In 12 minutes, without a template, without a developer, and without looking at the result before presenting it, Eliav Cohen built a working slide deck for a class of fifth graders about balloon history. He gave Claude the topic, the audience, and five design constraints, let it ask him a few clarifying questions, and walked out the door with something he could put on a screen. That is the operating standard he applies to every build, not just the fast ones.
Eliav is the founder of Seattle Ballooning, the largest hot air balloon company in the Pacific Northwest. He was also one of the earliest users of ChatGPT and has since built or rebuilt nearly every operational tool in his company using AI, from customer communication workflows to internal data pipelines to a client-facing invoice system he demonstrated live in front of a group of operators. He does not sell consulting. Everything he shares is something he is actively running in his own business.
What follows is a synthesis of what he walked through: the actual tactics, the decision logic for when to build versus buy, and the prompting approaches that separate fast useful output from generic noise.
Ask Questions Before Building Anything
The single most repeated habit across everything Eliav demonstrated was starting a build by having the AI ask him questions, not by telling it what to make.
The structure is simple: describe what you want to create, then instruct the AI to ask you ten questions, one at a time, before writing a single line of code or copy. He used this on the live invoice build. He used it on a documentary film treatment. He used it on a presentation deck. The use case does not matter. The questions are what matter, because the questions surface the things you would have forgotten to specify.
In the invoice demo, one of the questions Claude asked was about payment options. The operator had been working in Google Sheets and had a vague idea of what she wanted, but it was the question that surfaced something specific: she charged a trip-planning fee upfront, and it had never been tracked in her invoicing system. The client had paid it. It had not been credited. That oversight would have shown up in the first invoice she sent if she had skipped the question step and just started building. The questions caught it before anything was written.
This is not a prompting trick. It is a design constraint. The AI cannot build something well-suited to your context without information about your context. Asking for questions first is how you transfer that information in the right sequence.
The Invoice That Went Live in Twenty Minutes
The live build started with nothing. One operator said she had been struggling to create a proper client-facing invoice. Eliav opened Claude, typed the company name and website URL, and asked it to fetch the site before asking anything else.
Claude retrieved the site, identified the brand aesthetics as warm earth tones with a destination magazine feel, and incorporated that into every design decision that followed, without being told to. The operator did not have to describe her brand. The AI read it directly.
The questions covered charge breakdowns, payment options, the ability to download as PDF, and whether to accept Stripe payments. Within fifteen minutes, Claude had generated a complete HTML invoice page with line items, two payment methods, the company name, a destination photo, and proper typography. It was a working document, not a prototype. A client could have received it that day.
The files went to Netlify. Netlify is a hosting platform that accepts HTML files by drag-and-drop and gives you a live URL in about 30 seconds. The entire process, from blank prompt to live public URL, took under twenty minutes.
The operator asked about connecting Stripe. Claude walked through that too, step by step, for someone with no coding background. You do not need to understand what you are doing. You need to know how to ask what to do next, and to take a screenshot when you are stuck.
The Build-or-Buy Rule
Not everything should be built. That is probably the most important point Eliav made, and he returned to it throughout.
The rule he uses: if you can build something functional in under an hour, build it yourself. If it takes longer, buy a product. The categories that never cross the under-an-hour threshold for him are booking systems, OTA integrations, and CRMs. He is direct about why. They connect to payment processors. They integrate with external platforms. They maintain transactional records that clients rely on. When something breaks, you need a vendor you can call and hold accountable. You cannot call yourself.
He switched booking platforms not because he built something better, but because he found one that did one specific thing better than his old one: real-time two-way text messaging with clients. His team had been managing guest communication across multiple phones, with notes in Slack, with handoffs that got lost. He told both platforms exactly what he needed. Whoever built it first would get his business. The new one did. He is a close enough relationship with that provider that when problems come up, his feedback shapes the product roadmap.
That is the relationship you want with the systems your business depends on. It is not available to you when you build the system yourself.
The right candidates for solo AI-builds are the ones that would otherwise cost $2,000 to $5,000 a year from a software vendor, take you an hour a week of manual work, and could be replaced by something 90% as functional that you build in 45 minutes. Landing pages, event registration forms, invoice systems, internal scripts that automate repetitive reporting. Eliav rebuilt several of the tools he had previously paid for, eliminated the subscriptions, and got something better tuned to how his business actually works.
Your Data Lake
AI can generate outputs against almost any input. What it cannot do is operate intelligently over your business if your business data is fragmented across inboxes, booking confirmations in one system, client notes in another, and pricing history in a spreadsheet you update manually.
Eliav calls the central data repository a data lake. For him it is Notion. For another operator it might be Google Sheets or Excel. The format is less important than the principle: one place, connected to whatever AI you are running, updated consistently.
When his team switched booking platforms, they uploaded the entire booking history to Claude. It migrated the data and missed nothing. Now when he has a question about a client, he asks. When he wanted to know at what point the prior year July 17th had been fully booked out, Claude found it. That date feeds his dynamic pricing logic. He did not have to build a dynamic pricing tool. He needed one data point, asked for it, got it in seconds, and made a decision.
The data lake is not a project. It is a habit. You put business information in one place, keep it accurate, and keep it connected to the AI you use most. Everything else becomes easier after that.
What the Chatbot Actually Does
Eliav has been building chatbots since 2017. He had a Drift-based system for years, building out keyword trees and handling every unusual question manually. He threw it all out when the current generation of AI made it obsolete and rebuilt in about 20 minutes on a platform called Aniro.
Aniro pulls the information from your website directly and auto-generates accurate responses without hallucinating, because it is not generating answers from training data. It is looking up the answer from what is on your site. The accuracy is bounded by the accuracy of your site content.
He demonstrated it live by asking a set of increasingly unusual questions. Can I bring my blind dog? The chatbot acknowledged the dog, explained the policy, offered information about companions who could stay on the ground, and asked a follow-up question. My grandmother is a pirate with a peg leg, it is our anniversary, can she get in the basket? The chatbot addressed both parts of the question, handled the physical concern correctly, acknowledged the anniversary, and offered to help with logistics.
He has not looked at the chatbot in months because he does not need to. It handles the inbound question volume without any intervention and it is always accurate, because his site is accurate.
The one thing it does not do yet is complete a booking. It can hand off to the booking calendar with a direct link, but the actual payment still happens there. He noted that the integration is coming. For now, it eliminates the staff time spent answering questions that are already answered on the site.
How to Prompt Like It Owes You Money
Eliav’s prompting style is aggressive by design. He tells Claude that his friend who is an expert says the output is not the best it can do. He tells image AI that if it changes the geometry again it owes him money. He instructs it to make something a million times more authentic.
This is not about getting a reaction from a machine. Pushing back with specificity and firmness consistently produces better output than polite requests. The model responds to the same pressure that sharpens any piece of writing: directed criticism with a clear bar.
He also asks AI how the recipient will take an email before he sends it. Not the polished version, not a revised draft. He writes what he wants to say, however bluntly, then asks: how will this person take it? The response tells him whether the message accomplishes what he needs. Then he asks for a version with empathy added, and uses that.
For copy that needs to sound like a person wrote it, he starts by asking the AI to behave as a literary copy editor and critique what it just produced, then revise based on that critique. He gets near-final copy in one or two passes that way, because the revision starts from a diagnosis rather than a vague instruction to improve.
Before a meeting with a documentary filmmaker, he downloaded all the filmmaker’s podcasts, loaded them into Claude, and asked it to simulate that person’s perspective on a project he was developing. It asked him 65 questions. When he walked into the meeting he had already worked through most of the angles and had a clearer point of view. The filmmaker said the output was surprisingly sharp, and then told him the thing the AI had not arrived at, which was the right human anchor for the whole story. The AI improved the meeting. The filmmaker determined the direction.
That boundary matters. AI gets you to the meeting better prepared. The expert across the table still does the work that requires judgment.
What This Means for You
The ones getting the most out of AI right now use it daily, who take screenshots when stuck and ask what to do next, who push back on outputs that are not good enough, and who have stopped waiting for someone else to build them a tool.
A branded invoice page in twenty minutes. A slide deck in twelve. A chatbot that handles inbound questions without you. None of this required a developer, a software budget, or deep technical knowledge. It required knowing how to ask questions.
The question to ask before any build is not whether you can build it. It is whether building it saves time, eliminates cost, or gets you off something you should not be paying for. If yes: build it. If no: do not. That logic applies every time, and applying it is a faster decision than most people think.
About Eliav Cohen
Eliav Cohen is the founder and lead pilot of Seattle Ballooning, the largest hot air balloon company in the Pacific Northwest. He built and operated high-growth sales organizations before moving into aviation, and has spent the last several years applying AI directly to the operations, marketing, and customer communication systems of his own business. He does not sell consulting. He competes in the World Hot Air Balloon Championship and was part of an early summit at Richard Branson’s island focused on the future of AI and its impact on humanity.
Eliav Cohen shared these strategies and demonstrated these tools in a live Tourpreneur PRO Expert Session. Tourpreneur PRO members get access to the full recording, the cleaned transcript, and all 15 takeaways. Learn more about Tourpreneur PRO.