I’ve long been a fan of HotelTonight. In fact, Sarah and I were there at its launch, test driving the service in a fun dash through San Francisco. Since then, I’ve booked more than a dozen stays through the company — and the front end experience has always been a delight.

At noon each day, in any one of nearly 40 US cities (Canada, Europe coming soon), HotelTonight displays a list of available accommodation, at rates usually far below rack. Two clicks later and you’re booked.

Unfortunately, that’s where the trouble all too often seems to start.

Given how publicly bullish I’ve been on the company (and I remain a huge fan), it seems only right and proper that I’m equally public when things don’t go so well. And last week, when I travelled to NY for PandoMonthly, things went pretty damn far from well.

I was already a little nervous about using HotelTonight for a last minute booking. The previous week I tried to check in to Hotel Abri in San Francisco, only to be told that my HT booking had been moved to a different hotel down the street. The excuse given — “we had a water leak” — is classic hotel code for “we gave your room to a more valuable guest.”

But one swallow does not make a sucky service so, on landing at JFK, I clicked twice and was booked into Hotel Indigo. What happened next played out on my Twitter stream…

Ah yes, the old “pipe leak” excuse, this time at 1am. Next up: a walk across the street to a crappy Marriott that had absolutely no record of my transferred booking. Instead, they tried to sell me a one-night stay for $300 (I’d already paid $600 to HT for my three night stay at Hotel Indigo).

Meantime, Starwood jumped in with a textbook piece of social media marketing…

Finally, the next morning…

Actually, that last Tweet deserves a footnote. Yes, HotelTonight had booked me into the Ace, agreeing to pay the difference in rates, but in a last amusing twist, the company’s credit card was declined at check in. It took an extra twenty minutes before reception would actually let me into my room.

There are a few morals to the story. The first, of course, is that it doesn’t matter how slick your app is, disrupting an industry as complicated as the hospitality business involves building a huge number of real-world relationships. If your app takes customers’ money up front, then it’s your problem if any one of those relationships breaks down.

The second moral is that, while hotel owners will gladly agree to anything that might sell their last minute inventory, they’ll also throw your fancy app company under the bus if a full-paying customer walks through the door. That’s almost certainly what happened with my bookings in San Francisco and New York. In fact, it definitely happened in New York: hilariously, I discovered the next morning that the full-paying bookings that walked in were members of the PandoDaily team. Ho ho ho.

To deal with the above, HotelTonight has to do two things. For one thing, it has to keep growing. Once the service becomes a force to be reckoned with — like Orbitz or Expedia — hotels will think twice before bumping guests, knowing that doing so will surely lead to fewer bookings in future.

In the meantime, the second thing they need to do is act like they’re already a giant. Most customers will forgive a single dropped booking, but two “walks” and the third time they’re simply not going to take the risk. HotelTonight needs to have a zero tolerance policy with the hotels it works with. One strike and they’re out.

In the short term, that will reduce the number of hotels the company has to offer its customers, but better that than the alternative: a reduction in the number of customers it has to offer its hotels.