How do you position against a competitor with a larger model budget?

Picture a startup with eleven engineers up against a company that spends more on model training in a month than the startup has raised in its life. Marrow Notes writes session notes for outpatient mental health practices from a recorded therapy session. Meridian Scale is the well-funded general meeting-notes company whose product a therapist could just as easily point at the same session. Ines Bramwell founded Marrow.

How to answer this, stage by stage

Nobody is scoring whether you know the word "positioning." They're scoring whether you can name the one error that actually matters and prove you're not just hiding from a fight you'd lose.

Let's learn

Here is what happens when a small AI product tries to win the same fight a much bigger competitor is built to win.

Before Marrow Notes, a therapist typed up session notes by hand after each appointment, about eighteen minutes a session, working from memory and a few scribbled words. Marrow's early version, transcribing and drafting a note automatically, cut that to about four minutes of review. Meridian Scale's general meeting-notes tool could do something similar for any kind of call, backed by a model budget Marrow couldn't match dollar for dollar.

Marrow picked the corner Meridian's size can't easily follow it into.

For months, Marrow tried to keep pace on raw summary quality, benchmarking its output against Meridian's on general clarity and phrasing. Here's the turn: the slightly plainer summaries were never the real risk. The real risk was a single missing field in a clinical documentation format, the kind of error a generalist tool has no reason to even know exists, and Marrow's own generic-mode draft nearly let one through.

At its worst, chasing quality parity against a bigger model budget doesn't just waste engineering time. It leaves the one error that could end a practice's license unguarded, because all the attention went toward a fight the smaller product was never going to win anyway.

What I would leave alone: Marrow's general summary phrasing never needed to match Meridian's exactly. A slightly plainer note that still hits every required field is a fine trade a therapist will happily take.

The lesson: against a bigger budget, the fight to avoid is the one they're built to win by spending more. The fight to pick is the one where spending more doesn't help them at all.

Now here is the same thing as a story

The short version above is what you'd say defending this bet to your own board. Read this one for the night an audit almost went the other way.

Before Marrow, Ines Bramwell spent two years as a product manager at a hospital system, sitting in on compliance reviews with a headset clipped over one ear so she could half-listen to a call while reading a printed audit checklist. She knew, better than most engineers at a well-funded rival ever would, exactly which fields a state licensing board actually checks.

Marrow launched, and for its first year, therapists loved it. It saved real time, and on the rare afternoon someone compared its output to Meridian's newer, flashier version, Marrow's read a little plainer, a little less polished. Nobody at any of Marrow's dozen partner clinics ever mentioned it twice.

Meridian's size is real. It's also aimed at everyone, which means aimed at no one in particular.

Then, the night before a scheduled state audit at one partner clinic, the practice's compliance officer did her usual pre-audit read-through and found it: a session note, generated by Marrow's fastest, most generic drafting mode, missing the mandatory risk-assessment field entirely. She caught it at eleven at night, added the field by hand, and the audit went fine the next morning. Nobody outside the clinic ever knew how close it came.

Ines got the call the next morning. Her first reaction wasn't relief, it was recognizing exactly how close Marrow had come to competing on the wrong axis entirely: chasing Meridian's polish while leaving a licensing-grade gap sitting in its own fastest mode.

The team rebuilt Marrow's drafting mode around one hard rule: no note ships without every state-required field present, verified before a therapist ever sees a draft, even if that made the draft a few seconds slower and occasionally a little more repetitive to read. Meridian, still treating mental health as one small segment among many, had no reason to build anything like it.

One box is a shrug. The other box is a license.

Six weeks later, Marrow repriced entirely: a flat per-practice-seat fee with the compliance guarantee built in, instead of a per-minute rate copied from a company ten times its size. The near miss never happened again, not because the model got smarter, but because the one field it used to skip couldn't be skipped anymore.

PICK, when the other side has the bigger model budgetNot the classic "match the benchmark" pitch. This is what happens when the smaller player picks a fight the bigger one has no reason to show up for.

The recap, one line per letter: position is committing to compliant notes for one vertical instead of trying to out-summarize a generalist; impact is naming a thirty-second shrug against a compliance officer's overnight scramble; cost asymmetry is optimizing entirely against the second, rarer error; kill criteria is naming exactly what would make the narrow bet stop working.

And if you want to be sure it really works, try it somewhere elseSame four letters, an expense-report auditing tool instead of a therapy note-taker. This time the reversal is about packaging, not pricing.

Coalpine Audit flags questionable line items in employee expense reports for mid-sized companies, competing against a much larger general finance-AI platform with a far bigger training budget. Percy Adeyemi runs product there. Mapped onto PICK: position is guaranteeing every flagged item cites the exact policy clause it violates, not a confidence score, which the larger platform never bothers to surface since it serves every industry, not finance-policy detail. Impact is that a generic miscategorized expense costs an employee a mildly annoying resubmission, while an unflagged policy violation that reaches a public company's external auditors costs real regulatory exposure. Cost asymmetry means Coalpine spends its engineering effort entirely on policy-citation accuracy, accepting a plainer categorization experience elsewhere. Kill criteria is if the larger platform ever builds real per-company policy ingestion, not just better general categorization. The old decision here isn't a pricing call, it's a packaging one: Coalpine originally bundled its policy-citation feature as a paid add-on to keep the base price competitive with the larger platform's flat rate, which meant the one thing that actually mattered was the first thing customers turned off to save money.

The same four-part discipline holds up a therapy note and an expense-policy citation alike.

Swap the trigger and it still runs.
Speed: an interviewer caps you at sixty seconds. Say "don't fight on the error they absorb for free, fight on the one they can't afford to get wrong for your customer," and stop.
Cost: there's no budget to build the full compliance pipeline before the next renewal cycle. Say so honestly, and ship the required-field check first, the polish later, since the check is the part that actually protects someone.
The model gets better, for real: if a new model release closes the generic quality gap entirely, that's good news, it just means the position gets easier to defend, not that the position was wrong to begin with.

Where people run it wrong.
They try to match a bigger competitor's model quality dollar for dollar, a fight their budget can't sustain.
They price to look comparable instead of packaging around what they actually own.
They wait for a compliance officer to catch the real risk instead of naming it as the thing worth protecting from day one.

How to use it live. The moment someone asks how to position against a bigger-budget rival, ask yourself out loud: which of their mistakes costs their customer nothing, and which one costs mine everything? Build the whole answer around the second one.

The repricing came six weeks after the scare, not six weeks before it. That gap is the whole lesson.

Flashcards (tap any card to flip it)

Check yourself Score: 0 / 0