Golf: The Perfect Proving Ground for Experience AI
Why we chose golf travel as our first vertical, and what it teaches us about building AI for complex experiences.
"You could have picked something easier"
We hear this a lot. Golf travel, as a first market, seems counterintuitive. It's niche. It's seasonal. It's full of quirky traditions and unwritten rules. Why not start with something more straightforward — city tours, food experiences, adventure activities?
The honest answer is that we tried. And we kept running into the same problem: the simpler the experience, the less operators actually needed better software. A walking tour company with three set routes and online booking doesn't have a quote problem. They have a marketing problem.
But the operators struggling most — the ones losing deals because they can't respond fast enough, burning out from manual coordination, unable to scale without hiring more staff — were consistently the ones handling complex, high-value, bespoke experiences.
Golf travel sits at the extreme end of that spectrum. Which makes it the perfect place to build.
What makes golf travel genuinely hard
Let's walk through a real scenario. A group of 12 American golfers wants to do a Scotland trip in August. They want to play the Old Course at St Andrews (obviously), plus "the other famous ones." Budget is "whatever it costs for the best experience." They have 7 days.
Here's what the operator has to figure out:
The Old Course problem. You can't just book the Old Course. Visitor tee times are allocated through a ballot system that closes 48 hours before play. You can improve your odds through a tour operator allocation, but those are limited and relationship-dependent. Do you promise the Old Course and risk disappointing? Do you set expectations but potentially lose the booking to a competitor who overpromises?
The routing puzzle. Scotland's best courses aren't conveniently clustered. Carnoustie is an hour from St Andrews. Turnberry is three hours from Edinburgh. Royal Dornoch is practically in the Highlands. How do you sequence 7 days to maximise great golf while minimising time in minibuses?
The seasonality math. August is peak season. Prices are highest, availability is tightest, but the weather (statistically) is best. Do you suggest shoulder season dates at lower cost? How do you price a trip when green fees vary by 3x depending on the month?
The "other famous ones" interpretation. Which courses count as famous? To a golf purist, Prestwick and North Berwick are musts. To someone who's seen too many bucket-list articles, it's all about Turnberry and Kingsbarns. Getting this wrong means either an underwhelmed client or an awkward conversation mid-trip.
The group dynamics. 12 golfers rarely have identical handicaps or stamina. Do you suggest courses that will challenge the 5-handicapper without demoralising the 25? Should you build in a rest day, or do they want golf every day?
A single enquiry contains all of this complexity. And the operator has maybe 24-48 hours to respond before the client moves on.
Why this complexity is actually a feature
Here's the counterintuitive insight: the complexity that makes golf travel hard to automate is exactly why it's worth automating.
If the product were simple, operators would compete purely on price and marketing. The margins would be thin. The switching costs would be low. Software wouldn't provide much advantage because there's not much to optimise.
But because golf travel is complex, operators who handle complexity well can charge premiums. Clients pay for expertise, for relationships, for the confidence that their trip will actually work. The question is whether that expertise can be delivered efficiently.
Most operators handle maybe 50-100 trips per year because that's all their manual processes can sustain. With better tools, the same operator could handle 200, 300, or more — not by cutting corners, but by spending time on expertise rather than admin.
What we're learning
Six months into building Fareway, the lessons are humbling. A few stand out:
Natural language is harder than it looks. "A Scotland trip in summer" seems unambiguous. But does summer mean June, July, August? Does "Scotland" include northern England (some clients want to add Northumberland courses)? Does a Scotland trip imply St Andrews, or should we ask? Every interpretation choice compounds.
Operators know things they can't articulate. Ask an experienced operator why they'd suggest Cruden Bay over Murcar, and they'll give you an answer. But that answer often doesn't capture the full reasoning — the intuition built over hundreds of trips, the subtle read on what this particular client actually wants. Encoding expertise into software means constantly discovering hidden assumptions.
Pricing is context-dependent in ways that break traditional models. The "same" trip can vary by 50% in cost depending on when it's booked, how it's packaged, and what relationships the operator has. Most software assumes pricing is a lookup table. In reality, it's a negotiation.
Speed matters more than perfection. Early versions of Fareway tried to produce "perfect" quotes. But operators told us they'd rather have an 80% quote in 15 minutes that they can refine than a 95% quote that takes an hour. The workflow is iterative, not linear.
From golf to everywhere
The ultimate goal isn't to build the best golf software. It's to build infrastructure that works for any complex experience business.
But we're believers in focus. Golf teaches us how to handle scarcity (tee times), complexity (multi-stop itineraries), high stakes (expensive trips), and expertise (operator knowledge). If we can solve those problems well, the patterns transfer.
Wine country tours have the same scarcity issues with top wineries. Safari operators face similar routing optimisation challenges. Luxury concierges deal with comparable client interpretation problems.
Golf is the proving ground. What we're building is much bigger.