How Airbnb Quantifies Listing Lifetime Value

Airbnb measures the value of each listing by estimating three metrics: baseline LTV (the total bookings a listing is expected to receive in the next 365 days), incremental LTV (baseline minus cannibalized value), and marketing‑induced incremental LTV (the lift generated by internal marketing projects).

Baseline LTV is defined as the expected number of bookings a listing will generate over the next year. To predict it, Airbnb snapshots listing features (availability, price, location, host tenure, etc.) at a reference time t, trains a machine‑learning model on those features, and uses the future 365‑day booking count as the label. The model’s predictions are discounted to present value using a financial discount rate.

Incremental LTV accounts for the two‑sided market effect where a new listing may steal bookings from existing ones. Airbnb distinguishes “incremental value” (new bookings that only occur because the listing is present) from “cannibalized value” (bookings shifted from other listings). Incremental LTV = baseline LTV – cannibalized value.

Marketing‑induced incremental LTV isolates the lift from internal initiatives, such as a campaign that gives hosts optimization advice. By comparing the estimated marketing‑induced LTV with the total cost of the campaign, Airbnb can assess ROI.

Challenge 1: Accurate Baseline LTV – The model must wait 365 days to observe the true label, and market shocks (e.g., the COVID‑19 pandemic) can render training data stale. Airbnb mitigated this by shortening the training window, adding fine‑grained geographic features and external factors (border closures, reopenings), and switching to LightGBM for high‑cardinality geographic variables.

Challenge 2: Measuring Incrementality – Since the true incremental label is unobservable, Airbnb models a production function linking total supply (S) and demand (D) to bookings (O). Historical supply, demand, and outcome data are aggregated into low‑overlap segments, allowing the model to estimate the marginal bookings generated by additional supply.

Challenge 3: Handling Uncertainty – Airbnb continuously updates LTV estimates as real bookings accrue. For example, if a listing was projected to receive 16 bookings by year‑end but has already reached 16 halfway through, the forecast is adjusted upward (e.g., to 21). This dynamic updating reduces bias caused by unforeseen shocks.

The framework enables several product decisions: identifying high‑value listing segments for new hosts, pinpointing regions where listings could benefit from additional demand, and evaluating which internal marketing projects deliver the greatest community value. Airbnb notes that the same methodology could extend to other experiences, where value depends on travel trends and guest discovery.

In summary, Airbnb’s LTV system combines baseline forecasts, incremental adjustments for cannibalization, and marketing‑driven lift, while addressing data latency, market volatility, and uncertainty through model engineering and continuous recalibration.