The optimization engine for
AI-native advertising.
Benna is the learning layer that powers the Boost Boss network. On every impression inside an MCP-enabled application, Benna predicts conversion probability, selects the winning ad, and auto-optimizes bids in real time — using signals no legacy DSP can see.
Signals legacy DSPs never see.
Every bid request on the Boost Boss network carries a bb_context object populated by the host application's MCP layer. Benna learns from every dimension in real time — and none of it relies on cookies, device IDs, or third-party data.
Intent is a first-class feature.
On a news site, all you know is that a 34-year-old in Austin clicked "read more." On an MCP surface, we know they just asked Claude "why is my pytest failing on the CI runner" — and that's a completely different person to bid on.
Benna treats the prompt as a signal, the tool chain as a signal, and the conversational stage as a signal. All three together produce a prediction that is 12x more predictive of conversion than any cookie-based segment we benchmarked against.
Every signal is privacy-preserving: no PII leaves the host application. Embeddings are computed on-device where possible and transmitted as anonymous vectors to the Benna inference gateway.
A 3-stage pipeline, end-to-end in under 50ms.
Benna receives a bid request, encodes the context into a dense embedding, runs a two-tower ranking model over candidate ads, and returns a winning ad with a calibrated bid — all before the user's next token renders.
1 · Encode
bb_context → ctx_embed[512]
p99 < 3ms
2 · Predict
Two-tower ranking
p99 < 8ms
3 · Calibrate
Isotonic + floor check
p99 < 2ms
Measurable lift from the first impression.
Benna is benchmarked weekly against a holdout group served by rule-based bidding. Lift numbers below are median values across all campaigns on the network over the trailing 30 days.
Benna runs underneath all three Boost Boss surfaces.
You don't integrate Benna directly. It's the inference layer that powers the auction inside the Boost Boss ad network. Sign up for the surface you need and Benna comes on by default.
Boost Boss Ads
Benna auto-optimizes your bid on every impression. Set a target CPA or ROAS — Benna does the rest. No manual bid management required.
boostboss.ai/ads →Boost Boss Publish
Benna selects the highest-revenue ad for every impression on your app. No waterfall tuning. No mediation configs. Just maximum yield.
boostboss.ai/publish →Boost Boss Exchange
Benna annotates every OpenRTB bid request with AI-native context features, giving your DSP more signal than any cookie-based exchange provides.
boostboss.ai/exchange →How Benna learns.
Benna is trained on every bid, impression, click, and conversion that flows through the Boost Boss network. The ranking model is a two-tower retrieval architecture with a conversion head — a design that has become standard in high-throughput recommendation systems.
Context-aware bidding in AI-native applications
A technical note on how Benna models user intent inside MCP-enabled surfaces, including loss formulation, embedding strategy, and calibration approach. Released openly for the research community.
Get Benna on your next campaign.
Benna powers every decision on the Boost Boss network by default. To access it, sign up for an advertiser, publisher, or exchange seat — or reach out to the team for enterprise access.