direct interaction or self-help resources. And they hate when support button is extremely well-hidden…

  

• Transparency: Clear and honest communication about issues, delays, or policies builds trust. Customers appreciate it when companies are upfront about potential problems and set realistic expectations. But not all customers, especially if the answer doesn’t match what they were hoping for. 🙂

  

Ok, we reached our meme limit for this article, thanks for the attention. 🙌

Business expectations

Ah yes, the classic company goal: spend as little as possible to get as much value as possible. The less, the more — in a nutshell. But let’s break it down a bit more.

1. Save money This one’s obvious. But how much a company is willing to spend depends not only on its budget, but also on industry standards. Realism is often forced by competitors, customers, and yes — accountants.
2. Metrics Metrics are the obsession of every business. The golden rule seems to be: if you can’t measure it, you’ve already failed. But here’s the nuance — many companies don’t really know what they’re measuring, why, or how. I’ve talked to enough data analysts crying over creating convoluted stats just to satisfy some “vision” rather than common sense. Still, businesses expect support to have fancy, impressive KPIs: SLA, North Star metrics, LTV, FRT, resolution rates, and the list goes on.
3. Happy customers A vague, almost philosophical goal. “Make customers happy” — easy, right? Just buy them a bottle of wine… but I doubt that’s the intended solution. In reality, happiness in customer support is measurable by a set of practical criteria: fast replies, accurate solutions, smooth processes. It’s the kind of happiness that comes from competence and reliability rather than charm or gifts.

AI vs. Human: let’s compare

Let’s compare the two sides across five key dimensions:

• reply time
• correctness
• empathy
• money
• personalization

Reply time

No contest here. AI wins instantly. Even the slowest model will outperform any human in terms of speed. There’s no need to debate this one.

Answer precision

Now things get interesting.

AI is hard to beat here. It can store and process far more knowledge than any single human, potentially providing more accurate answers — in theory. But there are some blind spots:

• Data quality and coverage: AI learns from written material - documentation, wikis, past tickets. If the company hasn’t invested in comprehensive, well-structured guides and company documentation, the AI will only be as good as the resources it has. Many conversations still live in a few human brains, and those nuances can get lost.

Interestingly, integrating AI into my work has made me pay far more attention to how I write and structure documentation. I now consider a wider variety of cases, explain things clearly, and aim to make knowledge reusable, not just for humans, but for AI to learn and adapt as well. Yet, this level of documentation is still not a top priority across most companies, so it remains to be seen how this challenge will be solved industry-wide.

• Customer communication gaps: AI tends to trust people by default. It assumes customers know exactly what they’re asking. In reality, customers come with diverse backgrounds, unique understanding of problems, and their own words to describe issues. These words may not match documentation or prior examples, especially in complex products.

Another challenge: AI prefers to answer rather than question. If something is unclear, it might give a partially correct answer instead of probing deeper. Humans, by contrast, can more easily doubt, clarify, and dig into the root cause. For now, that’s a key weakness in customer-facing AI.

Also! Even with advanced AI, unique and important cases must be reviewed by real humans. Some situations are truly one-of-a-kind, or complex enough that a general AI approach just won’t cut it. That’s why access to human support needs to be quick, seamless, and pleasant. Customers should never feel trapped by automation - there must always be a clear path to a knowledgeable human who can handle exceptions and tricky cases. Fast. With no endless conversation with AI. Otherwise you will be hated. Maybe forever.

Empathy

Sorry, AI - you can write a symphony, craft the most heartfelt message, but it’s unlikely that people will truly believe you care.

A soulless machine pretending isn’t the same as genuine empathy. Sure, some humans in support sometimes pretend to care, but the key difference is that empathy comes naturally for those who are wired for it.

I believe that people who aren’t ready to empathize with others shouldn’t take human-facing roles. Empathy is the biggest game-changer in the AI battle. People still want to be cared for - they were raised that way. The perceived quality of support depends not only on solving the problem, but also on the human connection: how important the issue feels at the moment, the customer’s mood, and the relationship with the support agent.

Will AI create truly memorable experiences for customers? That’s tough to say. What is certain, however, is that human Customer Success Managers leave lasting impressions through genuine care and empathy. Positive interactions with humans - when customers feel understood, supported, and valued - tend to stick. These experiences don’t just solve problems; they build loyalty, increase retention, and enhance brand reputation. In short, humans have the unique ability to make customers feel something meaningful, something AI alone can’t yet replicate.

Oh, and don’t forget about advocacy! Client advocacy is a cornerstone of organizational success. Studies show that word-of-mouth marketing is often the most effective type of promotion, and the formula is simple: happy customers who speak positively about your brand drive outstanding outcomes. These positive experiences, however, rely heavily on strong human interactions. It’s the personal touch, empathy, and care from real support teams that transform satisfied customers into passionate brand advocates.

That said, I’m a realist. AI will evolve, and over time, human approaches may change. Twenty years from now, empathy might matter less; quality of answers could become the dominant factor, and human support might be seen as quaint or “old-fashioned.” Whether that generation will emerge - comfortable with AI and less reliant on human touch - is uncertain.

I still believe there will always be value in human touch. Those who focus on human-centered support may continue to win in certain contexts. Real-life experience supports this idea.

Money

Ah, yes, the classic battlefield where AI flexes its muscles again. AI agents are cheaper than human customer success professionals, and that gap is only likely to widen over time. Humans won’t suddenly start getting paid less (I hope…), but their cost relative to AI will grow as technology improves and production costs drop. If the goal is saving money, AI looks like the obvious winner.

But - and it’s a big but - balance is everything. Implementing AI successfully isn’t as simple as flipping a switch. You need to understand what AI can do, how it works, and how to train it effectively. Reality doesn’t always match the executive vision, and enterprise implementations fail far more often than you might think.

Executives often see a shiny AI tool and want igt immediately, but have little understanding of the foundational work required:

• Data cleanliness: How good is your company’s data?
• Resource allocation: How much time and effort are you willing to dedicate to building, maintaining, and refining AI?
• Feedback loops: Without real-time monitoring and adjustments, errors accumulate, especially in precise tasks.

MIT research shows the core issue isn’t the AI itself, but the “learning gap” between tools and organizations. While generic tools like ChatGPT excel for individuals due to flexibility, enterprise applications often stall because the AI can’t adapt to workflows automatically. Surprisingly, more than half of generative AI budgets are spent on sales and marketing, while the highest ROI often comes from back-office automation - streamlining operations, reducing external costs, and cutting business process outsourcing.

The hard truth? Saving money with AI often requires spending a lot upfront - to build, test, and understand. Misalignment, poor preparation, and missing feedback systems can lead to bigger losses than the savings initially promised. So yes, money is an advantage for AI, but only if the implementation is done carefully.

Here I wanted to thank this great article for inspiration.

Personalization

Once again, AI sits on the edge here - and it’s a strong position. AI can theoretically learn everything about a customer’s context. It can access secure customer data, track interactions over time, and remember details far better than any human could. It can also analyze patterns and draw conclusions that even the most attentive Customer Success professional could only dream of.

But there are still challenges. First, all of this depends on proper documentation and data quality. If the information isn’t recorded, AI can’t learn from it. Second, humans still bring something unique: the ability to build relationships, read subtle cues, and use gut instinct. Through conversation, a human can understand how a person thinks, anticipate priorities, and tailor support in ways that combine data with intuition.

Nah, technically, AI still wins. 🥲

OK, Polina. And what do you suggest?

I don’t even pretend to say that I know what to do. But when I (or anyone else in our team) is doing something, I carefully listen. So if customers are still happy - well, that’s a safe direction. If they are angry - and our customers are pretty loyal ones! - well, we fucked up. If there is lots of dopamine, more than before - hooray, the path is correct! (short-term at least).

At the moment, in Fibery, we have started offering two types of support:

• AI support: AI trained on educational data + aggregated customer feedback + the particular customer’s context - a pretty powerful system. You can ask any question. We keep improving it, so it just keeps getting better.
  • If you are curious, check out our User Guide Assistant.
• Human support: Any customer can explicitly choose to contact a human instead. No limits, only preferences.
  • Check out our latest detailed post on our approach to Customer Support.

Many customers enjoy AI, and yes - it works well. But interestingly, our overall number of chats hasn’t changed, nor has the distribution between new requests and those from paid customers.

Out of curiosity, I analyzed every customer interaction with AI. I fed their questions into the AI and evaluated the quality of the responses. Roughly 70% of the time, the answers were satisfactory. Yet, many customers still chose the slower, less sophisticated human agents.

This raises questions worth investigating:

• Is the AI button less discoverable?
• Do customers lack trust in AI?
• Does our strong support reputation influence their choice?
• Is it the perceived quality of human interactions? Or something else entirely?

Let’s check one more thing - proactive dopamine-unfused feedback.

To understand what brings the most joy to our users, we analyzed feedback highlights. In our scoring system:

• Negative scores indicate positive feedback
• While positive scores reflect pain points

Looking at the top praises, we see that users value flexibility, frequent updates, team connectivity, and the ability to use Fibery for everything. Features like AI - while appreciated and intensively used - appear much lower on the list, suggesting that users consider it a nice (or even must!)-to-have functionality rather than a core source of satisfaction. In other words, AI is seen as a “cool” feature, but when it comes to the experiences that generate genuine user delight and dopamine, human-centered updates, transparency, and versatile workflows take the crown.

For now, the numbers reassure me about my future in support - but also spark curiosity. Understanding why customers choose humans over AI could reveal a lot about the enduring value of empathy. And I would be glad to conduct such research in ~5 years, when AI would become an industry standard.

If you are interested in how we interact with customers, lucky you - here is a nice and very detailed article about it!

AI vs Human Support: quick take

AI dominates speed, cost, and technical personalization; humans shine in empathy and handling nuance.

Customers are becoming far less forgiving. 57% of consumers say they would switch to a competitor after just one bad customer experience - and this number has grown 5 percentage points year over year.

At the same time, AI is now seen as a standard part of modern customer service. 70% of customers expect CX teams to incorporate AI into their support workflows. And unlike in the past, bots are expected to perform flawlessly: 53% of consumers believe that in a few years they will prefer interacting with AI-powered agents because of the reduced risk of human error.*

The takeaway? Customers expect companies to deliver high-level AI implementation, but they don’t promise patience if mistakes happen. In other words, errors have become more costly than ever.

Some fresh metrics from the industry

This article was written, but then we dived into discussions about how to better structure it, yada yada. Before publishing, I found some interesting metrics- though I’m not ready to seamlessly add them to the article, they’re still worth mentioning. 😅

The Generative AI industry is maturing, and early excitement is meeting reality. Biweekly surveys from the U.S. Census Bureau covering 1.2 million companies show that AI adoption is slowing in larger firms (250+ employees). MIT research confirms this: even though 40% of companies have deployed chatbots like ChatGPT or Copilot, only 5% report tangible business benefits.

Enterprise AI often stalls in pilots due to poor context retention, limited learning, and integration challenges. Medium-sized companies can move successful pilots into production in ~3 months, large ones in 9+ months. Meanwhile, employees increasingly rely on personal AI subscriptions, a “shadow AI usage” that often boosts productivity more than official deployments.

The takeaway: AI hype remains high, but effective deployment requires expertise, robust engineering, and realistic expectations. The slowdown signals market adjustment, not collapse.

Here you may read the full article, I definitely recommend it.

Fun fact: Intercom’s confidence in AI

If you think AI can’t be trusted, take a look at how confident Intercom is. Their Fin Million Dollar Guarantee is bold - even eyebrow-raising.

Here’s the gist:

• Fin is Intercom’s top-performing AI Agent, consistently beating competitors in head-to-head evaluations.
• If you sign up for Fin and aren’t 100% satisfied in your first 90 days - or if it doesn’t meet a minimum resolution rate - you can get up to $1 million back, no questions asked.

How it works:

• Try Fin risk-free in at least 250 paid conversations in your first 90 days.
• If you’re unhappy or it doesn’t hit a resolution rate of 65% (roughly the human benchmark), Intercom will pay up.

The idea? In the near future, every customer touchpoint - marketing, sales, service, success - could be handled by expert AI agents 24/7. Intercom wants you to bet on Fin because they’re confident it’s the best AI agent on the market.

My take: I was really curious to try it. But that’s it. Not ready to risk even for a million-dollar opportunity.

But I still want humans in the mix because I don’t want to lose my job 😉 because giving people a choice in a world flooded with AI seems fair. Even the best AI shouldn’t replace the human touch entirely - at least not yet.

*numbers are taken from the Zendesk Customer Experience Trends Report 2024