The Restaurant That Refused to Take Bookings Online
Let me tell you a story about a restaurant owner who became obsessed with human connection.
He didn’t want people booking online. He wanted them to call. He wanted the ritual of a human voice, the small exchange about an anniversary or a first date, the warmth of being recognised. His team thought he was losing his mind. Online bookings were standard. Everyone did it. Why make customers work harder?
One of the managers finally asked him a pointed question: Have you ever actually made a reservation with us?
He hadn’t. So he did.
He called the restaurant. He was put on hold for thirty minutes. When someone finally answered, they were apologetic but firm — the restaurant was fully booked. No warmth. No conversation. Just a long wait and a closed door.
In trying to humanise the process, he’d made it worse.
Here’s where most stories would end with “so they moved everything online and fired the reservation staff.” That’s not what happened. They did move to online booking, but they kept the entire reservation team and repurposed them. These people now spent their days learning about the customers coming in that night. Who was celebrating a birthday? Who was on a first date? What had a regular not finished on their plate six months ago?
The team became mini-concierges. Every guest walked in to find someone who knew them — not in a creepy, surveillance-state way, but in the way a good friend remembers what you’re going through. The technology handled the transactional layer so the humans could focus on the relational layer.
I’m telling you this story because it captures something I think most organisations have forgotten in the rush to automate everything: the work of connection isn’t overhead. It’s the point. And as AI swallows the transactional layer of every industry at once, that relational layer is about to become the only point of difference left.
Trust Is Built One Marble at a Time
Brené Brown tells a story about her daughter Ellen that I realate to. Ellen came home from third grade devastated. She’d shared something vulnerable with a friend at recess, and by the time class started again, half the school knew. She told her mum, through tears, that she would never trust anyone again.
Brown’s first instinct, honestly, was to agree. Tell no one anything, ever. But she caught herself and offered a different idea — the marble jar. There’s a jar in Ellen’s classroom where good choices add marbles and bad ones remove them. When the jar fills up, the class celebrates. Trust, Brown told her, works the same way. You share the hard things with people who have filled up your marble jar over time. Marble by marble. Small moment by small moment.
Ellen immediately named two friends whose jars were full.
What stopped Brown cold was what counted as marbles. One friend had scooted over at the lunch table to make “half a heinie seat” when Ellen had nowhere to sit. Another had remembered the names of Ellen’s grandparents at a soccer game. Brown couldn’t believe it. Surely trust required grander gestures than this.
She went back to her research data and found exactly the opposite. The single highest-ranked trust-building behaviour in her entire dataset? People who attended funerals. Not the people who made big speeches or grand gestures. The people who showed up in small moments that mattered.
Trust isn’t a transaction. You can’t purchase it, automate it, or accelerate it with a clever marketing campaign. It accumulates in tiny, unremarkable moments that look like nothing from the outside — and everything to the person on the receiving end.
And here’s what I find most interesting about the marble jar: once it’s full, it’s incredibly hard to empty. You can weather bad days, miscommunications, even genuine mistakes. But if you’ve never bothered to fill it, a single bad interaction is all it takes. There’s nothing to cushion the fall.
Every organisation has customers whose jars are full and customers whose jars are empty. The ones with full jars forgive outages, laugh off a late delivery, stay through a price increase. The ones with empty jars churn the first time anything goes wrong. Most organisations don’t realise which is which until it’s too late.
Hospitality Is a Dialogue, Service Is a Monologue
Danny Meyer, the restaurateur behind Union Square Hospitality Group and Shake Shack, draws a great distinction. Service is the technical delivery of a product — the food arrives hot, the bill is accurate, the room is clean. Hospitality is how the delivery of that product makes its recipient feel.
Service is a monologue. You decide your standards and you deliver against them. Hospitality is a dialogue. You notice, you adjust, you respond.
Meyer puts it even more sharply: hospitality exists when you believe the other person is on your side. It’s present when something happens for you. It’s absent when something happens to you. Those two little prepositions contain the whole difference. You feel it instantly when it’s there. You feel it even more instantly when it’s not.
Meyer isn’t anti-technology, by the way. But he draws a firm line: the technology he’s interested in is technology that reinforces hospitality, not technology that replaces it. He gives sommeliers Apple Watches for real-time information. He doesn’t give them to waiters, because waiters need to maintain eye contact with guests.
Thats super important. A restaurateur whose restaurants have earned basically every award that exists decided his waiters should not be looking at screens because eye contact is more important than information. When was the last time you saw a decision like that made inside a large organisation?
The best story I know in this vein is from Will Guidara, who ran Eleven Madison Park when it was named the best restaurant in the world. One night, Guidara overheard a table of European food tourists lamenting that they’d eaten at Per Se, Le Bernardin, and Daniel — but they were flying home the next day and had never tried a proper New York City hot dog.
Guidara ran outside, bought a $2 hot dog from a street cart, convinced his chef to plate it with “swooshes of ketchup and a quenelle of relish,” and had it served as a surprise mid-course. Every person at the table said it was the highlight of not just the meal, but their entire trip to New York.
A $2 hot dog in a room full of the most expensive food in America. That’s the gap between service and hospitality. You cannot design an algorithm that eavesdrops on dinner conversation and dispatches someone to buy a street hot dog, because the person on the receiving end would immediately sense the machinery of it. The magic is precisely that a human heard, a human decided, a human cared.
Guidara was so moved by what that moment revealed that he created a dedicated role called the Dreamweaver — someone whose entire job was to make these moments happen. Researching guests ahead of visits, building cheat sheets so staff could greet people by name, listening for conversational cues during dinner. This is exactly what the restaurant in the opening did. The technology handles the boring parts so the humans can do the thing only humans can do.
The Things We Destroy Because We Can’t Measure Them
Here’s where I get frustrated, and this is the part of the article I’ve rewritten three times.
I’ve spent a lot of time in and around banking over the past few years, and I’ve watched a particular kind of decision get made over and over. It always starts the same way. Someone pulls a report showing branch transaction volumes declining. Someone else pulls a report showing the cost per square metre of the branch network. A third person models what happens if you close the bottom quartile. The numbers are clean. The deck practically writes itself. And the decision gets made in a room where nobody has ever stood behind the counter and watched what actually happens on a Tuesday morning at half past ten.
What you see when you actually stand there is not transactions. You see a retired schoolteacher who comes in every week not because she can’t use the app — she can — but because the teller knows her by name and asks about her grandson. You see a small business owner dropping off the weekly takings and mentioning, almost offhandedly, that he’s thinking about expanding into the next suburb, and the branch manager quietly filing that away because she knows a commercial lender who should probably call him. You see the elderly man whose wife just died, standing at the counter trying to change the joint account, and the staff member who recognises what’s actually happening and gently walks him through it for forty-five minutes without once looking at a clock.
None of that shows up in the transaction volume report. None of it shows up in the cost-per-square-metre model. But it’s where the loyalty comes from. It’s the reason the schoolteacher’s kids bank with the same bank and her business-owner neighbour refinanced there and the grieving husband never even considered moving his accounts when the rate went up.
When the branch closes, all of that evaporates. Not dramatically. Just quietly. The schoolteacher switches to the credit union down the road because they still have tellers. The business owner realises nobody at the new centralised lending hub remembers him, and starts taking meetings with competitors. The grieving husband’s children, watching their dad struggle with the phone banking menu, consolidate everything with a different bank when they inherit. Three years later the spreadsheet shows a modest increase in customer churn and nobody connects it to the branch closure, because the causal chain is too long and the data was never collected in the first place.
The broader pattern is consistent everywhere I look. Australian banks have closed something like 2,500 branches since 2017. The justification was always the same: digital adoption, customer preference, efficiency. But when whistleblowers from the Finance Sector Union testified at the Senate inquiry, a darker story emerged. Branch staff had been performance-managed to suppress in-branch activity — pushed to redirect customers to ATMs, pressured to sign them up for digital banking. The data that justified the closures had been partly manufactured by the closures themselves.
And what the banks never properly measured was any of this. Every one of those moments is share of wallet. Every one of them is retention. And none of it appeared on the spreadsheet that closed the branch.
The tragic irony is what came next. The same banks that couldn’t justify keeping branches open are now spending billions on AI personalisation engines designed to replicate the exact relationships they dismantled. JPMorgan Chase budgeted $18 billion for technology spending in 2025. Bank of America’s AI assistant has surpassed two billion client interactions. And yet only about a quarter of consumers say their bank provides tailored financial advice. They destroyed the organic version and are now paying orders of magnitude more trying to rebuild a synthetic one.
There’s a concept I keep coming back to called the McNamara Fallacy, named after Robert McNamara, the US Defense Secretary who tried to run the Vietnam War like a business. McNamara was brilliant with metrics. A general once told him he needed to add an “x-factor” to his measurement list — the feelings of the rural Vietnamese people. McNamara wrote it down, asked what it meant, and then sarcastically told the general he couldn’t measure it, so he erased it.
Daniel Yankelovich described the fallacy as a four-step descent. First, you measure whatever can be easily measured. Second, you disregard what can’t be easily measured. Third, you presume that what can’t be measured easily isn’t important. And finally — this is the fatal step — you presume that what can’t be easily measured doesn’t exist. Yankelovich called this last step “suicide.”
Every organisation I’ve worked with has done this at some level. Dashboards capture the measurable. But the most valuable things — trust, loyalty, the feeling a customer gets when someone remembers their name, the way an employee feels when their manager notices they’re having a hard week — live in the unmeasured space between the data points. And when we optimise away the things we can’t count, we destroy the foundation that made the things we can count actually work.
If you work in data, as most people reading this do. The reports that justified those branch closures were technically accurate. Well-modelled, properly sourced, beautifully visualised to requirements. Someone like us built every one of them.
Wells Fargo is the textbook case. Leadership set a target of eight accounts per customer. “Eight is Great” became the mantra. Over fourteen years, employees created 3.5 million fraudulent accounts to hit the number. They optimised the metric so aggressively they destroyed the thing — trust — that made banking relationships worth anything in the first place. The fake accounts generated about two million dollars in fees. The fallout cost them more than three billion.
Charles Goodhart’s law captures this beautifully: “When a measure becomes a target, it ceases to be a good measure.” The moment you start optimising for NPS, NPS stops telling you anything true about your customer relationships. The moment you start optimising for branch foot traffic, foot traffic stops telling you anything true about community banking. You end up managing the shadow and losing the substance.
Why the Apple Store Is Always Full and the Carrier Store Is Always Empty
Walk into any shopping centre and do a little experiment. Find the Apple Store. It’ll be packed. People learning photography, kids in coding workshops, someone at the Genius Bar getting help with an old laptop, others just hanging out because the store feels like a place you want to be. Then walk to the Telstra store, or Optus, or whichever carrier is nearest. You’ll find two or three bored staff members, maybe one customer, and the unmistakable atmosphere of a waiting room at a tax office.
They sell, to a remarkable degree, the same products. iPhones, accessories, plans. So why does one feel like a community space and the other feel like a place you’re only in because you have to be?
The answer starts with Ron Johnson, who built Apple retail alongside Steve Jobs in 2000. Johnson made a decision most retailers would never make: a store needed to be much more than a place to acquire merchandise. It needed to help people enrich their lives. Any website could transact. A store had to do something a website couldn’t.
He and Jobs asked themselves a question I still think about: what would the Four Seasons do? The Four Seasons doesn’t have cashiers. It has a concierge. So Apple introduced concierge greeters. The Four Seasons has a bar — a place where you can sit, ask questions, and get real help from a knowledgeable person. Apple created the Genius Bar. It dispensed advice instead of alcohol, but the idea was the same: a place to be cared for, not a place to be processed.
Then they did something that, from a commercial perspective, looks insane. They removed commissions. Apple Store employees earn zero percent on sales. No upsell quotas, no commission targets, no pressure scripts. Johnson’s framing was that Apple wanted to reach your heart instead of your wallet — and staff whose paycheque depends on closing a sale cannot be fully on the customer’s side.
Compare this to how a carrier store operates. Employees have sales targets. They’re incentivised to push you toward plans and devices that pay them more. Former carrier staff in Australia and the US tell the same story: focus on customer service and you get coached on pushing harder instead. The misalignment between what the employee needs and what the customer needs sits in the air. Customers feel it as distrust, even if they can’t name it.
Apple Stores generate roughly $5,500 in revenue per square foot — about twice the next-highest retailer on earth. They get over a million customer visits a day. But only about one in a hundred visitors actually buys something. Ninety-nine percent walk in, spend time, and walk out without making a purchase. And that’s the whole point. Apple built the most profitable retail operation in history by investing in the ninety-nine people who aren’t buying today, because every visit is a marble in the jar.
The carrier stores are trying to convert everyone who walks in. They fail. Apple is trying to make everyone who walks in feel good. They succeed, and they also happen to sell more per square foot than any retailer in history. The causation is not subtle.
AI Raises the Floor. Humans Raise the Ceiling.
This isn’t an anti-AI article. I work in data. I use AI every day. I think it’s genuinely transformative and most organisations should be using it more aggressively than they currently are.
But the conversation about AI has got the direction wrong. The question isn’t whether AI will replace human connection. The question is what AI should free us up to do. If the answer is “more automation, more optimisation, more throughput” — we’ve missed the plot entirely.
There’s a concept emerging in the marketing and design literature that I find genuinely useful: AI raises the floor, but it lowers the ceiling. A Fortune 500 study of customer-support agents found that AI boosted the average agent’s productivity by 14%, but boosted the least-skilled agents by 34%. That’s the floor rising. But when researchers tested AI on creative writing, they found something unsettling. Individual AI-assisted stories were rated higher than purely human stories. Read a whole collection of them together, though, and they were rated lower. They all felt the same. The ceiling dropped because the variance dropped.
Here’s what that means for organisations competing on anything other than price. If every company uses the same AI tools to optimise the same metrics, the optimisation itself becomes commodity. Everyone has the same chatbot, the same personalisation engine, the same churn predictor. The floor rises for everyone simultaneously, which means relative advantage disappears. The only remaining differentiator — the actual moat — becomes the irreducibly human: genuine empathy in a crisis, a person who actually cares, a brand that feels run by real people who listen.
John Naisbitt, the futurist who coined “high-tech, high-touch,” made an observation back in 1999 that feels almost prophetic now: the two biggest markets in America are consumer technology and escape from consumer technology. We’re in the same dynamic today, just at a higher frequency. The more AI we get, the more valuable the unmediated human encounter becomes.
I’m seeing this play out in hospitality right now. Mid-range hotels are racing toward automation — kiosk check-in, chatbot concierges, app-based everything. But luxury hotels are moving in the opposite direction. They’re bringing back butlers, personalised greetings, tailored in-person experiences. In an automated world, human service becomes the luxury.
Simon Sinek put it well: AI is a tool, not a replacement. The struggles and imperfections in human interaction are often where the most meaningful growth and strongest bonds happen. Those imperfections are not bugs to be optimised away. They’re the texture of real connection.
Your Employees Are the Moat. The Compounding Is Invisible.
There’s a piece of this conversation that most leaders miss, and it’s the part I care about most. You cannot deliver genuine human connection to customers if you haven’t first delivered it to your employees. The service-profit chain — a Harvard Business School framework that’s been around for decades — is unambiguous: internal service quality drives employee satisfaction, which drives retention, which drives external service value, which drives customer loyalty, which drives revenue. It’s a chain. You cannot cheat the order.
Herb Kelleher ran Southwest Airlines on this principle for decades. Employees come first; treat them right and they treat your customers right, and the customers come back. The things you can’t buy, he said, are dedication, devotion, loyalty — the feeling of participating in a crusade. Competitors can buy the physical stuff. They cannot buy the feeling. The results: number one in US Department of Transportation customer service rankings twenty-six times in thirty-four years, and forty-seven consecutive years of profitability in an industry famous for burning money.
His successor as president, Colleen Barrett, started at Southwest as his legal secretary. She said he treated her as an equal from the day they met, and he treated every employee the same way. That’s not a culture program. That’s a human being who decided to care about other human beings and then built a company around the decision.
This matters for the AI conversation because there’s a temptation, especially under cost pressure, to look at AI and think “great, we can run leaner.” That’s the trap. Use AI to squeeze more out of fewer people and you’ll hit short-term margins while quietly draining the moat. Use it to free people up to do the things only people can do — notice, listen, care, remember — and you’ll build something that compounds for years.
And compounding is exactly the right word: trust compounds, but the compounding is invisible for a long time. James Clear’s ice cube analogy applies. Heat a frozen room from twenty-five degrees to thirty-one (Fahrenheit — stay with me) and nothing appears to happen. Nothing, nothing, nothing — then at thirty-two degrees, everything happens at once. All the action is at the threshold, and all of it is invisible until it isn’t. Most organisations quit before the threshold arrives, because the accounting cycle is quarterly and the compounding cycle is years.
Seth Godin says: “You must build trust before you need it. Building trust right when you want to make a sale is just too late.” You get there drip by drip by drip, until people would miss you if you were gone.
So What Do You Actually Do With This?
Ask yourself which of your customer-facing moments are dialogue and which are monologue. If every interaction is something you’ve decided to deliver to the customer rather than for them, you don’t have hospitality. You have service. Service is fine. Service is table stakes. But it’s not a moat.
Ask yourself what your employees would say if someone asked them whether they were cared for. Not in a survey. In an honest conversation over coffee. That answer is the ceiling on what your customers will ever feel, no matter how much you spend on customer experience.
Ask yourself which of your metrics you’re managing and which ones you’re serving. If your teams are optimising for NPS, you’ve lost NPS as a signal. The moment a metric becomes a target, it stops telling you the truth. You need to look at the behaviour underneath the number — the actual quality of the human moments — and measure those, even if the measurement is messy.
Ask yourself where you’re using technology to replace human moments rather than amplify them. The restaurant in the opening didn’t remove the reservation staff. They redeployed them to do the thing only humans can do. That’s the move. Automate the transaction. Invest the savings in the relationship.
Remember that the dashboards capture the measurable but the valuable things live in the space between the measurements. The best data work I’ve ever seen hasn’t been about building better dashboards. It’s been about giving humans better context so they can have better conversations with other humans. That’s the quiet revolution most data teams are missing. We keep building tools to replace the human layer when we should be building tools to enrich it.
The moat isn’t your data pipeline. It isn’t your AI model. It isn’t your CRM. The moat is what your people do with the insights — the handshake, the eye contact, the moment of genuine care that no algorithm can fake and no competitor can copy. You can’t buy it and you can’t rush it. You can only show up, drip by drip, marble by marble, for longer than everyone else is willing to.
Instagram Engineering
Shopify Engineering