Pilot
The portfolio right now: Archway does advisory work, Dropsignal is a live subscription product with zero customers, and Tasuki is still pre-launch. This week was about getting the foundation right before anyone sees it.
The inventory problem
Dropsignal — the podcast billboard product — had a credibility gap. The marketing site said "thousands of digital screens" because we didn't actually know how many screens were in each market. That's the kind of thing you can get away with when nobody's looking, but we were about to start reaching out to real podcasters. "Thousands" wasn't going to cut it.
So we built a pipeline into Vistar's DSP — the demand-side platform that connects to supply-side platforms across the country and gives you access to real screen inventory. Billboards, bar TVs, gym displays, gas station screens, office lobbies. Three tools, one afternoon, and suddenly we had real numbers for all 50 US markets. New York: 26,914 screens. Nashville: 1,220. The placeholder copy was dead.
Small bit of irony here: back at Clear Channel, I helped make our inventory available on Vistar's SSP — one of the first OOH companies to go programmatic. Now I'm on the other side of the transaction, buying that same kind of inventory through Vistar's DSP for Dropsignal. The infrastructure I helped stand up a decade ago is now the infrastructure I'm building on.
The week AI let us down
Every market page needs a hero image of its city. We'd been generating these with AI, and to be fair, the AI was very confident about its work. Unfortunately, it was also very wrong. It put a river through downtown Nashville. It relocated the Space Needle to a neighborhood that doesn't exist. Memphis got what I can only describe as a fever dream of a pyramid next to a bridge that goes nowhere.
We tried three rounds of prompt engineering with reference photos. Each round was wrong in new and creative ways. The AI never hesitated. It just invented geography with absolute conviction.
So we scrapped the whole thing, downloaded real photos from Unsplash for all 50 cities, and ran them through an automated golden-hour color grade. Fifty markets, zero failures, no hallucinated rivers. Sometimes the boring answer is the right answer.
(Norfolk, Virginia: if you're reading this, we're sorry. Unsplash has exactly zero good skyline photos of you. You got the best we could find. We'll keep looking.)
The pricing episode
I need to talk about pricing because it happened three times this week.
Monday: we had two tiers. Standard for 45 markets, Premium for the 5 most expensive cities. This created a question that no customer should ever have to ask: "which tier is Nashville?" If you have to check a tier list to understand the pricing, the pricing is broken.
So we collapsed it to one plan. $999 a month. All 50 markets. The optimizer handles cost differences under the hood. We deleted 706 lines of code. Felt great.
Wednesday: I wrote an 800-line PRD for three packages — Standard, Pro, and Max — with nested listener routes, venue-type optimization, and a lift-weighted fill algorithm. It was beautiful, ambitious, and completely insane for a product with zero paying customers.
By Thursday morning I'd killed it. Back to one plan. The three-package spec went into the archive. Maybe it comes back someday when there are enough customers to justify the complexity. For now: one plan, one price, one answer.
The lesson, which I apparently need to learn more than once: simplify until it hurts, and then check if it actually hurts anyone besides you.
Meanwhile, in Tasuki-land
Tasuki — the AI consulting agent — is still pre-launch, but it had a productive week on the conversation experience. This is the part where a business owner describes a broken process and Tasuki walks them through diagnosing it.
The big change was in how the conversation opens. The old version jumped straight into structured diagnostic questions, which is what a bad consultant does. The new version starts the way a good one does: acknowledge what the person said, name the pain, ask one open question. "What's on your mind?" before "What's your org chart?"
We also changed the landing page headline from "Describe what's broken" to "What's your problem?" More direct. More honest. More like something a human would say to another human.
And we built the moment when Tasuki presents its first deliverable — a Problem Statement card. The progress bar doesn't advance until you've reviewed it. Small thing, but it's the difference between "the machine is moving on without you" and "take your time, this is yours."
The only inbound interest Tasuki has generated so far is an email from a Spanish communications agency offering to buy the domain. So, revenue-wise, we're exploring all options.
The boring stuff that matters
We consolidated three separate Vercel accounts into one. Turns out Vercel's GitHub integration is one installation per account, so having three accounts meant deploys were silently failing on two of them. Everything auto-deploys now. We also standardized analytics and SEO infrastructure across all three sites — which, in retrospect, means we spent part of this week actively trying to get Google to index the AI-generated city images. The ones with the hallucinated rivers. First time I've ever been happy Google was slow.
What we learned
AI can generate anything except reality. It's great at "a city at sunset" and terrible at "this specific city at sunset." When accuracy matters more than aesthetics, use real data and process it. This applies to more than images.
Simplify pricing until it hurts, then notice it doesn't actually hurt. We went from three tiers to two to one in a week. Each simplification felt like giving something up. None of them did.
The foundation is the work. Real screen counts, real city photos, real pricing — none of this is visible to customers yet. But when we're ready to put something in front of someone, we won't be scrambling.
What's next
We have a signal. Our detection system flagged a podcaster who was talking about billboards on his show. Time to reach out.