The messy middle is where the real learning happens — the part where you are not 100% sure things will work, you are building the product while talking to customers, and you are figuring out whether founders will even pay for this.

The Story

It was mid-September. After two weeks of intensive prototyping, I had validated that building an agentic AI co-founder was technically possible. But three massive unknowns remained: Do founders actually want this? Will they pay for it? And can we build enough of it to launch in 2-3 months?

My secret weapon was the Concierge MVP. Instead of building the product first, I became the product. I offered free business model reviews where founders could submit their Lean Canvas and receive a personalized diagnostic. But the real value was not filling my calendar with reviews — it was twofold. First, using each review to learn from customers. Second, incrementally replacing myself with LEANSpark by running each canvas through the prototype first, then tweaking the product to match how I would actually coach a founder.

I used a “give before you get” approach. Instead of emailing founders a diagnostic report, I offered to walk them through it on a call. That was the give. Then at the end, I asked if I could interview them separately — the get. This generated some of the most revealing conversations about how founders were actually validating their ideas.

When I had enough signal from 10 problem discovery interviews, I kicked off the broader Demo-Sell-Build campaign in October. Using my 10x Product Launch playbook, I scaled the hockey stick systematically: 10 customers first, then 100, then aim for 1,000.

For the landing page, I needed a demo — but the product was not built yet. This is where founders get confused. You do not need a working product to build a real-looking demo. I used Claude Code to build a demo runner that could simulate conversations between a founder and LEANSpark. The demos on the landing page were screencasts of these simulated exchanges. Everything shown was not working code at the time, but I was 80-90% confident we could build it.

The campaign ran in stages. Stage 1 launched October 9th with just 10 spots — handpicked from our most active customers. It filled in 6 days. Stage 2 sold out by October 30th. By the time I had stacked an email campaign, a Black Friday bundle, and a referral program in November and December, we had crossed 209 customers and roughly $35,000 in revenue — in about 40 days instead of the 90 I had planned.

The whole time, every Tier 2 customer who paid for a canvas assessment was actually getting a disguised problem discovery interview. The Concierge MVP was not just a sales tactic — it was a continuous learning engine.

Key Frameworks

  • The Concierge MVP: You become the product. Deliver value through high-touch services while simultaneously learning from customers and incrementally building the real product to replace yourself.
  • The 10x Product Launch: Scale the hockey stick instead of letting it play you. Go from 10 to 100 to 1,000 customers, with each level surfacing your riskiest assumption. At 10 customers, you maximize learning. At 100, you start automating. At 1,000, you go self-serve.
  • Offer Stacking: Incrementally transition from high-touch offers (concierge reviews) to more scalable offers (email campaigns with landing pages) as you gain confidence in your positioning and product.

Key Takeaways

  • You do not need a working product to build a convincing demo — you need to design the product, and a demo runner can simulate real interactions.
  • When a customer buys a demo, the demo becomes the best marketing requirements doc: all you need to deliver is what the customer bought.
  • The Concierge MVP is the most underrated prospecting recipe — it simultaneously validates demand, generates problem discovery interviews, and iteratively derisks your product.
  • Offer stacking lets you transition from learning-focused high-touch campaigns to scalable campaigns without losing the customer insights pipeline.
  • Selling in 10x stages (10, then 100, then 1,000) prevents you from drowning in scale problems before you have validated the fundamentals.