Your markets make every price prove itself against what actually happened. We do that for spend.
We price each media decision the same way, then let the outcome correct the price. Every cycle it gets sharper.
Where the money gets decided today
A dozen partners, one pot. AppsFlyer counts the deposits; which partner produced them is the open question.
Today the money moves on flight‑wide averages. We price each partner separately, and what the next dollar there will buy.
Which partner actually produced the deposits
Numbers here are illustrative; the screens are the week‑one build.
Precise addsmarginal CPA and headroom per partner, beside the blended numbers you already readWhy it mattersblended CPA says who looked good; marginal CPA says who earns the next dollar. Same logs, one trace, different question
What you see todayThe Trade Desk looks cheapest at $226 per deposit, averaged over the whole flight.
What you see with usThe Trade Desk's next deposit costs $268. DSP D's costs $540, even though its average looks like $419. Same money, half the price. The Trade Desk can take about $400k more before that $268 starts climbing.
tracing each deposit back to the spend that produced it, so the next dollar is priced instead of averaged
This is the same report you read every week. Your columns, your flight, your partners. The three teal columns are the only thing we added.
It updates as your logs come in, day after day.
And the same pricing goes below the partner: the data segments you buy, and the suppliers underneath each buy. Which segments earn their price, which supply paths deliver, where the next dollar goes at that level too.
polymarket × preciseWorld Cup flight · updating as logs landliverecommend mode · activation on your authority
Flight: Jun 11 – Jul 19Priced to the placement: $3.95M of $4.42M · 89%First deposits: 17,090Blended CPA: $259Outcome: first deposit · 7‑day window
Partner
Channel
Spend
First deposits
Blended CPA
Marginal CPA
Band (±)
Room before the price climbs
The Trade Desk · CTV
CTV
$1,610,000
7,120
$226
$268
± $31
$400k+
StackAdapt · CTV
CTV
$940,000
3,410
$276
$342
± $44
about $150k
DSP C · video
OLV
$780,000
3,890
$201
$219
± $26
$300k+
DSP D · display
Display
$620,000
1,480
$419
$540
± $88
already climbing
Creator lane
Social
$470,000
1,190
$395
by segment
—
segment, creator
Flight
$4,420,000
17,090
$259
$297 weighted
Precise · what traces, and what does not
89% of spend prices cleanly down to the individual placement. Social reports by campaign, ad set, audience segment and expansion, so the creator lane prices at those levels today, and those segments are exactly what a mix model wants. With post‑level tagging in place it prices per creator too, the same as everything else on the page. Marginal CPA is the fitted slope of traced deposits on spend at each partner's current delivery: the price of the next deposit, not the average of the last one, estimated from within‑partner day and geo variation. That variation is not experimental, so this is an observational response curve rather than measured lift, and the band is widened for it. The band is an 80% interval on next‑period realized CPA at the same delivery: we expect to land outside it about one time in five, and it is what the grading board scores. Marginal running above blended is what a diminishing‑returns fit implies, not a law we are asserting; the question is simply where the next deposit is cheapest, and how much you can move there before that price rises.
Board 03Your columns on the left, exactly as AppsFlyer and the DSP report them. The teal columns exist only because the join exists.
What you get from us each week
A running list of moves, refreshed as the numbers move. Each one specific enough to act on, or to decline.
Right now the next deposit costs $219 at your best partner and $540 at your worst. That spread is the whole opportunity, and it moves every day.
Precise addsranked moves with expected results and reasons, and a stated refusal when the math cannot support oneWhy it mattersa recommendation you can act on has a source, a destination, an amount, a prediction, and a band
What you get todayA direction: lean into CTV.
What you get with usMove $180k out of DSP D display into The Trade Desk CTV, where the same money buys deposits at $268 instead of $540. We watch it from the moment it moves, and if the price drifts we say so and move again. When the math cannot support a move, the row says HELD.
every recommendation says where the money comes from, where it goes, how much, what we expect back, and how sure we are, or it says we will not guess
Every line is a move you approve or decline. Nothing happens until you say so; when you grant the authority, an approved move activates against the platform, and the record captures who approved it.
The list also sends itself: the same report, into the same places your reporting already goes, on your cadence.
That is the whole point: decision velocity. Evidence in, decision made, money moved, lesson kept. No week of assembly in between.
polymarket × preciserecommended moves · awaiting your approvalliveday 12 of 28 · activation on your authority
Move
Cut
Fund
Amount
What the same money buys
Confidence
Reason
State
01
DSP D display
TTD CTV prime
$180,000
$540 → $268 per deposit same money, roughly twice as many: +200 to +470
high
display’s next deposit costs twice CTV’s, and CTV can absorb this much at its price
PROPOSED
02
StackAdapt late‑night
DSP C video
$120,000
$342 → $219 per deposit same money, +85 to +310
high
same reach costs $123 less per deposit next door
PROPOSED
03
—
TTD CTV sports
$90,000
capacity review
—
we have never seen this inventory deliver that much; funding it would be a guess
HELD
Precise · what happens to a row
These two moves touch $300,000 of a $4.4M flight, and they are the two we would stake a number on today. The point is not the two: it is that the price of the next deposit runs from $219 to $540 across your partners right now, and every day that spread sits unpriced, budget sits on the wrong side of it. On the season budgets you are running, a few points of efficiency is an eight‑figure number. The gain side is priced across the whole slice being added, not at today’s marginal, because cost rises as you spend into a partner, and both moves stay inside the headroom we can see. You approve or decline; either way the decision is written down completely (what was known, what was chosen, what we predicted, what happened, what it taught the system) and graded when deposits mature. Row 03 stays held until capacity is observed, and that refusal is the reason rows 01 and 02 deserve trust.
Board 04Three moves, one of them a refusal. You approve or decline each one; we are never moving your money.
Paths: the best route per ad type
Precise addsa continuously recalibrated route through your channels per ad type, where MMM gives one quarterly mix for everythingWhy it mattersthe brand film and the referral code should not buy the same surfaces; pricing the path per ad type is the question your ad‑type taxonomy can finally answer
What MMM tells you todayOne channel mix across CTV, video, display, social, OOH and creator, refit quarterly for the whole account.
What we tell youA route through those same channels for each kind of ad, updated as evidence comes in. The in‑match :30 runs live sports CTV where you are eligible, then retargets on social, at $238 a deposit. The creator lane prices by segment, by expansion, and by creator.
which channels each kind of ad should run through, updated as evidence comes in rather than once a quarter for everything at once
polymarket × precisepath pricing · by ad type · liveliverepriced as evidence lands
Ad type
Channels today
Priced route across channels
Next dollar
Traceable today?
In‑match :30
Live sports CTV, national
live sports CTV in eligible states → social retarget
$238 ± 29
yes
Brand film :60
Premium CTV, OOH
premium CTV → social video → branded search
$310 ± 61
partly: OOH untraced
Performance :15
Programmatic video, display
OLV in‑app → display recapture
$205 ± 22
yes
Creator post
Creator and influencer
by segment, expansion and creator
—
by segment
Referral code
Sports media syndication
syndication → code redemption, direct
$121 ± 9
yes
Precise · where we stop
We price a route at whatever level the money can be followed, and the last column says which level that is. Where a leg cannot be followed yet, the row says so and we leave it unpriced. Saying where we stop is what makes the priced rows worth acting on.
Board 05Paths per ad type. MMM answers what mix, quarterly, for the whole account; this answers which route, day by day, for each creative.
How it gets sharper
Every number we hand you gets checked against what actually happened, and the difference goes into the next one.
That runs at every level at once, day after day, so the estimates tighten as the season goes on.
polymarket × preciseweek 4 of the seasonliveevery level, every cycle
Partner
Our prediction (80%)
Baseline: last cycle
Realized
Inside our range?
Beat the baseline?
How it was tested
TTD CTV
$268 ± 31
$243
$259
yes
yes, by $16
we watched the data
DSP C video
$219 ± 26
$208
$231
yes
yes, by $11
we watched the data
DSP D display
$540 ± 88
$466
$497
yes
no, by $12
we watched the data
Holdout partner 1
$284 ± 34
$310
$291
yes
yes, by $19
held out of fitting
Holdout partner 2
$198 ± 24
$226
$243
no, we were low
no, by $28
held out of fitting
Precise · the miss stays on the board
The baseline column is the bar: carrying last cycle’s cost forward is what you would have guessed without us, and we only claim credit where we beat it. Three of five here, and the two we lost are on the board with their causes named. Each of those gaps goes into the next set of numbers.
Board 06The grading board: what this looks like after one cycle. Numbers are invented; the mechanism is the offer. The miss sits beside the wins on purpose.
Doors: what happens after the click
DoorsThe buy stops at the deposit today. Doors carries it through to what the customer turned out to be worth, and prices the next buy on that.
AppsFlyer gives you the two endpoints: the install and the deposit. Today, every partner is judged on those two events.
Doors fills in the journey between: onboarding, first trade, category, whether they came back.
More detail is what the pricing runs on. One signal ranks partners. Twenty tells you which creative, audience and platform actually produced value, while the flight is still live.
Then the buy changes: fund the customers who turn out to be worth it, stop funding the ones who never trade.
And if there is appetite, the same machinery goes well past media, inside Polymarket and inside Publicis.
What we want to learn on this call
Q1What does the app instrument today around onboarding, discovery, and first trade, and who owns that roadmap?
Q2If a lightweight SDK like Doors sat beside AppsFlyer, which behavior events would you most want feeding media decisions?
Q3Where do app and market events live today, and could a sample of them be traced against the logs you send us?
Q4Which downstream outcome is the real north star for ad spend: first deposit, first trade, seven‑day return, or trading quality?
Q5Who says yes to an SDK on one surface, onboarding being the natural first, and what would make it a no?
Q6Which growth decision are you making this month that we should put on the record and settle together?
Next steps
All we need to start is logs. One flight that already ran, whichever one you pick.
1AppsFlyer export for that flight
2One DSP's delivery and cost file
3A note on how the two line up, and what counts as a first deposit
Sample rows are enough to begin, and your systems stay exactly as they are: no re‑tagging, no integration, no access.
We come back with what traced, what did not, and where the next dollar should have gone.
What happens next
1Logs landOne flight, sample rows first
2Coverage readWhat traces, what does not
3What to changeRanked moves, with ranges
4It runs liveChecked and repriced as results land
The World Cup retrospective teaches the NFL pattern before a dollar moves.
Then the season runs, and the instrument runs with it: predictions re‑priced, weights re‑bet, allocations adjusted in real time as the evidence lands.
We pointed it at your markets. It refused to find an edge.
For fun, and to see whether our own machinery holds up, we pointed it at your public markets: snapshots of every open market, strategies that each had to pass a test written before they ran, and a paper book that scores itself.
Results so far: down $31.72. Every strategy failed the test it set for itself, and the error bars swallow whatever is left. We are publishing that, which is the point. Anyone can show you their wins; the losses are what tell you the measurement is real.
It cannot actually trade: no wallet, no signer, on purpose. Losing thirty‑one dollars on paper is the most we are willing to risk on our own opinions. But it means our machinery already speaks your market's language: predictions sealed before the outcome, payoffs graded at resolution, markouts, refusals.
