Not for sale.
At your service.
Infinima’s proprietary bidding engine, operated only by the people who built it.
No seats · No self-serve · Settled on CPI / CPA
The engine
Built for one method, not for the market.
Infinima’s work runs on finding small advantages and compounding them, and off-the-shelf platforms round those advantages away. So we built the bidder ourselves.
A performance bidder has one governing inequality. For every impression, the most it may pay is the outcome’s value, discounted by the probability that this impression produces it:
where p is the probability this impression ends in the outcome we are paid for, in practice the product of click and conversion probabilities, and V is the CPI or CPA payout. Estimate p too high and the engine pays for air; too low and it goes silent. On our terms the estimate is not a dashboard number, it is the price of being wrong.
Bid shading
First-price auctions punish honest bids. The engine learns clearing-price distributions from its own won and lost auctions, and shades toward the price the auction would actually have cleared at.
Conversion propensity
The p in the inequality. Models trained on our own auction and outcome data score every impression for the probability it ends in the install or action we are paid on. No platform-reported proxies: the training signal is what we measured.
Pacing
A budget spent by noon buys the morning’s prices. The engine allocates across the whole day’s price curve, because clearing prices move and patience is a margin.
Supply path
Short routes to inventory, with fewer fees inside them. Every intermediary is a percentage, and percentages are what we are in business to keep.
The loop
The loop is the product.
Every impression the engine buys produces a data point we measured ourselves: served, seen, clicked, installed, acted. Those outcomes retrain the models that price the next bid. Measure, bid, outcome, retrain; an engine that learns from its own ledger prices next week’s auctions better than it priced last week’s.
It is the same loop Infinima runs everywhere, compressed to auction speed. The measurement stack decides what counts as an outcome; the engine decides what an outcome is worth paying for; the gap between the two is where ε lives.
No seats
You cannot log in. That is the design.
There is no console to buy, no seat to license, no campaign to launch yourself. Every campaign on InfinimaX is operated by the people who wrote it, inside an Infinima engagement. When the builder runs the engine, quality carries one name.
The terms
Paid on outcomes, or not paid.
Platforms that sell seats earn on spend, whatever the spend buys. InfinimaX runs on CPI and CPA: the engine earns when the outcome happens, and not before.
We can price this way because measurement is the house discipline. Billing on outcomes is what confidence in numbers looks like.
Outcomes are counted where you count them. Your MMP’s attribution is the billing truth, our measurement runs alongside it, and for clients who want depth we reconcile both ways against your BI, number for number.
What we refuse to buy
The inventory we leave on the table.
Performance terms make junk unaffordable: when the engine only earns on outcomes, inventory that cannot produce them is a cost, not a bargain. A partial list of what the filters remove, each with its own room in our museum:
- Museum M.07 →
Made-for-advertising pages
Built to host ads, not readers. No outcomes live there.
- Museum M.06 →
Stacked slots
Ads sold in layers. Only the top one exists.
- Museum M.08 →
Unseen "viewable" impressions
Served is not seen, and we do not pay for served.
- Museum M.05 →
Stuffed pixels
Entire sites inside one pixel. Zero pixels, zero value.
- Museum M.09 →
Undeclared refresh inventory
Ninth-glance impressions at first-look prices.
The full collection is open to the public: The Museum of Ad Fraud →
Access
The way in is a conversation.
InfinimaX runs for clients of Infinima. If your growth plan has outcomes worth pricing, start there.