Intelligent Advertising in Real Estate: Challenges, Practices, and Insights from Beike

Beike, a leading real‑estate platform, introduced its intelligent advertising system that aims to maximize business goals such as MAU, GMV, and conversion rates. The system integrates a full‑stack DSP platform, smart bidding, incremental value estimation, user segmentation, and RTA strategy optimization.

The advertising workflow includes a seven‑layer funnel—from media request to conversion—requiring pre‑estimation of conversion rates at each stage. Multi‑task learning models (MMoE) predict combined click‑through and app‑launch rates (pCTDTR), while delayed‑feedback models (DFM) handle long conversion cycles typical of real‑estate transactions.

Key challenges identified are low repeat‑purchase ratios, long conversion cycles, fragmented media channels, and complex budget allocation across channels, plans, and units. To address these, Beike designed a budget smoothing mechanism and an ROI‑driven allocation framework that fits a monotonic increasing yet marginally diminishing curve.

For user targeting, Beike enhanced GEO targeting by fusing internal building‑dictionary data with external DMP/DSP signals, improving location accuracy beyond simple IP‑based methods.

Conversion‑rate modeling faces delayed label updates; Beike adopted a delayed‑feedback model that partitions labels into time buckets, enabling continuous model refresh and reducing bias caused by unobserved conversions.

Calibration techniques combine semi‑conversion signals (e.g., search and detail clicks) to adjust CTR predictions before full conversion, improving estimation accuracy.

RTA (Real‑Time Auction) strategies are evaluated through controlled experiments comparing various bidding perturbations, media direct‑buy, and different RTA algorithms, ultimately selecting the approach that yields the highest incremental ROI.

The presentation concludes with a summary of the intelligent advertising system, its technical components, and the measurable impact—over 10% improvement in conversion rates compared to baseline bidding.