Cost Per Use: The Subscription Math That Changes Decisions

A subscription is advertised as a flat fee, but it is consumed per use. The two numbers describe the same charge and tell different stories. Twelve dollars a month spent on a service opened twice is six dollars an opening. The same twelve dollars spread across twenty openings is sixty cents. Neither number is automatically good or bad. What matters is that the flat price is the one every renewal email shows you, and the per-use price is the one nobody calculates unless they choose to.

This guide does three things. It summarises what published research says about how people misjudge flat-rate deals. It shows where the major services keep the usage records you need for the arithmetic, so the count is real rather than remembered. And it gives a worksheet and decision rules that fit a subscription that is used unevenly, or not meant to be “used” at all.

Correction, September 26, 2026: an earlier version of this article was written as a personal account of running this calculation and cancelling on the strength of it. No records support that account, and it has been removed. Everything below is drawn from published research and vendor documentation, and every worked figure is either sourced or labelled hypothetical.

A receipt being divided over a calendar month, each small slice of the month marked with a tiny per-use cost Illustration only; not a record of any account.

What the research says about paying flat rates

The flat-rate bias is one of the better-documented findings in pricing research: people choose a flat fee even when a pay-per-use option would have cost them less, and they do it persistently. The studies below are the ones the rest of this article leans on. Sources checked September 26, 2026.

Source	What was measured	Finding relevant to cost per use	Limits
DellaVigna & Malmendier, “Paying Not to Go to the Gym” (American Economic Review 96(3), 2006)	Contract choice and attendance of 7,752 members of three US health clubs over three years	Members on a monthly contract of over $70 attended about 4.3 times a month, “a price per expected visit of more than $17, even though they could pay $10 per visit using a 10-visit pass.” They forgo about $600 in savings over a membership on average	Gyms, not software; US data from the late 1990s and early 2000s
Lambrecht & Skiera, “Paying Too Much and Being Happy About It” (Journal of Marketing Research 43(2), 2006)	Billing data for 10,882 customers of a European internet provider for up to five months, plus two surveys	Judged on three consecutive months of usage, 48.1% of customers on the flat-rate tariff would have paid less on a metered one. More than half of those with a flat-rate bias paid at least 100% more than the cheapest tariff would have cost. The authors attribute the bias to an “insurance effect,” a “taxi meter effect” and an “overestimation effect”	Internet access tariffs, 2000s; the effect sizes are for that market
Gourville, “Pennies-a-Day” (Journal of Consumer Research 24(4), 1998)	Laboratory studies of how reframing one cost as a stream of small costs changes acceptance	Framing an expense as a small recurring amount makes people compare it with other small ongoing expenses rather than with large one-off purchases, and raises compliance	Lab studies; explains why “only $12 a month” feels harmless, not how much any individual overpays
C+R Research, subscription spending survey (2022)	1,000 US consumers surveyed April 22 to May 2, 2022	Respondents first estimated $86 a month in subscriptions; the itemised figure averaged $219. 42% said they had stopped using a service but forgot they were still paying, and 74% said recurring charges were easy to forget	Self-reported, single survey, one vendor’s research blog
Deloitte, 2026 Digital Media Trends	3,575 US consumers surveyed October and November 2025	Households average four paid streaming video services at about $71 a month combined; 41% had cancelled a streaming service in the previous six months	Streaming video only; US sample

Two things follow from that table. First, the overestimation that Lambrecht and Skiera identify is exactly the failure this calculation targets: people predict more use than they will have, so the flat fee looks cheap per use at sign-up and stays uncorrected afterwards. Second, the C+R gap between the guessed and itemised totals says the problem starts earlier than the division. You cannot price a subscription per use if it is not on your list at all, which is why finding every recurring charge comes before any of this arithmetic.

The arithmetic, with the inputs labelled

The method is one division: monthly price divided by the number of times the service was actually used in that month. The prices below are the current US list prices shown on the vendors’ own pages on September 26, 2026. The usage counts are hypothetical, chosen to show how fast the per-use figure moves. Displayed prices vary by country, billing channel and promotion, so substitute your own statement line.

Service and plan (US list price, source)	Used 1 time	2 times	5 times	12 times	20 times
Spotify Premium Individual, $12.99 a month	$12.99	$6.50	$2.60	$1.08	$0.65
Apple TV+, $14.99 a month	$14.99	$7.50	$3.00	$1.25	$0.75
Peloton Strength+, $9.99 a month	$9.99	$5.00	$2.00	$0.83	$0.50
Any service at Deloitte’s average of about $17.75 (the $71 for four services above, divided by four)	$17.75	$8.88	$3.55	$1.48	$0.89

Figures are rounded to the cent. Zero uses is not a row, because the per-use price of an unused subscription is undefined: the whole monthly fee bought nothing, which is a clearer statement than any number.

Read the table across and the shape is the same for every price. Going from one use to two halves the per-use cost; going from twelve to twenty saves a few dozen cents. The decision the number informs is therefore a low-usage decision. At weekly use or better, almost any consumer subscription is cheap per use and the question becomes whether you wanted it at all. At two or three uses a month, the per-use price is in the range where a single one-off purchase or rental usually costs less, and that is the comparison the next section makes.

Where the real usage count lives

The step that separates this calculation from a guess is counting from a record instead of from memory. Lambrecht and Skiera’s overestimation effect operates on the memory. Most services keep the record for you, and the table shows where. Menu paths quoted from the vendors’ help pages, checked September 26, 2026.

Service	Where the usage record is	What it contains	Source
Netflix	Account page, Profiles, choose a profile, Viewing activity; “Download all” at the bottom of the page produces a CSV	“The TV shows and movies that have been watched on each profile on your account”, with dates	Netflix: viewing activity
Spotify	Account privacy settings, request Account data (streaming history for the past year) or Extended streaming history (the lifetime of the account)	Per-play records with timestamps, platform and how long each item played	Spotify: understanding my data
YouTube and YouTube Premium	myactivity.google.com, YouTube History, Manage history; export via Google Takeout	Every video watched, with date and time	YouTube: watch history, Google Takeout
Peloton App	Profile in the app, History	Completed workouts by date	Peloton App support
Any app subscription on Android	Settings, Digital Wellbeing	“What apps you’ve had on screen and for how long” and “how often you’ve unlocked your device and opened certain apps”	Android: Digital Wellbeing
Any app subscription on iPhone	Settings, Screen Time	Time per app and per category, daily and weekly	Apple: Screen Time

A practical note on what counts as a use. For a streaming or music service, count sessions, not items: an evening with three episodes is one use. For a fitness app, count completed workouts. For a phone app, Digital Wellbeing’s “times opened” is a reasonable proxy, though a two-second open that was dismissed is not a use in any meaningful sense, so pair it with the time-on-screen figure. The point is a consistent definition applied to a record, not precision.

A small card labelled ‘once a year’ with a calculator showing a cost that rises as the counter stays at zero Illustration only; the counter is not a measurement.

Rule one: compare with the one-off price

The per-use figure is only useful against a benchmark, and the natural benchmark is what the same use would cost bought singly. A film rental, a single class, a one-month subscription taken only for the month it is needed, a pay-per-use tariff where one exists. DellaVigna and Malmendier’s gym members had exactly this comparison available: a 10-visit pass at $10 a visit against a monthly fee working out at more than $17 a visit. The rule is then mechanical. If your per-use price is above the one-off price, the subscription is costing you money for the convenience of not deciding, and cancelling and re-buying as needed is cheaper.

The rule has one honest cost. Re-buying means re-deciding, and some services make returning expensive or awkward. Cancellation and refund terms differ sharply between vendors, and a few reset your data or your promotional price on the way out. Check what stops when payment stops before treating “rebuy later” as free.

Rule two: judge bursty use over a year

A single month’s arithmetic is unfair to anything used in bursts. A tax-preparation tool used hard for one month, a travel app opened daily on a trip and never in between, a language app used in January and again in September. For these, run the same division over twelve months: the annual cost divided by the year’s total uses. Using the Spotify list price as an example, twelve months at $12.99 is $155.88; if the year’s record shows thirty listening sessions, the per-use cost is $5.20, and thirty sessions concentrated in two months looks very different from thirty sessions spread across twelve. The annual frame rescues some subscriptions that a bad month would wrongly condemn. It does nothing for the ones that show zero across the whole year, which have no defence left.

The annual frame also collides with the annual plan. Prepaying a year lowers the monthly sticker but commits you before the usage record exists, which is the precise moment the overestimation effect is strongest. The annual discount behaves like a loan against use you may not spend, and the per-use calculation is the only way to check whether it paid back. Deloitte’s finding that 41% of streaming subscribers had cancelled a service inside six months is a reminder of how short the real horizon can be against a twelve-month commitment.

A twelve-month calendar with a chain wrapped around the first month, a pen hesitating above it Illustration only.

Rule three: do not apply it to protection

Some subscriptions are not for using. Insurance, a backup service, a password manager, a security product. Their value is the absence of a bad event, so a use count of zero is the good outcome, and cost per use produces the wrong answer every time. Judge these by a different question: what happens on the day you need it, and what would recovery cost without it. The same caution applies, in a weaker form, to the just-in-case subscription, where the real question is what you lose by cancelling rather than how often you opened the app. Lambrecht and Skiera call the desire to avoid a variable bill the “insurance effect” and count it among the causes of flat-rate bias, but for a genuine protection product the insurance is the product, not a bias.

Watch the trend, not the snapshot

A single per-use figure is a snapshot, and snapshots are easy to argue with. The stronger signal is the direction over three or four months. A service that cost $2 a use, then $3, then $5, with the price unchanged, is being abandoned slowly, and the falling denominator is the fact worth acting on before the count reaches zero. The usage records in the table above make this a two-minute check: the same download, filtered by month.

This is also why small subscriptions are the hardest to cancel. Gourville’s pennies-a-day framing means a $9.99 charge is compared with a coffee rather than with the $120 a year it adds up to, and a slowly falling use count never produces a moment that feels like a decision. Writing the per-use figure down monthly manufactures that moment.

A line on a graph steadily declining toward zero while a flat horizontal price line stays level above it Illustration only; the chart shows the pattern, not data.

Worksheet

One row per subscription, filled from statements and the usage records above. The first row is an illustrative example using a real list price and an invented count; it is not a recommendation about the service.

Subscription and plan	Monthly price	Uses last month (from record)	Cost per use	One-off price for the same use	Uses last 12 months	Annual cost per use	Trend over 3 months	Protection product?	Decision and review date
Example: Apple TV+, $14.99 (list)	$14.99	2 sessions (hypothetical)	$7.50	One month’s subscription bought only in a month with something to watch, $14.99	30 (hypothetical)	$6.00	Falling: 5, 3, 2	No	Cancel; rebuy for a specific month. Review January

How to fill it in:

List every recurring charge from three months of statements. Do not work from memory; the C+R gap above is the size of what memory misses.
For each, open the vendor’s usage record and count last month’s uses under one definition (sessions, workouts, opens).
Divide. Write the figure down even when it is embarrassing.
Find the one-off price for the same thing. If none exists, use one month of the subscription bought only when needed.
For anything used in bursts, add the twelve-month figure.
Mark protection products and skip the comparison for them.
Note the trend from the last three months’ counts.
Record a decision with a date, including “keep” decisions. A keep with a reason is a decision; a keep by default is the flat-rate bias in action.

Sources

All checked September 26, 2026. Prices are US list prices as displayed on the vendors’ pages from this location on that date; regional pricing, bundles and promotions differ.