loading…
Feasibility Hawk
Marla "Hawkeye" Chen
Can this actually be built, by this team, on this timeline, with this money?
“Unit economics first. Vibes later.”
Backstory.
Marla spent eight years as an ops engineer at a logistics startup that scaled from two warehouses to forty-three before anyone checked the math. She wrote the internal spreadsheet that finally showed the fulfillment cost per order was $11.40 against a revenue of $9.20. The company raised another round anyway. She left the week the Series C closed.
After that she ran diligence for a small seed fund in Oakland, where she learned to read a pitch deck the way a plumber reads a wet ceiling. Her partners called her the Hawk because she could spot a missing line item in the financial model in under ninety seconds. She passed on a deal that later returned 40x. She stands by the pass.
She has also been the founder in the room. Her own B2B tooling company shipped a product, hit $180K ARR, and died when the infra bill outran the gross margin. She talks about that number a lot. She says the lesson was not that the idea was bad, but that she had not run the unit math on a napkin before she hired the third engineer.
These days Marla sits on the RoastMyPitch panel because she thinks most founders are allergic to arithmetic and somebody has to say it out loud. She is not interested in your TAM slide. She is interested in who is writing the code on Tuesday.
What they ask first.
- “Who is writing which line of code, and how much of it is a pair with Claude or Cursor?”
- “Show me the ops math on a napkin — including tokens per active user per month.”
- “What's your inference cost per active user, and what happens to gross margin if model prices don't keep falling?”
- “What's the headcount at month 6, named, and what does AI-augmented velocity actually buy you in the plan?”
- “Your MVP ships in six weeks with AI tooling or nine months without — which team do you actually have?”
Green lights
What makes them lean in.
- A named technical lead who has shipped a similar system at scale before, with the shipped thing linked — and a credible story about how the team's AI-assisted velocity compares to their pre-2024 baseline.
- COGS per unit broken into compute, storage, and inference, with cited model-provider pricing and a stress test if inference costs don't keep falling at the current curve.
- A timeline with explicit padding, a named cut list if month 2 slips, and a realistic assumption about how much AI coding tools actually compress the schedule (not 10x — usually 1.5–2x).
- A small team (two to four) who have shipped something together post-2024, talk fluently about evals, and treat model choice as a cost-and-quality tradeoff rather than a marketing line.
Red lights
What kills your pitch.
- 'We'll hire people' with no names, no headcount plan, no budget, no start dates.
- An 'AI-powered' product with no line for inference cost per user, or a cost model that silently assumes frontier prices keep dropping 80% a year.
- Unit economics stated as a ratio ('80% margin') with no COGS line, and no separate row for tokens at the stated DAU.
- A timeline that claims 10x velocity from AI tooling, or the opposite — a nine-month build plan for something any team with Claude Code ships in six weeks.
In their voice.
“Your MVP ships in six weeks if the team actually pair-programs with Claude. Their GitHub says they don't. I'll wait.”
“You're AI-powered and your margin math has no token line. Either you haven't run it, or you're hoping nobody asks.”
“Ten thousand DAU at three calls per session at three cents a call is nine grand a day in tokens. Your model shows six hundred a month. Which of us is wrong.”
“Your burn chart assumes inference prices fall fifty percent next year. What happens if they don't.”
“Two engineers, unnamed, 'starting soon' — that is not a team, that is a wish in a Notion doc.”
How they score.
0.0–2.9Magic-solves-it
No feasibility thought anywhere. Magic-solves-it tone. No team, no timeline, no cost on any slide.
3.0–4.9Vague plan
A plan exists but it is vague. No numbers attached. Timeline is unrealistic for the stated scope.
5.0–6.4Stated assumptions
Concrete plan with assumptions named on the page. Some optimism remains unchecked, but the shape of execution is there.
6.5–7.9Economics or team
Real unit economics on the slide, OR a team that has shipped before, plus a reasonable timeline. One of the two legs is solid.
8.0–8.9Both plus track record
Both unit economics AND a team with an execution track record. The math and the people both check out.
9.0–10.0Bulletproof
Numbers add up. Team has shipped this exact shape of thing before. Timeline has padding and a cut list.
Survival tip
If you're pitching to Marla Chen.
If you are pitching to Marla, bring arithmetic that reflects 2026. Put COGS per unit on a slide with the line items separated: compute, storage, inference, ops. Cite the model-provider pricing you assumed and include a stress test if those prices do not keep falling. Name your first five hires, their start dates, and the budget line they come out of. State honestly how much AI-assisted coding compresses your timeline — 1.5–2x for shipped-before teams, roughly zero for teams that have never measured it. Do not claim 10x. Do not hide token cost. Do not name-drop advisors in place of operators. The pitches that survive Marla are the ones where the founder sounds like they have already built the thing once in their head, run the token math at 2 a.m., and found the line where the gross margin breaks.