Podscan: $4,000 in the Red Every Month, Two Months After Declaring Profitability — Arvid Kahl's Second Act Is a Fight With GPU Bills

The “second act” of someone who sold a SaaS for seven figures is usually told as a glossy success story. Arvid Kahl is doing the opposite. About his current business, Podscan.fm, he keeps talking through the numbers (MRR, expenses, the size of the deficit) on his weekly podcast. In April 2025 he reported “finally profitable”; two months later, in June, he acknowledged the slide back into the red: “around $10,000 or so a month in expenses, somewhere around $6,000 in monthly recurring revenue — about $4,000 short of breaking even.” Continuous disclosure that includes the bad months makes this a rare sequel to a success story.

Kahl is the founder who ran FeedbackPanda, a SaaS for teachers, as a two-person husband-and-wife operation for two years and sold it for seven figures at $55K MRR. He then moved to the commentary side of “build, grow, sell” through his books and the newsletter he grew to 7,600 subscribers, before returning to the field in early 2024 with Podscan. The service transcribes podcasts worldwide, detects mentions of company or product names in near real time, and offers the accumulated data through an API. It currently processes roughly 50,000 new episodes per day, with a database of 33 million episodes across 3.8 million shows.

The numbers over time

The cost breakdown at the one-month mark is concrete: $1,200 for GPU cloud servers doing transcription, about $500 for AWS’s managed database (RDS), and about $200 for the app server, full-text search and the rest. $300 of MRR against $2,000 of expenses, underwater from month one. At that point, over 400 users had signed up, a few hundred were on the 10-day trial (about half of them active), and ten had converted to paid. Half of the paying customers were using the API for programmatic data access rather than the mention alerts: that Podscan’s real product was the data platform, not the notification feature, was already visible in month one’s usage patterns.

Why this SaaS is heavy

An ordinary SaaS doesn’t incur more cost unless it gains more customers. Podscan is different. The value proposition itself, “we transcribe every podcast”, is a lump of fixed cost that accrues regardless of customer count. The GPU bill for processing 50,000 episodes a day is the same whether there are ten paying customers or a thousand. Kahl estimated that pushing the workload to OpenAI’s API would cost about $5,000 a day, and chose to run his own GPU fleet instead. The transcripts also document how he ground infrastructure costs down from a peak of about $30,000 a month to just under $10,000 through repeated optimization.

Not everything on the road to profitability was sales work, either. In the April 2025 transcript, what Kahl credits is a repositioning. Podscan’s original homepage was, in his words, a “weird chimera landing page that does everything and nothing.” He rebuilt it around the product’s nature as a data platform and gave each customer type its own landing page. The same asset, transcripts of every podcast, is shown to PR firms as “mention monitoring,” to developers as “an API,” to research shops as “a dataset.” The main driver of profitability, by his account, was not what he built but adjusting who it is sold to, and as what.

The cost structure also shaped his funding decision. A lifelong advocate of full self-funding, Kahl raised money in April 2024 from a bootstrapper-compatible fund (a small equity sale with weak growth pressure). Supporting a capital-intensive data platform out of savings alone was, he judged, not rational. Even so, his first goal was “become profitable before the funds run out”, reached one year later, in April 2025.

Why profitability evaporated in two months

The immediate cause of the June 2025 relapse was the churn of one large customer. Kahl explains it was driven by circumstances on the customer’s side, not dissatisfaction with the product, but whatever the cause, if the base of your MRR depends on a few large accounts, one cancellation flips the P&L. A fluctuation that would be rounding error for a small SaaS running on $2,000 a month becomes, for Podscan and its $10,000 fixed cost base, an immediate “$4,000 short every month.”

The response is specific: he brought in sales support to build outreach and demo-booking systems, added a premium tier at $2,500 a month, and shifted weight from product-led growth to high-touch sales aimed at agencies and companies. Rather than stacking low-price self-serve customers, go after a small number of customers who buy the value of the data platform outright, reshaping the revenue side to match a fixed-cost production side. He has publicly given himself “a couple of months” to prove the sales strategy before weighing alternatives, including a possible sale of the business.

The value of talking about what isn’t working

The documentary value of this case lies not in a recipe for success but in the resolution of its failure disclosure. Plenty of founders practice building in public. Most become talkative only while the numbers go up. Kahl reported the profitability milestone and the relapse into losses on the same channel at the same precision, so listeners can track GPU bills, RDS costs, churn, and pricing experiments. Placed next to a lightweight solo operation like Healthchecks.io, run by one person for eleven years, it shows how drastically cost structure changes the difficulty of the same “solo SaaS” label.

What generalizes, and what doesn’t

A few of these mechanics apply to any heavy-compute SaaS. If “we process everything” is your value proposition, cost detaches from customer count and turns into fixed cost. This class of SaaS has a high break-even point and pairs badly with bootstrapping. When MRR concentrates in large accounts, profitability can vanish overnight, judging a P&L requires looking past the MRR total to its composition. And cost-side improvements like cutting processing expenses with your own GPUs ($30,000 down to just under $10,000) can pay off before sales efforts do.

The limits are equally clear: this is a story of someone with a seven-figure exit and a large audience still flirting with losses. An unknown individual attempting the same business would find both fundraising and large-customer acquisition far harder. The most recent figures here are from the June 2025 transcript. Whether the sales pivot worked cannot be determined from these sources. We will append updates as the numbers move.