AI in the back office. Humans on the front line. Here's why.
In May 2025, Klarna's CEO Sebastian Siemiatkowski told the press something most fintech founders would never admit out loud. Two years earlier, his company had replaced roughly 700 customer service jobs with an AI agent built on OpenAI. The bot handled millions of conversations. Costs went down. Wait times went down. On paper, it worked.
Then customers started leaving.
"Really investing in the quality of human support is the way of the future for us," Siemiatkowski said. Klarna rehired. The public position now: AI solves the easy stuff. Humans handle the moments that matter. The CEO went further. He predicted that in a few years, talking to a real person will become a VIP feature customers pay a premium for.
We've watched this story closely. We've also watched the same pattern play out across hospitality, e-commerce, banking, and increasingly, residential property management. The technology is real. The temptation is obvious. The cost of getting it wrong is high. And the customer is telling the industry, loudly, what they want.
Here is where we've landed at Nested, and why.
What the data actually says
When Gartner surveyed customer service leaders late last year, 85% said they would explore or pilot customer-facing generative AI in 2025. Salesforce's State of Service report tracked the same trend from the inside: AI moved from the tenth priority on service leaders' lists to the second in a single year. Most companies are expanding their AI surface area as fast as their procurement teams can sign contracts.
Customers are not following the same curve.
A 2025 Zendesk study found that 64% of customers will abandon a brand after a single bad AI interaction. Only 15% of AI-to-human handoffs are described as smooth. Three quarters of customers say it is frustrating to retell their story when a bot finally passes them to a person. A Metrigy survey put the preference plainly: 84.7% of consumers would rather talk to a human agent than an AI one.
The picture is not that customers reject AI. The picture is more useful than that. Customers are perfectly happy to use a chatbot to check a balance, reset a password, or get a delivery update. The same Gartner research that found heavy enterprise adoption also found that on simple, fast tasks, 82% of customers prefer a bot over waiting on hold. The preference depends entirely on the stakes.
The moments where people want a human are the moments that define a relationship: complaints, emergencies, complex decisions, anything emotional. These are also the moments where most companies' AI fails. Three quarters of customers say chatbots struggle with complex issues. EdgeTier, an industry research group, identified the most common failure mode: bots that route a frustrated customer in a loop because the system was never designed to recognise frustration as a signal.
The hospitality version of this story
The 2025 reporting on AI in short-term rentals is worth reading if you own residential property. Roughly 70% of hosts now use some form of AI tooling, mostly for guest messaging. The pitch is sensible. AI replies to enquiries 24/7. Hosts get hours of their day back.
A 404 Media investigation last year captured the other side. A guest noticed something was off about a host's replies. The guest tested the bot by asking for a French toast recipe. The bot, dutifully, returned a recipe for French toast. The screenshots went viral. The story spread under a single headline: hosts don't want to talk to guests anymore.
The damage that does to a hospitality brand is not measurable in a deflection rate. It is the slow erosion of the one thing the category is selling: that you are not a transaction.
We think about this constantly. Nested is in a category where trust is the product. The conversations that define a tenant relationship do not get solved by a script. A delayed lease. A relocation decision. A complaint that needs to be heard properly. A family choosing where they will live for the next two years. These are not flowchart problems. They need a person who knows the building, the neighbourhood, and how to fix what is wrong.
Where AI fits at Nested
None of this means we ignore AI. It means we use it where it earns its place.
Our Customer Success team spends their day on tenant conversations. The work that has to happen around those conversations, sorting and prioritising inbound messages, drafting first-pass replies for agents to edit and send, summarising long email threads before a handoff, categorising maintenance tickets by urgency and trade, scheduling viewings, translating between French and English, used to eat hours that should belong to tenants. We have moved that work to AI. Quietly, behind the scenes. Every output is reviewed by a person before it leaves the building.
Salesforce's research found that agents using an AI copilot save roughly four hours a week on routine work. Our internal numbers are in the same range. That time goes back into the conversations we actually want our agents having.
The result, measured the way we measure it: less than two hours, on average, from a tenant message to a real human reply. Seven days a week. We have not added headcount to hit that number. We have given the people we already have more time to do the work that matters.
What this commits us to
The decision is not technical. It is positional. We are a premium operator in a city full of cheap shortcuts. The shortcut to "we use AI to handle customer service" is available to us. We are not going to take it.
The conversations that define a tenant relationship do not get solved by a script. A delayed lease, a relocation decision, a complaint that needs to be heard properly, these need a person who knows the building, the neighbourhood, and how to fix what is wrong. AI handles the work around those conversations so our team can be present in them.
AI is in the back office. Humans are on the front line. That order will not change.
There is a version of this industry, five years from now, where the only companies that bother to have humans on the line are the ones that decided to compete on care. We have decided. If you are an owner trusting us with your asset, or a tenant trusting us with where you live, you will continue to get a person on the other end of the call. The AI is just there to make sure that person has time to be a good one.
Can you confirm availability before the request arrives, or only after? What are your documented maintenance response times and how are they tracked? Can you provide compliance certificates for every property on request? What does your invoicing look like — one document per placement per month, or multiple vendors?
The answers show quickly whether a provider is built for volume or for individual transactions. They also show whether the agency will spend its capacity managing the housing relationship or delivering the relocation programme.
The agencies best positioned for growth in Brussels have already had this conversation. Most of them changed their housing partner as a result.
Want to discuss what a structured Brussels housing programme looks like in practice? Connect with Nested.
