The Founder's Decision Framework

Seven questions, asked at every stage, with a different right answer each time.

Three case studies in this series looked at three companies solving three problems on three clouds. Underneath all of it, the same seven questions did the work of deciding what to build. Making them explicit means they can be asked directly instead of re-derived at the next stage, the next company, or the next architectural decision this series didn't cover.

What changes between stages

The questions stay the same; what counts as a good answer changes. A company optimizing for learning should give a different answer to “what should we buy as a managed service” than a company optimizing for global regulatory compliance, even though the question is identical. Five stages recur throughout this series:

Stage	Optimizing for	Illustrated by
Pre-seed	Learning whether the problem is real	Casewell: an AI legal-research assistant for small law firms, built to move fast and learn before its narrow runway ran out
Seed	Product-market fit
Series A	Repeatable growth	Sentrio: an AI-assisted cybersecurity analysis platform for SOC teams, built to prove its model repeats across enterprise deals
Series B/C	Scalable execution	No dedicated case study in this series; the bridge between proving a model repeats and operating at Virelane's scale
Late stage	Reliability, governance, cost, security, and global scale	Virelane AI: a global enterprise AI platform for regulated financial, healthcare, and government customers

1. What problem are we solving now?

The current bottleneck standing between where the company is and its next milestone, which is a narrower thing than the company's mission.

Stage	The current problem
Pre-seed	Nobody has confirmed the problem is real yet. The only thing that answers that is talking to the people who supposedly have it. Casewell's founders spent this stage in conversations with law firms, using a rough prototype to make those conversations concrete.
Seed	Early users exist; usage and retention are the open question, ahead of feature breadth.
Series A	Sentrio's problem was proving that the deal that closed last quarter wasn't a fluke, that the sales motion and the product both repeat.
Series B/C	Scaling a proven model without the organization or the architecture fragmenting under its own growth.
Late stage	Virelane's problem is operating at a scale where a single failure reaches a regulator before it reaches a support queue, while holding its latency commitments and its gross margin, both of which are now measured by people outside the company.

2. What milestone must the company reach next?

Stage	The next milestone
Pre-seed	A working prototype and candid reactions from a handful of prospective users.
Seed	Evidence of product-market fit strong enough to raise a Series A on.
Series A	Sentrio needed a growth engine that produced the same result quarter after quarter, with the reasons it worked understood well enough to repeat on purpose.
Series B/C	Efficient scaling: growing revenue without growing headcount and cost at the same rate.
Late stage	Virelane's next milestone is external: an IPO or a further growth round, either of which puts its financial and operational controls in front of scrutiny no customer procurement process ever applied.

3. What failure would be fatal?

Stage	The fatal failure mode
Pre-seed	Building a polished solution to a problem nobody has.
Seed	Mistaking founder enthusiasm and interested conversations for product-market fit, and scaling before it exists.
Series A	A growth engine that worked once by luck, and can't be repeated on purpose.
Series B/C	Losing architectural and organizational coherence while headcount outpaces the systems meant to support it.
Late stage	For Virelane, a serious breach or a failed audit at a top account, the kind of event that can cost the largest customers and draw scrutiny from more than one regulator at once. Slower and equally fatal: infrastructure cost growing in step with revenue, which removes the margin story an IPO depends on.

4. What complexity can we avoid?

Stage	Complexity worth avoiding
Pre-seed	Almost everything beyond the prototype. Casewell had no reason to build multi-region infrastructure it had no customers to justify.
Seed	Premature scale: building for load that hasn't been asked for yet.
Series A	Operating its own Kubernetes cluster. A managed container platform could already express everything Sentrio's workloads needed from a scheduler.
Series B/C	Building bespoke platform tooling before enough teams exist to make it worth the investment.
Late stage	Almost nothing. Nearly every piece of complexity in Virelane's architecture traces to a requirement someone can point at, which is what makes the remaining kind so recognizable: complexity that maps to no requirement at all. A fourth cloud. A bespoke database where a managed one already fits.

5. What should we buy as a managed service?

Stage	What gets bought instead of built
Pre-seed	Everything that isn't the product itself: auth, hosting, the database, the model API.
Seed	Still nearly everything; the product is the only thing worth custom-building.
Series A	Sentrio bought SSO, audit logging, and its compliance tooling rather than building any of it in-house, because none of it differentiates the product.
Series B/C	Managed databases, managed Kubernetes, managed observability; the operational load of self-hosting any of it stops being worth the control it buys.
Late stage	Virelane buys Kubernetes control planes, key management, multi-region databases, and its SIEM, every one of them a solved problem a specialist provider runs better than an internal team would.

6. What is strategically differentiating enough to build ourselves?

Stage	What's worth building in-house
Pre-seed	The core product experience, the thing being tested for demand.
Seed	The specific workflow and retrieval logic that makes the product useful, not the infrastructure underneath it.
Series A	Sentrio built its own detection and analysis models, the part of the product customers were paying for, and bought everything around them.
Series B/C	An internal developer platform, once enough product teams exist that a shared paved road saves more time than it costs to build.
Late stage	Virelane builds its own fine-tuned models and routing layer for its most sensitive workloads, and its own compliance and audit tooling tuned to its specific regulatory footprint. Both differentiate; the managed infrastructure underneath them doesn't.

7. What would force us to revisit this architecture?

Stage	The trigger that forces a revisit
Pre-seed	Finding out the problem isn't real. The architecture rarely survives a pivot, and shouldn't be built as if it needs to.
Seed	Signs of product-market fit, the point at which avoiding scale stops being the safe choice.
Series A	Sentrio's trigger was enterprise demand outpacing what a single-region, lightly governed deployment could support.
Series B/C	Growth outpacing what the current platform team and tooling can support without every team building its own version of the same thing.
Late stage	For Virelane, it's a new jurisdiction requiring data residency the current region footprint doesn't cover, or a regulatory change. The trigger is an external requirement, not a technology preference.

The framework produces one right question at a time. It also gives a way of noticing when the last answer has quietly stopped being true. Casewell, Sentrio, and Virelane arrived at three different architectures by asking these same seven questions at three different points on the table above. Decision quality wasn't the variable; stage was.

The other half of this framework

This series has covered the technology side of that table: what to build, buy, or leave alone at each stage. A parallel set of decisions governs the capital side of the same company: how much to raise, from whom, and on what terms, at each of these same five stages. The founder's funding checklist is that series' capstone, and the two are a pair. Capital decisions without architectural discipline burn the runway that capital bought, and architectural decisions made without knowing the company's stage produce the over-engineering this framework exists to prevent.

Previous ← Virelane AI