Framework
Timing as Product Strategy
The founders who tried to build restaurant voice AI before GPT-3.5 had to train their own models - at great expense, with poor results. Loman launched two months after GPT-3.5 and got the same capability for the cost of an API call. Christian's principle: if an LLM cannot yet do the thing your product requires, waiting is sometimes the right strategic move. The technology timeline is part of your roadmap.
How it works
- 1The cost of being early: pre-LLM voice AI companies spent years and significant capital building custom models that performed poorly - the first movers in the restaurant phone AI space had a worse product at much higher build cost than those who waited for foundation models
- 2The right question: before committing to a technical approach, ask whether a foundation model will solve this problem in 12-24 months; if yes, building a proprietary solution before that happens creates technical debt that becomes a liability when the LLM arrives
- 3Two months of timing advantage: Loman launched approximately two months after GPT-3.5, which is when the latency and quality of voice generation first became viable for a live phone conversation - being early by two months, not two years, captured the market without the cost of being too early
- 4Fine-tuning over building: rather than training models from scratch, Loman fine-tunes foundation models (OpenAI, Anthropic, Gemini, open source) for specific use cases - capturing the capability of large-scale training at a fraction of the cost
- 5Model selection as an ongoing capability: different foundation models perform better for different tasks; testing all of them and matching each to its best use case within a product is an ongoing competitive activity, not a one-time decision
- 6The platform shift thesis: just as waiting for GPT-3.5 was the right call for voice AI, AI builders today should watch for the next capability threshold - agentic reasoning, multimodal understanding, real-time data access - and position their products to land at each new capability floor
Hear it on the podcast
Building a Restaurant AI Startup from Zero to 1,000 Customers in a YearListen to the episodeMore from this episode
The Restaurant Revenue TriangleThree levers, two revenue and one cost: answer 100% of calls instead of 60% (direct revenue recovery), upsell on every single order instead of 5-8% of them (average order value lift), and redirect or eliminate the labor hours absorbed by phone handling. The math stacks across all three lines simultaneously - making this a rare product category that improves a business's P&L in multiple ways at once.Vertical AI GTM PlaybookFor vertical AI companies, the GTM formula is: identify exactly where your ICP lives online (not where other founders live), start SEO from day one, build customer video stories as your primary sales asset, and create word-of-mouth through product quality before building a referral program. The distribution channel must match the buyer's media diet, not the founder's.
Get the next framework first
One email a week with the frameworks, tools, and founder stories from the podcast. Free.
More frameworks for Strategy & Moats
The AI Holding Company ModelBuild a portfolio of AI-native service businesses under shared infrastructure rather than scaling a single product. The holding company model captures diversified upside from multiple category disruptions while sharing GTM, talent, and capital allocation expertise. Each company stays small by design - a team of five is a feature, not a constraint.The 20 Percent AI RuleWhy Filip thinks the model is the least defensible part of the product.Quality-Depth Over Geography-BreadthWhen facing pressure to expand into adjacent markets, choose depth in your current market over thin coverage across many. The quality of your service depends on institutional knowledge, relationships, and operational infrastructure that cannot be replicated overnight in a new geography.The Shelf RiskMatt's framing for what agentic commerce does to a brand.Understand the EngineWhy the moat is not what you built.10-Year Problem DesignBuild solutions for the problem that will exist five to ten years from now, not the problem visible today. Requires outsider perspective, first-principles thinking, and organizational structures that support multi-year investment horizons.