Buying an abandoned AI project and growing it to $300,000 in gross bookings in 90 days

Over the last three months, an AI startup called Roam Around has done $300,000 in gross bookings and become one of the world's largest AI travel planners. A company which didn't even exist at the beginning of the year has now produced over 6 million travel itineraries for its customers.

To me, this first sounded like business on easy mode. But I was wrong.

When I reached out to Roam Around's CEO Shie Gabbai he told me the company didn't just get off to a rocky start. The situation had been existential. He spoke of 21-hour workdays. $35,000 in personal credit card debt. Unmanageable API bills from OpenAI, which powers the software. At one point, he said, he'd even found himself three days away from shutting the company down and walking away financially ruined.

But if that's how it started, it's certainly not how it's going. Here's the story of how Shie steered Roam Around through multiple crises and began building a vision for the future of travel.

The problem that Roam Around solves

Channing Allen: Hi Shie! Fill me in on your career before Roam Around.

Shie Gabbai: Hey! I’ve been in tech my entire career, mostly in customer experience. Prior to Roam Around, I managed a team of 50 support agents in the Bitcoin space, and before that I was at Google for 6 years, supporting some of the G’s largest advertisers.

CA: And what is Roam Around?

SG: Roam Around is an AI travel agent. At the moment we leverage OpenAI to create bespoke itineraries for millions of people. But in the not-too-distant future, we’ll automate any task of a dedicated personal travel agent.

Imagine your connecting flight is delayed and you’re in the air with no signal. You land and receive a text from Roam Around saying, “You missed your connecting flight, but I’ve already rebooked you on an alternate flight, leaving from JetBlue gate 34. I’ve already emailed all the hotels in the area, and negotiated a great room at the airport Hilton, located at address…”

That’s the future we’re working towards. Just as chauffeurs were reserved for the elite not so long ago until Uber made them commonplace, we’re going to put a travel agent in every person’s pocket.

But for now, people seem to love what we’re doing. In the 90 days since we’ve launched, we’ve generated 6 million itineraries and $300k in gross bookings, completely organically.

How Shie discovered and acquired Roam Around

CA: So how'd you go from running customer support teams in the crypto industry to running Roam Around?

SG: In my previous role, I was working to create a ChatGPT focused on customer service. But I was running into AI hallucination issues in which ChatGPT would make up answers to customer questions and, with full confidence, just shoot off nonsense responses and provide completely invented URLs.

This was disheartening until I came across Roam Around. The V1 of the code was originally written by Nader Dabit. Nader didn’t want to maintain the site, and I was looking for my next project. I like to say that, just as travelers meet at hostels, Nader and I connected at the internet’s hostel, Twitter.

CA: Wow! How did that transfer process work? Did he sell the codebase to you or something? I assume there must have been monetary terms of some kind.

SG: Nader posted something pretty innocuous like "i built this site, it's costing me a lot of money and I no longer want to support it. i'll take it offline unless someone wants to buy it."

A bunch of people DMed him and asked him really tedious questions. But I came in and said "show me the traffic, and tell me where to send the $."

Ultimately he went with me because I was the easiest.

It was pretty seamless. I paid him 50% upfront via crypto, and that was to secure it. Then he transferred over the GitHub repo and domain name and signed an IP sale agreement, and then I sent over the remaining 50%. Whole thing took maybe two hours.

I didn't expect him to support me or the project the moment he clicked "transfer." I saw Roam Around as something pretty different than the original vision, and so I didn't expect much in return.

That being said, he was very generous with his time during the first weekend when I took over. I was still getting used to the codebase and things would constantly break. So despite him having a lot on his plate, he was generous and decided to help me out. And I really appreciate that. Months later, as a token of goodwill, I awarded him advisor shares.

CA: That's amazing. Hard to imagine a transaction happening that way without Twitter. But what was it about Roam Around that inspired you to part with that kind of cash, no question asked?

SG: Roam Around really hit close to home for me because I’m an avid traveler (9 countries in 2023 so far), but I’m a terrible planner. I’m not the type to do hours of research before visiting a city, which often leaves me scrambling to find the best things to do upon arrival.

I also saw the potential Roam Around had as a business. During a recent visit to Barcelona (after I'd started running the company), I realized that with a few slight tweaks, we could make these itineraries actionable. Once we made those changes, we saw a meteoric spike in traffic, and at one point we generated one million itineraries per day for three consecutive days.

The issue was that we weren’t monetizing. At the time, we were using Text-Davinci-003, and between OpenAI and Vercel costs, I had racked up $35,000 in credit card debt within a short 30 days. I was three days away from taking down Roam Around and walking away (basically) financially ruined when OpenAI announced 3.5-turbo, which was 10x cheaper.

We lowered our Vercel bill 90% through Edge functions (instead of Serverless), and added a monetization layer.

I still think back to how incredible the timing of all of that was. We were three days away from death, but the universe refused to let that happen.

In a world ever more divided, travel is the antidote that connects people of different cultures and makes the world feel a bit smaller. Anything that enhances, promotes, or accelerates this connection is good. Roam Around, by way of saving people dozens of hours planning a great trip, is bringing people together by reducing the barrier to travel.

Tech stack and product development

CA: How was the initial product built? And what's the development process been like since then?

SG: The current version of Roam Around was built via an army of contract developers found on Upwork and located around the world. (Our main devs are in Pakistan, Ukraine, France, USA.)

Personally, for the entire first month of building, I was working 18–21 hours each day while juggling part-time work as a CRM consultant.

I remember how refreshed I felt after my first five-hour sleep in weeks. I’d wake up at 7am to catch up with the devs in Pakistan and India as they were finishing their shifts, then I’d meet with the devs in Europe. In the afternoon, I’d shift to partnership and fundraising meetings with people based in the US, and by nighttime, Asia would come online, and I’d work with them until 4am. Going to bed, my mind would race about the millions of things I wanted to accomplish the next day. Then I’d get a restless three hours of sleep and begin the cycle all over again. It was incredibly exhilarating and exhausting at the same time, which drove me to take a leap of faith to tackle Roam Around full-time.

The features we build today are primarily dictated by customer feedback and my own personal dogfooding while traveling abroad. I read every single feedback email (itineraries@roamaround.io) and try to contextualize why people are asking these questions, which informs our product roadmap.

Financially, our original funding was my personal American Express account. As I mentioned, I maxed out every personal credit card I had in an effort to bring this dream to fruition. Once we had gained and sustained traction, we received a $100k pre-seed check from Jason Calacanis’ Launch accelerator. That kept the lights on (in a very literal sense) and also bolstered our credibility with other partners and investors.

CA: What's your tech stack?

SG: Our main tech stack consists of OpenAI’s LLM (GPT 3.5 and GPT 4), various point of interest APIs, NextJS, Vercel, and Firebase.

Originally, we used Serverless functions, but transitioned to Edge functions due the significantly higher cost of Serverless.

Our main struggle has always been balancing quality with speed. Obviously quality is important, but each itinerary initially took 45 seconds to generate, and we were seeing really high bounce rates. This remains a big focus area, and we’ve worked hard to figure out workarounds that allow us to provide high-quality content at acceptable speeds.

Marketing and user growth

CA: What's the story behind Roam Around's first users and initial traction?

SG: Our initial traction came as a result of going viral on Indian WhatsApp groups. Though we’ve tried, we were unable to ever trace back exactly how that happened. Our best guess is that it was a combination of AI hype plus people realizing how useful this tool could be.

Interestingly, we believe people were excited that Roam Around put their home town on the map, so to speak. Conde Nast Traveler has already covered itineraries for Manhattan, Rome, Paris, and Madrid ad nauseam. But Roam Around was able to create a detailed itinerary for Pavagada, India and Clarksburg, California — cities that would otherwise never be covered by major newspapers. So at first, people were really only using Roam Around to test us. “There’s no way this AI could know my hometown,” they thought. Wrong! Yes, we know about the bridge where locals hang out, and yes we know which cafe is best and which is a tourist trap. And so once we passed these tests, we began getting into the hands of people actually looking to book trips.

Travel is also an inherently social activity. Sure, people sometimes travel solo, but for the most part you’re traveling with friends, family, or business. And so we’re seeing that a large percentage of our generated itineraries are also shared via WhatsApp or email. This obviously fueled our growth.

Aside from that, we worked really hard to get into the hands of as many end-users, publications, influencers, and bloggers as possible:

• We replied to numerous Reddit posts about itineraries for City X with a custom-built Roam Around itinerary for that city.
• We landed a few interviews, and even a couple VC pitches, through DMing hundreds of people on Twitter.
• We studied some of the great growth hackers of our day and tried to learn from their tips on building sites that are shareable and have social hooks.
• And we definitely rode the wave of AI threadbois sharing us in their daily “13 AI tools that will blow your mind.”

From the very beginning we focused on SEO. We realize that there’s an opportunity to specialize in itineraries for the long-tail of cities (think: 5-day itinerary for Pavagada, India) and that helped us grow organically.

Business model and revenue

CA: What's your business model?

SG: We’re in the business of saving people time on their travel plans. We do that via a two-way affiliate model. Local businesses pay us to connect travelers with their activities, and we compensate influencers for any activities booked as a result of their marketing.

On our end, we recommend activities and hotels that link out to Viator and Kayak, respectively. We’re adding in more sources of activities so that we can better serve our users, especially those in Asia and LatAm. We get paid anywhere between 8-10% of each activity bookings, with the average package ranging around $250. In the 90 days since we began monetizing, we’ve generated $300,000 in gross bookings.

Our unit economics are pretty strong and only getting better. 50k OpenAI requests costs us about $15, and given our median booking price, it means our break-even is 0.5 bookings per 50k requests. We’re actually seeing about 20x that.

CA: Have you learned any interesting lessons about revenue growth?

SG: I’d say that founders should demand a seat at the table.

With Viator, I was generating a ton of revenue for them and I was communicating via their normal support methods until I decided to reach out to the C-suite and make a case for being assigned an Account Manager. I don’t know why I had the gall to make that request, but I’m really happy I did, as that relationship has been really fruitful for both parties.

Overcoming adversity

CA: Roam Around has only been a thing for a few months, but it's faced numerous existential crises. Of all the challenges you've faced, which ones have been the biggest?

SG: We’ve faced many challenges. From tech to UX to fundraising.

On the tech side, wide use of LLMs is a pretty new phenomenon which makes sourcing good talent a somewhat difficult task. (If this is you, please email me at shie@roamaround.io!) And even when you’ve found good engineers, the tech itself is constantly changing.

From a consumer-behavior perspective, users aren’t quite ready to fully interact with AI. If you ask our users what they want, they’ll tell you “more dropdown menus!” But dropdown menus are a symptom of the past, of a time where you had to query a database in a structured way. With the paradigm shift brought by natural language processing, our one chat box actually contains infinite dropdown menus. But conveying that to the end-user not familiar with ChatGPT has proven to be quite a hurdle. Users are comfortable generating an itinerary, but they're much less comfortable imagining all the different ways they customize the itinerary to their needs.

Despite our viral growth and substantial success, fundraising has been difficult. After the Web3 boom and bust of the last two years, VCs are wary of jumping head first into AI. We’re constantly facing questions about moat and defensibility, which I think will be the case for any startup building on top of any of the major LLMs.

Looking forward, the main roadblock we’re facing is the nascency of the AI industry. Every week another seemingly mind-blowing and world-changing AI announcement is made and it’s hard to not want to drop everything and build on that. It can be challenging to continue on our path and stay laser-focused on making a great experience and obscuring the complexity of the tech fueling the product.

CA: Knowing what you know today, would you do anything differently if you could go back in time?

SG: I’d do everything exactly the same, but I’d take the leap of faith even earlier.

At the beginning I had a hard time believing this was real, and I was constantly afraid it would one day just instantly die. So even though I was spending 20 hours a day working on Roam Around, subconsciously I didn’t commit until we received the pre-seed check from Launch. Only then did I allow myself to think “OK, this is somewhat real.”

Helpful approaches and resources Shie has taken advantage of

CA: Setting aside things outside your control, what's the main thing you'd attribute your success to?

SG: The main thing we attribute our success to is obsession. Your team’s first idea is typically not going to be its best, and so the quicker you can get to your 100th idea, the better off you’ll be. If this is a part-time gig, you’ll probably get around to the 100th just as you’re running out of runway on your 5th pivot. If you’re absolutely obsessed with your product, you’ll get to the 100th idea by lunch time.

We are three individuals willing to give 110% every day to make this shared vision into a reality.

None of the founders are travel-industry veterans. At first we were a bit insecure about this, but now we see it as an advantage because it means we approach everything without ego and are completely receptive to feedback. And part of the upside of being self-funded is that we are free to experiment, take risks, and make decisions in pursuit of the best user experience possible.

On top of that, utilize Twitter more than you think, and rely on LinkedIn less than you think. Some of the greatest minds of our generation are on Twitter, and inexplicably their DMs are open! Write them a detailed message with a specific ask, and more often than not, we’ve found that they're open to helping out. Jay Hack was instrumental in helping us visualize our roadmap after we struck up a conversation via Twitter.

Three different travel unicorn founders took calls with us because we were persistent in their DMs. Heck, I initially came across Roam Around on Twitter, and it's where I met our CTO Hassan.

CA: One thing I like about your story is that it's clear you've had to work extremely hard, but it's also clear you haven't had to do everything on your own. Often, we hear about people's success without hearing anything about the support systems that were instrumental in bringing that success about.

SG: Aside from my wife who supported me and put up with living with a zombie for months, I was fortunate enough to receive guidance from many generous people throughout this process.

Anushka Jain has been an incredibly wise resource, helping me contextualize much of the consumer behavior and the competitive landscape within India. Sarah Dines is the Chief Commercial Officer at Viator and obviously has many big things on her plate, but she was always willing to take time to help me navigate this new space.

But above all, it was our passionate customers who helped guide every action we’ve taken since day one. It turns out that people have very strong opinions about the itineraries we generate about their hometowns, and they aren’t shy in sharing these opinions with us. We appreciate it more than you can imagine.

Words of advice

CA: Do you have any advice for indie hackers who are just starting out?

SG: This isn’t going to be easy. Even if everything falls into place perfectly, you’re always one or two steps away from complete failure. Prepare yourself for many highs and even more lows. However, if you’re focused on something that you’re passionate about, the work can be incredibly rewarding.

Early on, I came across a post that was extremely helpful to setting my own mindset. It went something like, “If you know that you’re 100 'NOs' away from success, you won’t feel disappointed when you inevitably face rejection.” You’re going to face rejection. I can’t stress that enough. So just look at it as another NO on your journey to YES, and you’ll find the strength to power through.

Most of all, understand that ultimately everything falls on you. You’re now the CEO, CTO, CMO, HR department, lawyer, secretary, customer support agent, social media manager, end user, and yes, even the janitor.

CA: Awesome. Where can readers go to learn more about you?

SG: Any of the following:

• Website: RoamAround.io
• Instagram: instagram.com/roamaround_io
• TikTok: tiktok.com/@roamaround_io
• Twitter: twitter.com/roamaround_io