Author
TL;DR: A fair usage policy is boundedness retrofitted at the policy layer. Vendors write one when a plan sold as flat or unlimited sits on a metric with no natural ceiling, and the cost of that gap becomes a real number. The fix lives upstream in the licensing model: define the value metric, license a quantity of it, and price commitment past that quantity. Caps no account ever touches, and overage rights that never bill, are repricing evidence for the next renewal.
A fair usage policy is drafted by lawyers and read by almost nobody, but the work it does is commercial. When a software company sells a plan as unlimited and then reserves the right to throttle, upgrade, or renegotiate its heaviest accounts, the policy is doing work the pricing model declined to do. It answers the question every serious buyer eventually asks: how much of this can we consume at this price? That answer was moved out of the pricebook and into a document nobody opens until enforcement.
What a Fair Usage Policy Is in Software (and What It Is Not)
In B2B software, a fair usage policy (FUP) is a contractual cap attached to a plan that was sold without one. The plan says unlimited users, unlimited API calls, unlimited AI actions, or simply one flat price. The policy says: within reason. It defines, usually in deliberately general language, the point where the vendor may throttle performance, require a higher plan, or reopen the contract.
Two neighboring terms cause most of the confusion. Copyright law’s fair use doctrine governs when protected material can be reused without permission, and it has nothing to do with software pricing. Telecom fair usage policies manage congestion on shared networks, where one subscriber’s consumption degrades another’s service. A software FUP borrows the telecom name, but the exposure it manages is usually the vendor’s own cost rather than a neighbor’s experience.
Part of any software FUP is genuine abuse control: no resale, no scraping, no shared credentials. The rest is economics. It caps the consumption the price never accounted for.
The Defect Is Upstream: An Unbounded Metric Priced Flat
Flat-rate pricing keeps taking the blame for this exposure, and the blame lands one layer too high. A flat price is a price-point decision. Whether it is safe depends on the value metric underneath it, and specifically on whether that metric is bounded. A flat price over a bounded quantity, a tranche of named users or a defined volume of transactions, is sustainable by construction: the grant itself limits what the price must cover. A flat price over an unbounded metric, unlimited consumption for one fee, is an open liability with a subscription attached.
Under traditional software economics the liability rarely billed. Serving one more user or one more API call cost close to nothing, so unlimited was a marketing word with no cost underneath it. Generative AI changed the mechanism. LLM inference carries a real per-request cost, which puts a cost line under every unit of consumption, a shift we cover across AI software pricing. Agentic usage then removes the old natural ceiling: a human user is bounded by hours and attention, while an AI agent consumes on whatever loop it runs.
The fair usage policy is where the vendor concedes all of this. It is boundedness retrofitted at the policy layer, and it arrives with the defect intact: the metric inside the licensing model is still unbounded, and the policy only argues with the consequences.
Policy Caps vs Licensed Quantities
A cap can live in two places, and the two behave differently under pressure.
Why a Policy Cap Erodes
A policy cap is discovered at enforcement time. The account that trips it never priced it in, the rep who sold the deal never mentioned it, and the conversation now happens mid-term with relationship capital on the table. Under that pressure the cap is waived, and the waiver becomes the precedent the next account cites. Policy limits erode this way everywhere we see them, in discount floors as much as in usage caps.
What a Licensed Quantity Changes
A licensed quantity holds. When boundedness lives in the licensing model, the cap is part of the grant: this many named users, this volume of consumption, this pool of agentic actions. Each is stated the way per-seat grants state their assignment terms. The quantity is sized to the customer at the deal, priced as commitment, and expanded through a purchase rather than an argument. Larger commitments earn better unit pricing, which is how volume discounting rewards commitment instead of subsidizing drift.
Peer-reviewed work on cloud pricing points the same direction: committed structures sort buyers by how certain their demand is, and they outperform both pure flat-fee and pure usage designs.
Who Designs the Boundary
Procurement will build the boundary if the vendor does not. In our client work we watch enterprise procurement teams convert any flexible exposure into a negotiated fixed cap, and when the vendor never decided where that cap belongs, deal pressure decides it.
Move boundedness into the grant and the fair usage policy shrinks rather than disappears. What remains is the abuse clause, and nothing else.
Is Your Usage Cap a Policy Promise or a Licensed Quantity?
The difference between a cap discovered at enforcement time and one baked into licensed quantity determines how your pricing holds under pressure. Find out which structure your licensing, packaging, and pricing actually reflects.
Unexercised Overage Rights Are Repricing Evidence
An enforced cap produces revenue; an unenforced cap produces information, and most vendors throw that information away. Look at what the policy record says: accounts that never approach the cap, overage rights the vendor holds but has never billed, enforcement clauses that have never fired. Each one is a measurement. A customer consuming far beyond the plan’s intent, under a right the vendor never exercised, is demonstrating willingness to pay above the invoice, in conduct rather than in a survey answer.
That read should be systematic, not anecdotal. The evidence sits in the contract file already. Nobody has to run a study to collect it.
The wrong move is the back-bill. Retroactive enforcement converts years of accumulated evidence into one quarter of revenue and one burned relationship, and it teaches every other account to fear the meter. The right move is repricing at renewal: resize the licensed quantity to observed consumption and price the new commitment on the schedule. The evidence funds a price, not a penalty.
Tests to Run on Your Own Fair Usage Policy
Five questions expose where your fair usage policy is carrying a licensing decision. Ask them in order.
- If no account ever touches the cap, what is the cap measuring? A boundary nothing reaches was either priced too loosely or placed by guesswork.
- Which document answers “how much can we consume at this price,” the license grant or a policy page? The buyer will find the answer either way. Only one location lets you price it.
- If your heaviest account doubled its consumption next quarter, which line responds first: the invoice, the policy, or nothing? An architecture that answers “nothing” has already decided who absorbs the cost.
- Do you hold overage rights you have never exercised? Restraint is a datapoint. It says where the price should sit at the next renewal.
- Who owns the decision of where boundedness lives, the team that writes policies or the team that designs the licensing model? The answer predicts whether the fix holds.
Working the answers into an architecture is judgment work, per vendor and per metric. A fair usage policy that carries your whole boundedness story is a pricing decision already overdue. Talk to a pricing expert, describe the plan and the line the policy is currently holding, and a pricing architect reads where that boundary belongs in the grant.
