Building PropertyScribe: From Idea to Launch
How we went from a conversation with an estate agent to a live product in 6 weeks, including the decisions we made, the corners we cut, and what we learned along the way.
PropertyScribe launched 6 weeks after the initial idea, and here's what happened in between.
Week 1: Validation
The idea came from Tom, an estate agent in Manchester who complained about writing property descriptions. "I write 15-20 a week," he said, "and each one takes about 20 minutes. It's the same thing over and over."
That works out to 5-7 hours per week on repetitive writing for just one agent. We asked 12 other agents and they all said the same thing, so validation was done and it was time to build.
Week 2-3: MVP
We started with the simplest version where you upload property photos, add some details, click generate, and get a description. The AI would analyse the images and write copy ready for Rightmove, Zoopla, and OnTheMarket.
The tech stack was deliberate:
- Next.js for the frontend because we know it well
- OpenAI API for generation and image analysis
- Vercel for hosting with one-click deploy
We chose boring technology so we could move fast and it worked because we had a working prototype in 8 days.
Week 4: Real Testing
We sent the prototype to 12 agent contacts and they gave us feedback that was immediately clear and specific:
"The descriptions are too formal. Estate agents don't write like that."
That was a fair point, so we added three writing styles:
- Premium for luxury properties with aspirational language
- Modern which is family-focused, warm and inviting
- Concise for quick listings that stick to just the facts
The image analysis turned out to be the surprise hit. Agents upload photos and PropertyScribe identifies features automatically like the kitchen island, the south-facing garden, or the period fireplace, which means no more typing out every detail. The whole generation process takes under 30 seconds.
Week 5: Payments
This is where it got real because we needed to start charging money. The options were:
- Stripe which was familiar but complex
- Paddle which handles VAT but takes higher fees
- Lemon Squeezy which was new but simple
We went with Stripe because yes, handling VAT ourselves is annoying, but we already knew the API and speed mattered more than convenience.
Pricing took three iterations to get right:
- Free with 3 descriptions per month, which is enough to try it
- Starter at £15 per month for 15 descriptions
- Professional at £29 per month for 50 descriptions with advanced image analysis
We added a free tier late in development which was risky, but it lowered the barrier to trying the product.
Week 6: Launch
We built a simple landing page, wrote clear copy without any "revolutionary" nonsense, and went live. The first customer signed up 3 hours after launch and it was Tom, the agent who started this whole thing.
What We Learned
1. Talk to users constantly. Every assumption we made without asking users turned out to be wrong, but every feature we built after talking to users actually worked.
2. Ship before you're ready. We launched without a blog, without testimonials, and without half the features we planned. None of that mattered because what mattered was solving the core problem.
3. Boring technology wins. We didn't use the newest framework or the coolest database, we just used what we knew and it saved us weeks.
4. The free tier was worth it. Agents are skeptical of new tools, so letting them try 3 descriptions for free built trust and most free users convert within a month.
Where We Are Now
PropertyScribe has a 4.8-star rating from 127 users and agents tell us they're writing descriptions 10x faster.
The product handles both sales and lettings, all descriptions are SEO-optimised for UK property portals, and we added a feature to avoid specific phrases ("boasts" is banned by popular demand).
More importantly, it works. Agents save hours every week and they tell other agents about it. Some have cancelled and that's fine because we learned why and made it better.
The next product is already in progress and we're following the same process: find a boring problem, build fast, ship early, charge money, and repeat.