Now onboarding launch partners # Every auction teaches the next. The DSP that never stops learning. Ailumic’s bidding models are built to learn from live auction outcomes as they happen, so each bid reflects what the market just taught. Generative AI turns a one-paragraph brief into a plan, creative and reports. [Become a launch partner](https://www.ailumic.com/contact/#launch-partners) [See how it learns](https://www.ailumic.com/#learning-loop) **Ailumic Inc.** builds a demand-side platform: software that buys ad impressions for advertisers in real-time auctions. 1. [01 Continuous learning — models update from live outcomes, not weekly retrains](https://www.ailumic.com/#learning-loop) 2. [02 Generative AI — from a one-paragraph brief to plan, creative and report](https://www.ailumic.com/#brief-to-bid) 3. [03 The auction’s clock — a C++ core with privacy checks in the request path](https://www.ailumic.com/#auction-clock) Continuous learning ## Minutes, not weeks. Most bidders learn in batches: train, deploy, wait a week. Ailumic’s models are built to keep learning. Every outcome — a win, a loss, a click, a conversion — flows back into the model within minutes, and the next bid is priced with it. 1. **Observe.** The request: page or app, format, supply path, consent. 2. **Predict.** What this impression is worth to this advertiser’s goal. 3. **Bid.** A price shaded to what the auction is likely to clear at. 4. **Learn.** The outcome updates the model, and the loop closes. One bid, start to finish Illustrative 1. 00:00.000 request Exchange A · 300×250 · news / sports · US 2. 00:00.004 predict p(click) 1.2% · p(trial | click) 3.1% · worth $0.94 CPM 3. 00:00.006 bid $0.71 CPM, shaded from $0.94 4. 00:00.112 outcome Won at $0.66 · impression served 5. 00:42:10 event Free trial started · attributed 6. 00:47:00 learn Model updated with 18,402 new outcomes Generative AI ## From brief to bid. Describe the goal the way you would to a colleague. Ailumic drafts the plan — inventory, bidding, pacing — proposes creative variants, and reports results in plain language. You approve every step, and every creative before it runs. Brief → plan Illustrative example Your brief Launch a four-week campaign for our fitness app’s free trial in the US. Keep the cost per trial under $25. Mobile first. Drafted plan · awaiting your approval **Goal** : Free-trial starts · cost per trial under $25 **Inventory** : Mobile web and in-app · United States **Bidding** : Value-based, shaded to the clearing price **Pacing** : Even over 4 weeks Creative variants - A Your first workout is free. - B 20 minutes. No gym. - C Start today. Cancel anytime. week 1 Cost per trial $21.80. Variant B leads; A is paused. Built for real time ## Built for the auction’s clock. An exchange gives a bidder a fraction of a second to answer. Ailumic’s bidding core is written in C++ and designed to decide in a small slice of that window, with the consent, supply-chain and brand-safety checks inside the same request — not in a separate pass afterwards. One auction, 0 → 100 msDesign target **Typical exchange deadline** : ≈ 100 ms **Ailumic’s decision, inside our servers** : ≤ 10 ms target **Network, there and back** : varies by region **Checks in that same request** : consent · supply chain · brand safety Privacy in the bid path ## Privacy is checked where the bid is made. This website, measured **cookies** : 0 **trackers** : 0 **third-party requests** : 0 [Privacy & trust](https://www.ailumic.com/trust/) [Opt out of interest-based ads](https://www.ailumic.com/opt-out/) - ### Signals honoured Global Privacy Control, GPP strings and US state opt-outs are read in each request and respected. - ### No sensitive targeting No targeting on health, religion, sexuality, political views, precise location, or anyone under 18. - ### Supply you can trace `ads.txt`, `sellers.json` and the supply-chain object are checked before we bid. - ### Data on a clock Raw logs 90 days, detailed records 13 months, then de-identified aggregates only. Where we are ## Built in the open, launched in steps. Ailumic is not buying media yet. Here is what exists and what comes next. ### Built - A C++ serving runtime and release system, running in Google Cloud regions - Operations consoles for logs, metrics, releases and datasets - Pipelines that compute and serve the features models learn from ### Next - Exchange integrations over OpenRTB 2.6 - First campaigns with launch partners - Generative planning and creative in the advertiser console Who’s behind Ailumic ## Write to the people building it. **Company** : Ailumic Inc. **Email** : [support@ailumic.com](mailto:support@ailumic.com) [Company facts and contacts](https://www.ailumic.com/company/) - Advertisers & agencies [Become a launch partner](https://www.ailumic.com/contact/#launch-partners) - Exchanges & publishers [How we buy](https://www.ailumic.com/partners/) - Saw an ad from us? [Opt out of interest-based ads](https://www.ailumic.com/opt-out/) Bid with what the market just taught. [Become a launch partner](https://www.ailumic.com/contact/#launch-partners)