research/
Reads before it spends
Reads your site, pricing, docs and payment data, then writes the first plan as hypotheses you can correct: who is in pain, which promise to test, what a trial has to prove.
You ship. Agents pinpoint demand. 24/7, recursively self‑improving.
A team of agents watches the market in real time, puts your self-serve software in front of exactly the people who need it now, runs the trial, the payment check and the ledger — and each round trains the next. You connect the product and approve one page.
Watch it work.
Three views of one loop. The search: a market of twenty thousand searches and eight scouts that start everywhere and end on the cluster that pays. The funnel: a Markov chain whose odds are estimated live and rewritten every round. The log: what the team actually did across a seat’s first twelve days, including the moment it stopped itself.
Replay of the team’s log for one example seat over twelve days: research writes hypotheses, build deploys the trial page and campaign, monitor verifies checkout, pauses spend when a deploy breaks the trial, and resumes; ledger closes three rounds with verdicts hold, keep running, then raise; research re-plans from each ledger.
fig. 3 — the team’s own log for one example seat, replayed: what it built, what it verified, when it stopped itself, and how each round’s ledger rewrote the next plan.
Same spend. Three verdicts.
The figures below are invented; the mechanics are not. Every round closes with the real numbers and one verdict — hold, keep running, or raise — and the next plan is rewritten from the ledger, not from a calendar.
Round 2 exists because round 1 was measured. Round 3 is better because round 2 was.
# trial: Upload 30 seconds of your own recording. Hear it cleaned before you sign up.
| round | media out | trials | paid | collected | media / customer |
|---|---|---|---|---|---|
| 1first plan, written from your site, pricing and docsNot paying back — hold spend | $600 | 64 | 8 | $232 | $75.00 |
| 2re-planned: promise narrowed to single-speaker recordingsPromising — keep running, no new budget | $600 | 71 | 14 | $406 | $42.86 |
| 3re-planned: trial moved to the first screen, checkout path shortenedPaying back — raise to $1,800 | $600 | 69 | 23 | $667 | $26.09 |
Paying back — raise to $1,800
Three rounds at the same spend, improving each time. Worth more money — if you approve it.
> waiting on your approval before a dollar of the $1,800 is spent
Four agents. One operator.
build/
Builds the whole path
The campaign, the trial page that calls your real product, and the tracking that ties a click to a payment. Publishes only after you approve the page.
monitor/
Watches every hour
If your deploy breaks the trial, it pauses spend within minutes and tells you what broke. Spend caps are yours. Nothing raises without your approval.
ledger/
Closes every round with a verdict
Real spend, real trials, real payments — then one instruction: hold, keep running, or raise. It holds spend itself when a round is not paying back.
operator/ — One person you can name handles exceptions, signs off on every raise, and is on the hook for the guarantee.
It rewrites its own plan. Every round.
Most customer acquisition runs on a calendar: a monthly report, a quarterly review. This team runs on evidence. Each round closes into a ledger; the ledger rewrites the next plan; the next plan is tested against the same spend.
What one seat teaches stays on file for the next — which promises earned a trial, which trials earned a payment, which fix moved the number. Every seat inherits the rounds before it.
cross-seat transfer is the claim we test hardest. we publish whether it holds.
Customers, or it’s free.
A founding seat is US$1,500 to set up, refundable in full — the standard seat is US$4,500 after October 31, 2026. If the team has not brought you a paying customer within 45 days of launch, you pay nothing — and you keep everything it built: the trial page, the tracking, the ad account setup.
After that, the team earns 30% of what each new customer pays in their first 90 days, billed after the money clears. It is paid on customers who stay, not on signups.
Media budget is yours, capped by you, and never raised without your approval. Media, trial compute and the seat appear as three separate lines.
what the guarantee needs from you
- - your product stays up for the 45 days
- - checkout works — the team tests it before a stranger does
- - a media cap of at least US$500 a round, set by you, never raised by us
- - approvals answered within two working days
- - “a paying customer” means: came through our channel, payment cleared, your refund window passed
- - “launch” means the first ad approved and live — Google usually reviews within a working day, sometimes longer; that queue is not ours
- - no API or SDK we can call means 3–10 working days of integration before launch, and we say so before you pay
The offer, side by side.
| criterion | marketing agency | ad tools | markov |
|---|---|---|---|
| up front | US$3,000+/mo retainer | US$97/mo | US$1,500 founding (US$4,500 standard), refundable |
| who runs it | account managers, business hours | you | agents, 24/7 · one operator on exceptions |
| trial page + tracking | extra | yours to build | built, verified, yours to keep |
| if it doesn’t work | you still pay | you still pay | you pay nothing |
| paid on | hours | seats | customers who stay |
| learns between rounds | whoever is on your account | no | every round, every seat |
The retainer is gone because the hours are gone. Agents do the work that used to be billed by the hour. The one person left is the one who signs off on your money.
Thirty seats. Then the doors close.
Thirty seats in the founding batch — enough to learn from, few enough that every seat gets the operator’s attention and the guarantee stays real. Applications close October 31, 2026.
Founding seat US$1,500. Standard seat, from November: US$4,500. You take the risk of being first; we take your results as proof.
who the seats are for right now
- - self-serve software — a stranger can try it alone, online
- - something the team can call — an API or SDK — so it runs the trial itself and knows when it breaks
- - at least 20 paying customers today
- - at least US$500 a round you are willing to put toward testing
what you connect, and where each connection stops
- ad account — the team drafts and manages campaigns. # drafting is not raising the budget
- payment + product events — the team confirms trials, payments and refunds. # reading is not refunding
- test environment or scoped API — the team runs the trial itself first. # the trial page is not your production
Apply for a seat.
Four questions. If it is a fit, you see the first plan within a working day — before you pay for anything.