An engineering analogy for seed investing

This series of blog posts will break down my (current) framework for investing in early stage SaaS companies. My intention writing this down is for my framework to be 1) falsifiable and improvable and 2) hopefully useful - both for founders I partner with and for investors and others who disagree (please do!)

This series is focused on known markets and SaaS only. Truly new category creation I plan to write about separately.

Idea #1: Seed = Search/Optimization Function under a hard budget

I think of early investing as backing a systematic search rather than a fixed plan. Each test costs time and money; you only get so many iterations to find an attractive optima before money (and team energy) runs out.

My mental model:

Setup / Foundation:

• The search landscape: You're trying combinations of who it's for (ICP), what you sell (wedge), how you sell (GTM), and how you charge (pricing metric). There are attractive maxima in some markets. Other markets are inherently limited in size/scope. Others still may be structurally biased towards fragmentation.

• Evaluation cost: Runway = number of experiments a company can afford to run. Every "ship + measure" cycle costs time and money. A sufficiently attractive search landscape might merit multiple parallel searches, but most teams can only afford sequential exploration. N = runway / cost per eval.

Search Execution / Strategy:

• Learning from experiments: Learning is rarely direct. Companies get noisy, delayed proxy signals (acquisition → activation → retention → expansion → payback). These are lagged signals, so at pre-seed/seed, metrics like time to value and weekly active users/teams operate as surrogates.

• Explore first, then focus: Iterate too fast and broad = thrash, move too slow = stall (burn runway before signal). Successful companies often follow an annealing-like trajectory: high learning rate early on (explore segments, pricing, channels), lower learning rate post-signal as signal concentrates. Explore then exploit.

• Escape local maxima: Don't camp on a pretty but capped hill. Schedule restarts before the fuel light comes on. Failure modes include dying on said pretty hill.

Pitfalls:

• Don't worship early revenue: Early revenue is almost always too small to matter and is often noise. What's an extra $1M or $2M when the goal is to get to a few hundred $M+ in a few years? Prioritize trajectory. Prioritize time to initial core value, activation, early retention shape by ICP and use case.

• Parallelization: Great in theory, often fails in practice. Bigger batches confound attribution, increase operational overhead, and ultimately lead to sub-optimal point evaluation. Focus matters.

• Don't overfit: a big logo or edge-case user might lead you in the wrong direction.

What this means for investing:

• Identify founders working on conceptual search landscapes with potential for attractive maxima that:

  • Have a unique learned/earned insight on why either:

    • A particular point in said space is worth evaluating

    • A particular space is worth evaluating and likely has an optima

  • Can take a sufficient number of shots on goal (N experiments) over the course of the next 12-18 months to maximize the potential of finding a likely optima:

    • In pre-seed this means partnering with "complete" teams, or teams I'm reasonably confident I can help complete with specific people in my network (i.e. teams that can both build and sell in the space they're operating in)

    • In seed stage (i.e. have been operating for at least a few months), treat as an optimization/search algorithm that's had a few iterations - understand and evaluate prior iterations and likelihood of converging to an attractive maxima

• Ensure founders are comfortable iterating. One of the most damaging things I've seen happen to companies is they raise too much money too soon from an investor with expectations that a company will do a particular thing. Putting more gas in a car going the wrong way and slamming the accelerator just takes you further out in the wrong direction.

• Ensure that teams are capable of fully capitalizing on an optima if one exists - can the team attract and hire talent?

• Help founders:

  • Iterate with intention

  • Build with urgency

  • Avoid local maxima and areas that have already seen evaluation (and limited success)