i was killing two healthy revenue streams to keep a corpse alive. and i did not know it until an ai agent sat me down, looked at my numbers, and showed me exactly where my time and money were going. it was not a gentle conversation.
here is what the audit found, and here is what i did about it.
the stream i loved was the one bleeding me dry
every creator has a favorite child. the work that feels like the real work. the project that made you a creator in the first place. for me, that was original long-form content. the deep pieces. the ones that took weeks.
and it was dying. not slowly. not subtly. it was consuming resources at a rate the data made embarrassing to look at.
the research on this is brutal. a 2022 viewpoint in angewandte chemie laid it out plainly: sunk-cost bias keeps you investing in a failing project because of what you already put in, not because of what it will return. and the wharton study on 558 public-company acquisitions found the exact same distortion at corporate scale. companies that had sunk more into an acquisition were measurably less likely to divest it, even when the data said to walk away.
i was running a one-person version of a fortune 500 mistake.
what the ai agent actually did
people hear "ai agent auditing your creative output" and picture a robot judging their art. it is not that. it is a margin analysis on your actual business. the same framework a CFO would run, but pointed at your streams instead of their product lines.
the cfobridge framework breaks it down: segment revenue, attach direct and variable costs to each line, and calculate contribution margin per stream. that last number is the one that hurts.
here is what mine looked like after the audit:
- stream one: high-effort, high-emotion work. the stuff i framed as my identity.
- stream two: steady, unglamorous, repeatable. the stuff i never talked about.
- stream three: the new thing that was quietly growing. smaller numbers, but the margin was real and the momentum was compounding.
i was pouring my best hours into stream one. the audit showed stream one was being subsidized by stream two and stream three. it was not just underperforming on its own. it was actively cannibalizing the health of the other two.
that is the part that broke me. it was not that one stream was weak. it was that i was strangling two healthy streams to feed a dying one.
the diversification myth i had been telling myself
here is the story i had been running: i am diversified. i have three streams. that is the smart move. look at me, being financially responsible.
diversification is not three streams. it is three streams that each stand on their own. the research from the artist community said what i refused to hear: one artist described relying on original paintings as a painful, vulnerable model and found smaller originals, prints, and functional art were actually attracting buyers while the big pieces sat. the demand was there. it was just in the streams i was neglecting.
the folksy forum thread was full of the same story. creators subsidizing their art sales with workshops, teaching, pension income. not because they wanted to. because the flagship stream alone could not carry them. and some deliberately limited those supplemental streams because the scale required conflicted with the work they actually wanted to make.
that is the trap. you tell yourself the supplemental streams are beneath you. the data tells you they are the only thing keeping the lights on.
the question that changed everything
the sunk-cost research offers a cleaner test than anyone wants to admit. instead of asking "is this stream failing?" ask "what information would tell me whether to stop, and can i test for it cheaply?"
that reframe is everything. termination is not failure. it is the evidence-based decision to redirect. the hacker news thread on equity nailed the deeper issue: planning effort and past assumptions should not anchor your present decisions. current execution signals, paying customers, real conversion, those are the evidence. everything else is a story you are telling yourself.
so i ran the test on stream one. i asked what would actually prove it was viable or not. and the answer was not "try harder." the answer was "the data says it has been dying for six months and you have been ignoring it because you love it."
love is not a business model. it is a reason to start. it is not a reason to stay.
what changed when i stopped strangling the healthy streams
i did not kill stream one entirely. but i stopped feeding it first. the margin analysis made the allocation obvious: stream two and stream three had the contribution margin. they got the hours.
within a month, the numbers shifted. not dramatically. not overnight. but the direction changed. the healthy streams accelerated because they finally had resources, and the dying stream stopped draining the others.
the trusted article on product profitability made the point that ranked me out of my denial: margin analysis should drive portfolio decisions, not affection. capacity is finite. every hour i gave the dying stream was an hour taken from a stream that would have returned something. that is the real cost. not the money i lost on the dying one. the money i never made on the healthy ones.
opportunity cost is invisible until someone shows you the other column.
the takeaway
if you are a creator and you have more than one way to make money, run the audit. not next quarter. this week. attach real costs to each stream. calculate the contribution margin. identify which stream is subsidizing which other stream. and then have the hard conversation with yourself about whether you are keeping something alive because it works or because you love it.
the research is clear: sunk cost bias is real, it hits individuals harder than sophisticated operators, and the only mitigation that works is independent review. for me, that independent review was the ai agent. it had no emotional attachment to my identity as the person who makes the deep, slow, beautiful work. it just looked at the numbers and told me where to put my hours.
you can be that reviewer for yourself. but you have to be honest enough to let the data override the story.
the two streams i was strangling are not strangling anymore. the dying one got cut down to size. and for the first time in a long time, the numbers and the work point in the same direction.
that is not a small thing. that is the whole thing.