Six dollars sounds like nothing
It does. It is less than a cleaning-supply line item. It is a rounding error on a Saturday in July.
Run it out anyway.
Six dollars a night is $2,190 a year on a single listing. Fifty listings is $109,500. Two hundred listings is $438,000. Four hundred is north of $875,000.
That is not a fee you are paying. It is revenue that was available on nights you already sold, at prices you already set, with a tool you are already paying for. Nobody sends you an invoice for it, which is exactly why it survives.
You would not keep a revenue manager who left $438,000 on the table. Most managers renew the tool that does.
Where the number comes from
We do not estimate this after the fact. We calculate it before anything changes.
Every account that comes to us from a base-price tool goes through the same step. We take the client’s live calendar — same listings, same nights, same demand, same comp set — and we compute two things for each night:
The expected revenue at the price of their current tool set.
The expected revenue at the price our model solves for.
Expected revenue is price multiplied by the probability the night books. A $400 Saturday with a 60% chance of selling is worth $240, not $400. That is the only honest way to compare two prices for the same night, because a higher price that does not book is worth nothing and a lower price that always books is leaving the rate behind.
Across those comparisons, the average difference is about $6 a night.
It is an average, so it moves. Tight urban markets with deep comp sets run thinner. Seasonal beach and mountain portfolios, where a single peak weekend is worth a slow month, run wider. Individual nights swing much harder than the average suggests — which brings us to why the gap exists at all.
Why the gap exists
Take one night. Base price $130, soft demand.
A base-price model answers one question: how should I adjust the base rate you gave me? So it adjusts. Add $10 for seasonality. Add $51 for the day of week. Add $25 for a local event. Apply a 1.66× market factor. Out comes $216.
Every one of those adjustments is defensible. The arithmetic is clean. But it is arithmetic performed on a number you chose yourself, and the model never asks whether the answer it produced is the best available price. It was not built to. It was built to move your base price in a sensible direction.
An optimization model answers a different question: what price earns the most tonight? To answer it, the model has to solve the whole curve — what happens to booking probability at $150, at $220, at $290, at $400 — and find where price multiplied by probability peaks.
On this night, that peak sits at $288, with an 80% chance of booking, for $231 in expected revenue. The base-price answer of $216 books more often and earns $202. The difference is $29 on that single night.
The base-price model is not tuned wrong. It was never looking for the peak.
That night is a demo, not a typical result. Real portfolios average $6 because most nights the two models land close, and the base-price answer is genuinely fine. The money is in the minority of nights where it is not — the peak weekends, the event dates, the sudden soft patches — and those nights are impossible to catch by hand and invisible to a model that is not solving for them.
“Expected revenue is just a model”
It is. And it is the right objection.
A pre-cutover calculation says what our model thinks the optimized price is worth. It does not prove a guest showed up and paid it. Any vendor can build a model that flatters itself, and plenty have. Projected uplift is the easiest number in this industry to manufacture.
So we grade the model against what actually happened.
We measure forecast error — the distance between the revenue our model expected and the revenue that booked. Inside a 30-day window, for clients live with us more than three months, that error runs under 5%.
Three months matters, because the model needs your booking history to calibrate against your listings rather than against the market in general. Thirty days matters because forecasting a specific night six months out is a genuinely harder problem than forecasting it four weeks out, and we would rather quote you the number we can stand behind than the one that sounds best.
Under 5% is what makes the $6 meaningful. Without it, the uplift calculation is a projection about a projection. With it, the expected revenue we compute is a reasonable proxy for the revenue you collect.
A pricing model you cannot audit is a guess with a dashboard.
What to ask your current vendor
Two questions.
First: what is your forecast error, over what window, on my portfolio? Not a case study, not an aggregate uplift claim across their whole book — your listings, measured against what actually booked.
Second: when your tool set last Saturday’s rate, what did it compare that rate against? If the answer is a base price and a set of adjustment rules, the tool is telling you where to move. It is not telling you where the money is.
You can measure this yourself, on your own calendar, before you change a thing. That is the whole point of doing the calculation before the cutover rather than arguing about it after.