How do I know if my startup has a GTM problem or a product problem?

A signature is not the outcome

A startup has a product problem when the people it meant to serve use the core and still miss the promised outcome. It has a go-to-market problem when a defined set of customers hits that outcome and would miss the product, but the company cannot find, explain, or convert more of them.

Jyoti Bansal defines fit as value delivered after the sale, not the signature. No happy, successful customers means he treats the stall as a product problem. His enterprise bar is value to 25 customers. At seed, zero outcomes means do not scale. Outcomes in a lookalike set mean the stall is distribution.

Mark Roberge calls a pitch that closes people who do not need the thing market-message fit. Fit is delivery. He wants the early event stayers hit and leavers miss. His examples are 2,000 Slack messages in month one, and five or more of 25 HubSpot features in that month and after. Write the event for your job.

Marc Andreessen defines fit as a good market plus a product that can satisfy it. His “not happening” list sets missed value beside word of mouth that does not spread, usage that does not grow, a long sales cycle, and deals that never close. Lovers who would miss the product and still send nobody are a go-to-market problem. Usage that never grows inside the segment is a product problem. The essay also assumes acquisition does not cost more than the customer returns. Treat a slow close as go-to-market only after closers are getting value.

The break has an address

Name the stage where behavior stops.

Elaine Zelby gives each stage a read. They act and never sit down: follow-up, not the product. A first meeting that does not become a second: the sentence does not match the product. A late death: the sales process. Procurement: the champion, or a deal the buyer cannot take upstairs. Silence can be the message or a weak pain. A failed proof of concept can be the process or the product. A security stall is a normal pause, not a rebuild. A data-processing review or a works-council step is the same pause.

What you observe	Read it as	Flip the label if
The segment never replies	Reach, or a sentence they cannot repeat	A direct note still gets no talk about the job
They arrive and do not start	Message, or a pain that is not urgent	One use case loves it, and a narrower page moves the next cohort
A pilot misses the event	First run, a hollow pilot, or the product	The pilot never tested the objection they named
Every account needs its own build	You are selling a service	The same setup works for the next lookalike
Lookalikes each use a different slice	You do not have one segment yet	They name the same job and the same missing outcome
They leave and have no alternative	The product	They leave for a tool that does the job
One use, then silence	The job does not recur, or the product does not hold it	Only outsiders leave
They stay, and you cannot find more	Distribution	Do not open the roadmap
Renewal needs a discount, or legal kills it	Who feels the impact, the champion, or a normal pause	Usage died, or the user never hit the event

If people can start alone, read activation the way Zelby reads a second meeting.

The word GTM hides four different fits

Go-to-market is not one failure. A product can satisfy a segment and still stall because the channel, the price, or the way that market buys does not fit.

Brian Balfour says product-market fit is not the only thing that matters. Companies show the signals and still struggle to grow. His four fits for venture-speed growth are market-product, product-channel, channel-model, and model-market. Products are built to fit channels. Channels do not mold to products. Calling fit, then shopping channels, is the wrong order.

1. Market-product. If this segment misses the outcome, stop the channel. Andreessen’s point is that a great product in a market that does not exist still fails.
2. Product-channel. A guided setup looks dead in a self-serve ad. A product users can start alone looks slow if you only sell it through a long committee.
3. Channel-model. Balfour’s later illustration pairs a self-serve product a buyer can start alone with an enterprise sales team, and a large annual agreement with a viral social channel. Neither pair holds.
4. Model-market. Emery Rosansky asks whether the user can buy or must fetch a boss. If they cannot buy, a self-serve page is the wrong product test.

Score the people who used it, then split them

Ask recent users of the core how they would feel if they could no longer use it. Split the answers before you call the product a failure.

Sean Ellis uses that question. Above 40% very disappointed, he tends to call fit. Below that, he tells teams not to grow hard, and not to treat a launch date as the milestone. Ignore people who would not be disappointed. Read somewhat-disappointed notes only where they sharpen the lovers’ use case.

Rahul Vohra surveys people who used the product at least twice in the last two weeks. Around 40 answers, the read is directional. Below that, skip the percentage. He also asks who benefits, the main benefit, and what to improve. Superhuman sat at 22%, then 33% after keeping the personas inside that group, then 58% within three quarters of building for them. The same account of Hiten Shah’s 2015 check is 731 Slack users, 51% very disappointed. His write-up that month says Slack was already used by over 750,000 people a day.

Ellis describes a suite at about 7% whose lovers all used one piece. Leading with only that piece, and aiming the first session at it, put the next cohort over 40% a week later, with no real product change. A low score can be the sentence. Vohra drops requests from people who would not miss the product, and from fence-sitters who wanted a different benefit. Same-benefit fence-sitters named the gap. For Superhuman it was mobile. A request is a symptom. Keep it only when it is what stops the outcome they already want.

Record the job and the workaround too. No time already spent on the job means unproven pain, not a channel miss.

Fourteen days, one lever

Change one lever in a fortnight: the audience and the sentence, or the first run, or the product path. Write the break before you touch a second lever.

1. One segment you can refuse, and one outcome event in the first days or weeks. See what to measure in early tests.
2. Everyone who used the core at least twice lately. A few design partners means read each account and skip the percentage.
3. The disappointed question, the job, the workaround, the event, and what stopped the next use.
4. Lovers, fence-sitters who share that benefit, and everyone else. Do not average them.
5. If a segment hits the event and would miss the product, freeze the roadmap. Put their words on the first screen. Test that angle on a small batch of lookalikes.
6. If the right people start and miss the event, stop volume and watch three. Explanation belongs in the first run. A failed job belongs in the build.
7. If the event happens only while you are present, write your steps. Labor you cannot remove is still a service. A missing path is a build.
8. Day 14, one sentence: this stage, these accounts. Pick the next experiment from it.

Skip the test when nobody has used the core. Get a trickle through the product, then the score, then growth. Two messages, a pilot, and a blast in one fortnight make the cohort unreadable.

What has to be true before seed marketing gets louder

Before you scale marketing at seed, have a segment you can refuse, an outcome you can see, and some right-fit users who hit it without you doing the job by hand.

A company Rosansky worked with aimed at security chiefs in younger companies and found the pain two or three levels down, in large enterprises. She warns that spray-and-pray leaves customers you cannot draw a pattern from. Selling machine health to manufacturers, win rates rose after a narrower cut. Define the ICP so a stranger can refuse it, and check that refusal on behavior before you raise volume. Self-serve still needs a start without a seller, a user who can buy, and value without a long setup.

Do not scale paid volume, a launch, quotas, or a seller from a company that already had a playbook. Roberge wants the first sales hire to bring back patterns from the conversations, and he would hold part of the variable until the retention event hits. Scale the motion after the event repeats in one segment.

Do not rebuild, and do not spend, in these cases

Do not rebuild when a narrow segment already reaches the outcome. Do not scale marketing when nobody in that segment does. Do not spend when the segment, the promise, and the channel will all change this month.

I name the break before I touch a channel

I would not add volume while the label is still a guess. I would read who hit the outcome, who only signed, and whether the page says what the lovers already get.

I am one person, under the name Poldermarketing, and I work remote in Dutch and English. Freshly funded startups hire me freelance, or for a few days a week, as one person for the marketing, the AI, and the automation, from the first sentence through to the first customers, without a separate specialist for each part. I am strong in AI, content, automation, and building workflows in your accounts. Google Ads and Meta Ads are newer for me. I can set them up and review them. I am the wrong person when the work is scaling a large media program.

The first week is the cohort. One change: the audience, the sentence, or the first run. The free growth scan shows how the site reads to a stranger. Positioning and messaging is where that sentence is rewritten. How I work is the shape of an engagement.