Travel search was engineered to stay broken

From Flink’s grocery boom to a second consumer bet

Muzamil opens by introducing a guest who has already lived through one full cycle of consumer hype. Saad co-founded Flink, the German rapid grocery delivery company that crossed a billion-dollar valuation, scaled past 10,000 employees and became one of Europe’s fastest-growing unicorns before he exited in 2022. What pulled him into travel was a pattern he had seen twice before: an industry that decades of founders had tried and failed to fix, suddenly meeting a technology that could finally hold its complexity.

“I get super passionate about building consumer products and also those products that really change the way people are using some things,” Saad tells Muzamil. Fintech taught him what digitalisation does to an old industry. Grocery taught him what happens when a behavioural shift collides with mobile apps and e-bikes. Travel, he argues, is next, because for the first time “you actually have a brain that can actually take all of that and process it in a good manner.”

Why travel search was engineered to stay broken

Muzamil pushes on the obvious objection: travel has been disrupted before. Expedia, Booking, Google Travel. The promise was that travel agents would die and consumers would win. Instead, planning a trip in 2025 is more anxiety-inducing than it was a decade ago.

Saad’s answer is the sharpest framing in the episode. The mess is not an accident, it is the business model. “The most profitable part of travel is travel search,” he says. A keyword like “hotels in Barcelona” triggers a bidding auction where each link “are going to be bidding upwards of $30 to $40 to get you.” The thirty open tabs are the product. “Every time you bounce around these different links, these companies or basically Google is going to be making money.”

The shift, in his telling, is not the chat interface itself. It is what happens to that funnel when the top of it collapses. “Now since it has changed the game at the top of the funnel, it’s going to have these ripple down effects downstream.” The incumbents are slow because they “haven’t basically had something of this magnitude of a disruption to face in over since they’ve been existing.”

The content layer, and what China already showed

Muzamil notes that when he tested Layla the night before, what stood out was not the chatbot, it was how short-form video sat inside the planning flow. Saad is direct about why: travel is visual, and the link between Instagram inspiration and a real booking has never existed in the West.

He points to China as the existence proof. “In China, social commerce and travel has is is uh rising in like crazy. So you can see somebody live streaming a hotel and you can book it right there and then using their coupon code.” That worked because Chinese monopolies own both the social surface and the booking inventory. The West never had that vertical integration. What changed, Saad argues, is that COVID forced Western travel companies to digitalise their inventory and expose clean APIs so they could diversify distribution. “Now all of the travel data is accessible, and now it’s more or less just a matching problem.”

The 12-suitcases problem, and why humans stay in the loop

Muzamil tells a long story here that becomes the emotional centre of the episode. Moving from Dubai to the US last year with a three-month-old, a five-year-old and twelve bags, his human travel agent’s Turkey stopover collapsed at the airport. No pickup. Wrong vehicle. Three hours outside the terminal in summer heat. His point is not that humans fail. It is that travel fails in places where a single missed detail cascades.

Saad does not flinch. His view is that current AI is not ready to absorb that risk alone, and pretending otherwise is how AI products lose trust permanently. Layla runs a full travel agency, based in Pakistan, inside the product. “Where before they could serve basically four people a day, now they can be serving around 20 people a day. And they can basically constantly keep on improving the models based on their experiences as well.”

He compares it directly to how the frontier labs got built: “With open AI they had an army in Kenya who was training and and all of the models with all of the data… the big buzz word which was alignment, how do you make these models more aligned. And I think this is the same thing for us in travel.”

Reading the AI bubble through the Flink exit

Muzamil takes the conversation to the bubble question, framed carefully. He believes AI is real. His worry is the narrative whiplash, where the average person rejects the entire technology once the hype flips.

Saad’s framework comes from leaving Flink. He uses two questions to evaluate any consumer wave: how large is the behaviour shift, and how sustainable is it? For grocery during COVID the shift was massive but sustainability was always doubtful, because the moment restrictions lifted the willingness to pay the delivery margin was an open question. That is why he exited.

For AI, he reads both numbers as high. “Search, which is literally… the most profitable company in the world in history Google, has now fundamentally shifted in such a short amount of time.” He cites Anthropic doing roughly $10 billion in the second quarter and adds that “money speaks louder than anything… the only reason why they have had such big revenues is because companies have decided that this technology brings meaningful impact.”

His one real bubble risk is not adoption. It is supply. “The only thing that could really blast the bubble of AI is the the compute and oil prices. So if all of the the billions that are being invested in making these new data centers, if they don’t happen, then for sure whole AI industry is going to face a massive um backlash.”

The timeline argument, and why geography decides

Muzamil pushes back on the timeline. His thesis is that the developed world has an existential incentive to automate labour and intelligence, because it is sitting on a sovereign debt crisis it cannot solve at current cost structures. But the shift is generational. “7 to 10 years are a generation. It’s enough time for people to pivot and come to terms with a new reality.” The bubble narrative gets fuelled, in his view, by people compressing that into one to three years.

Saad mostly agrees and adds geographic texture. Europe moves slower because Europeans are more sceptical. China is aggressive but has passed a law saying you cannot fire an employee just because AI can do their job. India has not put up the capital for frontier models. The US is the outlier. “The US is extremely capitalistic and people are also very capitalistic mindset.” His test is blunt: if AI can build a better marketing campaign than your marketing manager, and there is no regulation, will a US company pay $5,000 a month for the human or a $100-$200 subscription?

He offers a small case study from inside Layla. They have an AI running their entire marketing department. It understands the pain point of an incoming user, generates a video tackling it, produces six format variants for TikTok, Instagram, Google and YouTube, runs the A/B tests, and optimises in real time. “Which was quite shocking for us to see this, that it can handle so much context and in real time react so fast.”

Traction, and why the investor list matters

Layla has raised around $7 million. Five million users have come through the platform. Roughly 30 million messages. Trips generated worth about a billion dollars in intent. The current phase, Saad says, is conversion. He references a family last week that booked a $30,000 safari “where the human had to press just done.”

The investor base is where Muzamil presses, because his concern is the classic one: investors pushing founders to scale before the product is sticky. Saad’s answer is that the cap table is deliberately operator-heavy. The founder of Skyscanner. The founder of Booking.com. United Airlines. First Minute Capital, run by Brent Hoberman of lastminute.com. Bu, “the Google in China.” M13 in San Francisco. “These companies have taken decades to come to where they are. So they are definitely very supportive and very much understanding that scaling the wrong thing is the worst thing you can do because there’s no turning back after that.”

That experience shapes product focus, not just funding. “There are people that would really say you need to get my restaurant right but then they are not going to pay you for recommending the perfect restaurant. Which is very different to you need to get my hotel right, where then you’re going to be making hundreds of dollars just on that recommendation.”

Distribution when search is dying

Muzamil asks the question every consumer AI founder is now living with: if users no longer go to Google, how do they find you? Saad gives a number that lands hard. Twenty percent of Layla’s users arrive having already failed on ChatGPT. They paste in the output and ask for a real planner. “Chat at the end of the day does not get all everything live stitched together… Chat actually does no mathematics, it’s just predicting the next letter. So it’s notoriously bad in mathematics, and travel is all about numbers and budgets and distances.”

The rest of the distribution is social. Travel creators post itineraries with deep links back to Layla, where the trip becomes a customisable template. Layla also lives inside ChatGPT as a connector, built fast using a startup called NoodleSeed. “That allows any business to come and be on ChatGPT within basically a matter of days.”

The 10-year vision: the always-on luxury travel agent

Muzamil closes by asking what travel looks like in ten years. Saad rejects the first generation of travel chatbots, which mostly tried to turn the filter-and-sort interface into a conversation and failed. The real prize is recreating the luxury travel agent. One that knows you well enough to anticipate, not just respond.

Muzamil presses on the real-time problem. If his son decides on day two of a Mexico trip that he wants a water park, can Layla re-plan and re-book in the moment? Saad is honest. The technology is mostly there. The blocker is economic. “To have an agent that’s always on and online is still too costly. The tokens that it costs and so on… but the willingness to pay for this is not high because you don’t really care that much about this pain point.”

The future state he describes is one where the agent sees that day two’s beach forecast has turned to rain and suggests an indoor alternative before the user notices. Where a foreigner travelling Pakistan gets continuously guided through which districts to enter and which to avoid, because the agent has read every relevant blog comment and TripAdvisor review. Muzamil’s reaction lands the vision: a solo traveller walking around with an earpiece and a local fixer who never sleeps.

It is also, notably, a vision that does not require the AI to replace the human agent. It requires it to replace the thirty open tabs.