Great post. The gap between software-reported and customer-reported attribution interests me, and I think the problem is more nuanced than the multiple-choice options allow.
A few others stated this concisely. But I think this is also shifting with the increased use of AI tools.
At it's core, what is meaningful or helpful information to a new potential customer?
How a customer tried to find you and how they eventually engaged with you can be entirely different. I search for a product, fail to find one, ask around, and someone in a forum recommends you. I report "word of mouth." The founder reads that as a signal to spend more on social, when the actual failure was that the product wasn't discoverable through search — or that I found the site but couldn't get the information I needed to shortlist or disqualify you. The survey attempts to answer the founder's question ("which channel deserves budget?") with the least possible context, and the answer misleads. Free-form responses with LLM aggregation, random sampling, and outlier highlighting would produce a far better signal.
Blogs build trust and search ranking, and a good post mortem gets cited and (hopefully) builds confidence. But that's secondary. I have to find companies that match my requirements before I start ruling them out on anything else, and what I need for that is product detail, not content.
Consider two competitors.
Company A publishes product summaries, technical specifications, and an API reference I can exercise (Swagger or similar) without talking to anyone. Pricing is published and predictable as I scale. Uptime, incidents, and post mortems are public. The privacy policy is readable. The roadmap, support channels, and known issues are visible. GitLab is the reference case — they publish docs.gitlab.com and their entire company handbook (handbook.gitlab.com). Any startup that can do this competently already has their foot in the door.
Company B spends its budget convincing a VP the solution is right for them without saying what the solution is technically. Pages of marketing copy and case studies, no architecture, no feature breakdown, no knowledge base without a support contract, and the only path to learning or testing anything is "speak with sales."
If Company B redirected a fraction of that spend toward transparency and a frictionless trial, their chance of winning my business would go up considerably. Frequent marketing emails, bold claims, "customers don't know what they want, it's our job to convince them"... this is precisely why engineers go to great lengths to avoid sales conversations.
Help me find you, not the other way around.
And the discovery mechanics are shifting under all of this. I'm now more likely to hand an LLM a list of requirements and let it crunch whatever is publicly available than to type keywords into a search engine. Or maybe I ask the model to build a working draft of a workflow. A company gating its product information gets no representation in that answer; it will simply be ignored at that stage. Machine-readable product information like feature tables, published specs, open docs, are all cheap to create, cheap to maintain, and it's free marketing.
Software said 78% of signups came from search. Ask the customers directly and only 12% said search the rest pointed at podcasts, communities, someone mentioning it in a group chat.
That gap isn't a measurement error. It's a budget mechanism. The channel that captures demand gets credited, the channel that created it gets nothing, and next quarter money moves from the second to the first. Do that for two years and you've defunded the thing that was actually working while every dashboard tells you performance improved.
I ran marketing inside a commercial P&L and this is the fight I lost most often. Brand spend shows up as cost with no attributable revenue, so it dies in every review, and by the time the search volume dries up nobody connects the two.
The free-text box beats an attribution platform for the same reason. It measures memory, and memory is where the decision was made. The cookie only saw the last click
The insight on software attribution vs. self-reported dark social really hits home from an operations perspective. So many early-stage teams waste budget scaling paid ad channels because a dashboard credits the final click, completely ignoring the word-of-mouth or community post that actually built the intent. Adding a simple free-text input during onboarding seems like the most practical, low-friction way to capture that missing context without overcomplicating the analytics setup.
The PostHog handbook story is the whole argument in one move: credibility artifacts beat content calendars. Nobody remembers a blog post, everybody remembers the company that published its playbook. I'm running this as a live experiment - a one-person business where every decision, price, and miss is published as it happens - and the receipts outperform every "content strategy" I have watched founders burn a year on.
Great post. The gap between software-reported and customer-reported attribution interests me, and I think the problem is more nuanced than the multiple-choice options allow.
A few others stated this concisely. But I think this is also shifting with the increased use of AI tools.
At it's core, what is meaningful or helpful information to a new potential customer?
How a customer tried to find you and how they eventually engaged with you can be entirely different. I search for a product, fail to find one, ask around, and someone in a forum recommends you. I report "word of mouth." The founder reads that as a signal to spend more on social, when the actual failure was that the product wasn't discoverable through search — or that I found the site but couldn't get the information I needed to shortlist or disqualify you. The survey attempts to answer the founder's question ("which channel deserves budget?") with the least possible context, and the answer misleads. Free-form responses with LLM aggregation, random sampling, and outlier highlighting would produce a far better signal.
Blogs build trust and search ranking, and a good post mortem gets cited and (hopefully) builds confidence. But that's secondary. I have to find companies that match my requirements before I start ruling them out on anything else, and what I need for that is product detail, not content.
Consider two competitors.
Company A publishes product summaries, technical specifications, and an API reference I can exercise (Swagger or similar) without talking to anyone. Pricing is published and predictable as I scale. Uptime, incidents, and post mortems are public. The privacy policy is readable. The roadmap, support channels, and known issues are visible. GitLab is the reference case — they publish docs.gitlab.com and their entire company handbook (handbook.gitlab.com). Any startup that can do this competently already has their foot in the door.
Company B spends its budget convincing a VP the solution is right for them without saying what the solution is technically. Pages of marketing copy and case studies, no architecture, no feature breakdown, no knowledge base without a support contract, and the only path to learning or testing anything is "speak with sales."
If Company B redirected a fraction of that spend toward transparency and a frictionless trial, their chance of winning my business would go up considerably. Frequent marketing emails, bold claims, "customers don't know what they want, it's our job to convince them"... this is precisely why engineers go to great lengths to avoid sales conversations.
Help me find you, not the other way around.
And the discovery mechanics are shifting under all of this. I'm now more likely to hand an LLM a list of requirements and let it crunch whatever is publicly available than to type keywords into a search engine. Or maybe I ask the model to build a working draft of a workflow. A company gating its product information gets no representation in that answer; it will simply be ignored at that stage. Machine-readable product information like feature tables, published specs, open docs, are all cheap to create, cheap to maintain, and it's free marketing.
Software said 78% of signups came from search. Ask the customers directly and only 12% said search the rest pointed at podcasts, communities, someone mentioning it in a group chat.
That gap isn't a measurement error. It's a budget mechanism. The channel that captures demand gets credited, the channel that created it gets nothing, and next quarter money moves from the second to the first. Do that for two years and you've defunded the thing that was actually working while every dashboard tells you performance improved.
I ran marketing inside a commercial P&L and this is the fight I lost most often. Brand spend shows up as cost with no attributable revenue, so it dies in every review, and by the time the search volume dries up nobody connects the two.
The free-text box beats an attribution platform for the same reason. It measures memory, and memory is where the decision was made. The cookie only saw the last click
The attribution section mirrors what I see with clients: they spread thin across channels, then buy expensive tools to justify it.
A simple rule: give each channel a fixed small budget, test it, and measure with the free-text box.
Only double down where you see real signal.
The insight on software attribution vs. self-reported dark social really hits home from an operations perspective. So many early-stage teams waste budget scaling paid ad channels because a dashboard credits the final click, completely ignoring the word-of-mouth or community post that actually built the intent. Adding a simple free-text input during onboarding seems like the most practical, low-friction way to capture that missing context without overcomplicating the analytics setup.
The PostHog handbook story is the whole argument in one move: credibility artifacts beat content calendars. Nobody remembers a blog post, everybody remembers the company that published its playbook. I'm running this as a live experiment - a one-person business where every decision, price, and miss is published as it happens - and the receipts outperform every "content strategy" I have watched founders burn a year on.