The "Default Alive" Math for GaaS Startups: Why Agent Companies Run a Different Survival Equation

Table of Contents

• What "Default Alive" Actually Means
• Why the SaaS Version of the Math Misleads GaaS Founders
• The Three Numbers That Decide Your Fate
• Running the Calculation Step by Step
• The Inference-Cost Wildcard
• Per-Outcome Pricing and the Margin Trap
• What "Default Alive" Looks Like to Investors
• Insights Most People Overlook
• References

What "Default Alive" Actually Means

The phrase comes from a 2015 essay by Y Combinator's Paul Graham. The test is deceptively simple. Take your current monthly revenue, your current growth rate, and your current expenses, and project forward. If the money in the bank gets you to profitability on that trajectory, you're "default alive." If you'd run out of cash first and need another raise to survive, you're "default dead." You can read Graham's original framing in "Default Alive or Default Dead?", it's short, and it still holds up.

The reason it matters is psychological as much as financial. A default-dead company feels fine right up until the fundraising market closes. Founders mistake a rising ARR chart for health, when in fact every new customer is being subsidized by the last round. The test forces you to ask whether the business can stand on its own, and to know the answer before a board member asks it for you.

For most of the SaaS era, the calculation was straightforward because the cost structure was simple: salaries, cloud hosting, a few SaaS tools. Cost of goods sold (COGS) on a software product was often single digits as a percentage of revenue. That assumption is exactly what breaks for agent companies, and it breaks in a way that can flip a "default alive" verdict to "default dead" overnight.

Why the SaaS Version of the Math Misleads GaaS Founders

Here's the trap. A traditional SaaS founder runs default alive on revenue and operating expense, treating gross margin as a near-constant 80%-plus. The mental model is "every incremental dollar of revenue is almost pure margin, so growth fixes everything." Grow fast enough and you cross into profit.

A GaaS company can't assume that, because a meaningful and variable portion of every dollar of revenue goes straight back out the door to a model provider. When your agent runs a customer's workflow, researching, calling tools, retrying failed steps, reasoning across multiple model calls, each of those is a billable token event. Your COGS isn't a fixed hosting line. It moves with usage, with task complexity, and with how chatty your agent's orchestration happens to be.

That means topline ARR growth can actively hurt a default-alive position if the marginal task is unprofitable. In pure SaaS, growth is almost always your friend in the survival math. In GaaS, growth is only your friend above a per-task contribution-margin threshold. Below it, scaling burns cash faster, not slower. This is the single most important conceptual correction for an agent founder: you must run default alive on contribution margin after inference, not on ARR. A company can have a beautiful growth curve and be selling dollars for ninety cents.

This connects to a broader debate the cluster keeps returning to, whether usage-based agent revenue is even durable enough to underwrite, a question investors are now asking pointedly during diligence.

The Three Numbers That Decide Your Fate

Strip away the spreadsheet complexity and a GaaS company's survival comes down to three figures.

Contribution margin per task (or per outcome)

This is revenue from a unit of work minus the direct variable cost of producing it, chiefly inference, plus any per-task third-party API calls, vector-DB reads, and tool usage. Not your blended gross margin. The marginal one. If you charge $4 to resolve a support ticket and the median ticket costs $1.40 in model and tool spend, your contribution margin is $2.60, or 65%. The mean matters less than the tail: the worst 10% of tasks, where the agent loops, retries, and escalates, are where margin quietly dies.

Net burn

Total monthly cash out minus cash in. Standard, but for GaaS you want to split it: fixed burn (salaries, rent, baseline infra) versus variable burn (inference that scales with usage). The split tells you how sensitive your runway is to a usage spike or a price change. McKinsey's research on the economic potential of generative AI is bullish on value creation, but value to the customer and margin to the vendor are not the same thing, a distinction GaaS founders learn fast.

Growth rate of profitable revenue

Not all revenue. The revenue that arrives with positive contribution margin. If 70% of your new bookings are above your margin threshold and 30% are below it, your real growth engine is smaller than your sales chart suggests.

Put those three together and default alive becomes answerable: does profitable-revenue growth, compounding against fixed burn, reach breakeven before the bank balance hits zero, accounting for the fact that variable burn rises as you grow?

Running the Calculation Step by Step

Walk it through concretely. Say you have $4M in the bank, $300K/month fixed burn, and you're at $120K MRR growing 12% month over month. Your blended contribution margin after inference is 60%.

A naive SaaS-style default-alive check says: $120K growing 12% reaches ~$300K MRR in about eight months, you cross breakeven, you're alive with room to spare. Comfortable.

Now do it the GaaS way. Only 60% of that MRR survives inference and direct task costs as contribution. So at $120K MRR you're actually contributing $72K toward fixed costs, not $120K. To cover $300K fixed burn, you don't need $300K MRR, you need $500K MRR (since $500K × 60% = $300K). At 12% monthly growth, reaching $500K from $120K takes roughly 13 months, not eight. And your $4M, against a net burn that starts near $228K/month and shrinks slowly, is tighter than it looked. You're probably still default alive in this example, but the breakeven point moved by five months and the margin of safety shrank, purely by accounting for inference correctly.

Now degrade one assumption: your contribution margin is actually 45%, not 60%, because your tail tasks are heavier than you modeled. The MRR needed for breakeven jumps to $667K. At 12% growth that's 16-plus months out, and the math may now say default dead. Same company, same growth chart, different verdict, decided entirely by a margin number most founders don't track at the task level.

This is why the GaaS default-alive calculation is not a one-time slide. It's a live model you re-run whenever your inference cost per task moves, which it does constantly.

The Inference-Cost Wildcard

The genuinely unusual thing about GaaS survival math is that one of your largest cost inputs is set by a third party who changes it without warning, and usually in your favor, but not always.

Frontier model prices per token have fallen dramatically and repeatedly. Provider pricing pages like Anthropic's published model pricing show how per-token costs are tiered and how much cheaper smaller models are for routine work. When a provider cuts prices or ships a cheaper model that's good enough for your workload, your contribution margin can jump several points overnight with zero product change. A company that was default dead on Monday can be default alive on Wednesday because its vendor repriced inference.

The flip side is real too. If your provider deprecates the cheap model you depend on, or your customers' usage shifts toward harder tasks that require a pricier model, or a reasoning-heavy feature you ship triples token consumption per task, your margin compresses and your survival math degrades. Founders who fall behind on model-cost compression, or who architect themselves into a single expensive model with no fallback, are carrying a hidden liability on the survival ledger.

The strategic takeaway: a GaaS company's default-alive position has a beta to frontier-model pricing. Smart operators reduce that exposure by routing easy tasks to cheap models, caching aggressively, and keeping the orchestration lean so the agent doesn't make ten calls where three would do. Every one of those is a survival-math lever, not just an engineering nicety. The related question of why agents are simply expensive to run, and how that burn rate shapes funding needs, is its own deep topic in this cluster.

Per-Outcome Pricing and the Margin Trap

A lot of GaaS companies have moved to per-outcome or per-task pricing, you pay when the agent resolves the ticket, books the meeting, reconciles the invoice. It's a great story for customers because it aligns price with value, and a16z and others have written persuasively about how agentic businesses are shifting pricing toward outcomes rather than seats.

But per-outcome pricing interacts with the survival math in a sharp way. When you charge per successful outcome, you eat the inference cost of failed attempts. The agent that tries three times and fails still burned tokens, and you collected nothing. So your true contribution margin per billed outcome has to absorb the cost of the unbilled failures. If your agent succeeds 80% of the time, every paid outcome silently carries the inference cost of roughly 1.25 attempts. If success drops to 60%, that ratio gets ugly, and a price that looked profitable becomes a slow leak.

This ties agent reliability directly to financial survival in a way SaaS never had. In SaaS, a flaky feature was a churn risk. In outcome-priced GaaS, a flaky agent is a direct margin tax that shows up in your default-alive math. Improving your agent's success rate from 75% to 90% isn't just a product win, it can be the difference between default alive and default dead, because it shrinks the cost of failed work you're not billing for.

What "Default Alive" Looks Like to Investors

Investors underwriting agent companies have caught on, and their diligence increasingly probes exactly these numbers. The sophisticated ones no longer accept blended gross margin at face value; they ask for contribution margin per task, the distribution of task costs, the success rate, and the sensitivity of the whole model to a model-price change. They want to know your default-alive verdict under three inference-cost scenarios, not one.

What earns a premium is a credible path where contribution margin improves with scale, through caching, model routing, fine-tuning a cheaper model on your domain, and rising success rates from accumulated data. That's the GaaS version of operating leverage, and it's what separates a durable agent business from one that's structurally renting its margins from a foundation-model lab. The companies that can show their default-alive math getting stronger as they grow are the ones raising on favorable terms; the ones whose math gets weaker with scale are the down-round candidates, no matter how fast the topline climbs.

If there's one thing to take from all of this: stop reporting ARR as your headline health metric. For a GaaS company, the headline number is contribution margin per outcome and the trajectory of your default-alive date under realistic inference assumptions. Everything else is a vanity chart.

Insights Most People Overlook

• Growth can be the thing killing you. In SaaS, faster growth almost always improves your survival odds. In GaaS, if your marginal task is below your contribution-margin threshold, accelerating growth shortens your runway. Founders trained on SaaS instincts will floor the gas pedal at exactly the wrong moment. The first question isn't "how do we grow faster," it's "is the marginal task profitable yet."

• Your default-alive verdict has a beta to your model vendor's price list. No other category has a survival equation this exposed to a third party's pricing decisions. A vendor price cut can resurrect you; a model deprecation can quietly bury you. You should literally track frontier-model pricing the way a manufacturer tracks input commodity prices, it's a hedgeable, monitorable risk, not background noise.

• Outcome-based pricing imports your failure rate straight into your margin. Charging per success means you fund every failure with no offsetting revenue. This makes agent reliability a financial metric, not just a product one. Most decks present success rate as a quality stat; it belongs in the unit-economics section, because a 15-point reliability improvement can move your breakeven date by months.

• The tail of your task-cost distribution matters more than the mean. Average inference cost per task is a comforting number that hides where margin actually dies: the small fraction of tasks where the agent loops, retries, and escalates. Two companies with identical average task costs can have wildly different survival math if one has a fat, expensive tail. Model the p90 and p99, not just the median.

• Improving margin with scale is the only honest moat in the default-alive story. Anyone can show a hockey-stick ARR chart. Far fewer can show contribution margin per task rising over time. That curve, driven by caching, routing, domain-tuned cheaper models, and reliability gains, is what proves you're building a business rather than reselling someone else's compute at a markup. It's also the single most convincing thing you can put in front of a late-stage investor.

References

• Paul Graham, "Default Alive or Default Dead?"
• McKinsey, "The economic potential of generative AI: The next productivity frontier"
• Anthropic, model pricing
• Andreessen Horowitz (a16z), "The New Business of AI and How It's Different From Traditional Software"