Data‑Driven Film Marketing and Real‑Time Box Office Prediction

Author: Feng Xinping, Senior Data Technology Expert, Alibaba Pictures.

Film promotion is shifting from traditional offline channels to the Internet. The budget ratio of offline to online promotion has changed from a predominance of offline to roughly 1:2 (offline:online). Effective promotion lays a solid foundation for box‑office performance, helps word‑of‑mouth spread quickly, and penetrates third‑ and fourth‑tier markets.

Because film promotion involves long chains and many complex links, the data‑driven process faces many challenges.

At the 2019 Hangzhou Yunqi Conference "Smart Entertainment Technology" forum, Feng Xinping shared Alibaba Pictures' data‑driven promotion solution from a technical perspective.

1. The Internetization of the Film Industry

According to the 2018 data from the Film Industry Development Office, 250 million users purchase tickets online each year; 85 % of box‑office revenue comes from online sales, and 90 % of cinemas sell tickets online, indicating a high degree of internetization.

Based on this online foundation, we have accumulated user data (basic profiles, preferences, viewing paths, decision paths, trailer views, final ticket purchase) and film/cinema metadata, which are essential for box‑office forecasting.

However, high internetization does not automatically solve all problems. The promotion channels are fragmented, costs remain high, full‑link ticket data analysis is difficult, and cinema data reporting is often delayed, leading to unsynchronized box‑office figures.

Therefore, the talk focuses on two aspects: promotion and distribution .

2. Promotion – Channel Integration and Rapid Push Capability

We integrate promotion channels such as social media, new media platforms, and Douyin accounts to achieve fast, coordinated campaigns. Although channels are diverse, 85 % of tickets are still purchased online (e.g., via Taopiaopiao). Even with fragmented data sources, we can perform partial link analysis.

Example: For the film "Detective Chinatown 6", we first target existing high‑scoring audiences from "Detective Chinatown 5" for preview screenings, then guide core audiences identified during pre‑production, and finally analyze potential audiences using Alibaba's "Lighthouse" platform (features like "Big V Treasure" and "Channel Pass") to reach hundreds of millions of users within an hour and provide systematic data reports.

3. Distribution – Real‑Time Box‑Office Forecasting

Pre‑release box‑office forecasting is challenging due to many influencing factors (public opinion, hot topics, etc.). Our solution provides real‑time forecasts that calculate effective sessions, sold seats, and average ticket price.

We have direct connections with about 90 % of the 11,000 cinemas in China (through Taopiaopiao). The challenge is to extrapolate national box‑office figures from these directly linked cinemas.

Technical difficulties include:

Identifying valid sessions.

Recognizing true sales situations.

Estimating average ticket price.

We also need to filter out abnormal sessions such as overlapping sessions, "ghost" full houses, and private bookings. For example, overlapping sessions occur when the end time of one screening (22:39) conflicts with the start time of the next (22:30). Nationwide, there are up to 500,000 sessions per day during peak periods that must be screened for such anomalies.

Seat‑availability recognition is another challenge. Different ticketing systems report seat status differently (unavailable, sold, reserved, locked). Some cinema managers mark reserved seats as unavailable, which must be detected. We compare the platform’s seat status with authoritative national data and apply algorithms to resolve inconsistencies.

Average ticket price varies across cinemas due to infrastructure, promotional activities, and subsidies. We fuse three price sources—effective order price, schedule price, and historical official price—to fit the real average price.

Finally, we calibrate the direct‑link data (covering 90‑95 % of cinemas) to the whole market. Calibration coefficients depend on release periods and weekly cycles. Using time‑series analysis and seasonal models, we adjust forecasts for weekdays versus weekends.

Results: During the 2019 Spring Festival period, our forecast error was about 1 % for total box‑office, with top‑10 films averaging less than 3 % error. Larger‑scale films showed smaller errors, while smaller films had larger deviations. About 50 % of theaters had average ticket‑price errors within 5 %.

In summary, the data‑driven promotion process still faces challenges in data completeness and user‑path analysis, but our goal is to build a new intelligent infrastructure for the film industry that makes promotion effortless.

Join the Alibaba Entertainment Technology community by adding the "Entertainment Technology Assistant" on WeChat and providing your contact information.