Adam, and Why Our First AI Feature Is a Reservation Agent

Softinn shipped Adam.

Adam is a reservation agent, and it is the first AI-powered feature Softinn has ever put in front of real hotels.

I want to write down why we did it this way, mostly so I can come back and read it in a year and see how wrong or right I was.

How it Works

We are testing the market, not declaring victory

The honest framing first: Adam is out there because we want to learn, not because we think we have figured out AI in hospitality.

There is a lot of noise right now. Every hotel tech vendor has an AI page. Most of it is a chatbot with a new label. I did not want us to add to that pile with a feature we built because the market expected us to have one.

So we treated the launch as an experiment with two questions attached:

1. Will guests actually use it? Not "will they try it once," but will they reach for it when they have a real question at 11pm.
2. Will hoteliers trust it? Because a tool the front desk quietly works around is a dead tool, no matter how good the demo looks.

We are watching the usage data, and we are talking to hotels. Both matter, and they don't always agree.

Usage tells you what happened. Feedback tells you why. Early on, with a feature this new, I trust the conversations slightly more than the dashboard — the sample is small, and a single hotelier explaining what confused their receptionist is worth more than a percentage point of engagement.

We will let it run, watch it fail in ways we didn't predict, and adjust.

Why a reservation agent, of all things

This was the part we debated the most. When you get one shot at a first AI feature, the choice of where to put it matters more than the model behind it.

We picked the reservation agent because it sits on both sides of the business at once.

There is a guest-facing front. The guest asks about availability, rates, what's included, whether they can check in early. These are the questions that arrive at all hours and rarely get answered fast enough.

And there is a hotel-facing front. Every one of those conversations touches inventory, rate plans, policies, and eventually the PMS. It is not a standalone chat widget bolted onto a website. It has to be correct about things the hotel cares about being correct on.

Most candidate features only give you one of those. A guest-facing FAQ bot teaches you nothing about operations. An internal ops assistant teaches you nothing about guest behaviour. The reservation agent forces us to learn both at the same time, from a single feature, with a single group of pilot hotels.

That is expensive in engineering terms and cheap in learning terms. For a first attempt, I'll take that trade.

There's a second reason, less strategic and more practical: reservations are where the money is. If AI can move anything for an independent hotel, it should show up in bookings. If it can't, we would rather find out early than after we've built five smaller features around the edges.

On top of the above, we learned from the existing chatbots in the market, and we believe adding the AI element to a chatbot amplifies its usage. Prior to this, chatbots were mostly configured using templates and canned answers. With AI, we believe it will bring the guest service to life, and it's easier for the hotel to set it up.

Two weeks to build it, two weeks to make it safe

The prototype took two weeks. We presented the prototype to the team, and they were impressed.

That number is not a brag. It is the point. Getting something that works in a demo is genuinely fast now. Anyone reading this who runs a product team already knows that. The prototype was working, convincing, and completely unfit to be near a customer's data.

Then it took another two weeks to integrate it into our ecosystem and test its security.

That second fortnight was the real work: connecting it properly to our booking engine and PMS, deciding what it is allowed to see and what it is not, making sure it cannot be talked into doing something it shouldn't, and confirming that a bad response degrades into a handover instead of a wrong answer with confident wording.

The ratio is worth sitting with. Half the calendar went to building the thing, half went to making it responsible enough to ship. I suspect that ratio only gets worse — in a good way — as we do more of this.

If there's one lesson I'd pass to another founder starting their first AI feature, it's this: budget for the second half. The demo is not the product. The gap between them is where trust is either earned or lost, and in hospitality software, trust is the whole business. A hotel is handing us their inventory and their guest relationships. We don't get to be casual about that.

What happens next

Adam is early. It will get things wrong. Some hotels will love it and some will turn it off, and both outcomes are useful to us right now.

What I'm most curious about is not whether the AI is good. It's whether it changes the shape of the work at the front desk — whether it removes the repetitive questions and gives the team back the time to do the parts of hospitality that no software should be trying to automate.

We'll see. I'll write again when we know more.