What “optimization” actually means
In mathematics, optimization means solving a function to find the point of highest expected revenue. That is what distinguishes an optimization model from the industry-standard base-price model: one adjusts prices logically, the other computes the mathematically best price.
The $250 seat or the $115 seat?
Consider a classic revenue-management decision: a seat with a 50% chance of selling at $250, versus a 90% chance at $115. Multiply probability by price and the answer flips intuition.
50% × $250 = $125 · 90% × $115 = $103.50
The first option yields higher expected revenue despite the lower booking probability. An optimization model reasons in exactly these terms; a rules engine does not.
Rules-based models: logical, but not optimal
Base-price models use if-then criteria — if season is X, change price by Y — rather than probability analysis. The adjustments are sensible, but they follow heuristics instead of identifying the highest expected-revenue point.
Why airlines went to rules — and came back
In the mid-2010s, some airlines shifted toward rules-based systems when probability estimates degraded amid low-cost carriers and capacity swings. But the majors kept optimization models with analyst support, and the industry eventually recognized that, set up correctly, optimization models simply make more money.
The single-listing problem
Vacation rentals face a binary booked-or-vacant reality, unlike hotels with many rooms. Historical data is messy — owner stays, maintenance blocks — and with one unit there is no chance to recover from a forecasting error. In that context, a well-built rules model can outperform a poorly configured optimization approach.
What Quibble forecasts instead: shopper choice
Rather than traditional demand forecasting, Quibble estimates the probability that a guest books your listing when it appears on their screen versus competitors. That shopper-choice model yields both revenue-maximizing pricing and competitive insight.
Finding the true maximum
An optimization model tests expected revenue at every possible price point, and the real challenge is finding the global maximum rather than a local one — accounting for how each price change shifts booking probability for both your listing and the comps.
The method: discrete choice
Quibble’s model is a discrete choice model, specifically a multinomial logit. Discrete choice theory was developed largely by Daniel McFadden, who received the 2000 Nobel Memorial Prize in Economic Sciences for theory and methods for analyzing discrete choice. The premise is simple: a shopper picks one option from a small set, and the chance they pick yours depends on the price, amenities, ratings, reviews and photos of every option in that set.
Expected revenue = price × probability of booking
Quibble sets each night’s price to maximize expected revenue, which is the price multiplied by the probability the night books at that price. Price is the output of the optimization, not an input you choose. The science of pricing is in estimating booking probability accurately; once it is known, choosing the price is arithmetic, the same $250 versus $115 comparison above repeated across every price for every night.
The consideration set of 10
Shoppers do not weigh hundreds of listings; they screen down to about 10. About 30 comps are attached to each listing, and for each night the 10 best matches that are still available go into the model. On the Comp Model tab a comp can be replaced only when it is flagged Needs revision (Quibble offers up to 5 candidates, and the replacement reprices on the next nightly run), and Comp Model → Simulator shows share of a night, the best comps and what-ifs.
Thin data, and market share as a probability
One listing rarely has enough history to estimate choice on its own, so Quibble predicts total demand across the small comp set and splits it by choice probability: expected bookings = set demand × your choice probability. Market share is that probability: the chance the next qualified shopper, someone who will book in the set, picks your listing. Price, amenities, ratings, reviews and photos all move it.
Choosing between a base-price model and an optimization model
There are two kinds of model. Rules-based models are if/then statements, such as “if the average market price drops on day X, drop mine by the same amount”; the user sets a base price and the model moves it up or down. Optimization models solve for the best price in two steps, a forecast step and an optimization step, and remove the need for a base price at all. What you actually pay for in revenue management is the model that makes the pricing decision, and the better the model, the fewer manual changes a revenue manager makes.
No model is perfect, because every one is predicting human behavior. Manual pricing is limited by time; a base-price model is limited by the human-set base it builds every future date on; an optimization model is limited by the data it can ingest and how well it is tuned. Some gains come before any model: simply pricing by season and day of week already moves RevPAR. Optimization is the step up when you want more nuanced decisions without maintaining base prices.
Go deeper: How base-price models work, Dynamic pricing isn’t optimized pricing, What is a blocked day worth?, Discrete choice experiments.