Author
TL;DR: A trial license, a pilot, and a proof of concept are three different license grants, and most vendors give whichever one the deal produced rather than the one they chose. An evaluation grant is a duration decision with two riders stapled to it: assignment, who inside the customer holds it, and boundedness, whether the evaluation is capped by time, quantity, capability, or nothing. The riders decide whether it converts. Freemium is a packaging decision about what is permanently free; an evaluation grant is a licensing decision about what is temporarily granted and how it ends. With generative capability in the product, an uncapped evaluation becomes a cost commitment filed under marketing.
A sales engineer stands up a proof of concept in the second week of a quarter. By week six the prospect has three departments inside it, a connector wired into a workflow that runs overnight, and a request to extend past their fiscal year end. Finance is looking at an inference line nobody assigned to an account, because there is no account. None of it was decided. All of it was produced by terms the company never wrote down.
A trial license is a temporary license grant that conveys use of software for evaluation, for a defined period, on terms that end. It names who may use the software, where, for how long, how much they may consume, and what happens when the term ends. Those are the four dimensions a software license grant always settles: assignment, deployment, duration, boundedness. An evaluation answers all four, and when nobody answers deliberately it answers by inheritance.
This is not the freemium argument. Freemium is a packaging decision about what is permanently free, covered in when it works and when it does not and what to do when a free edition’s economics stop working. An evaluation grant decides what is temporarily granted, to whom, and on what terms it ends.
Three Things Called a Trial License, and They Are Different Grants
The self-serve trial. The grant runs to whoever completed a form, frequently someone with no budget and no authority to sign. Deployment stays in your environment. Duration starts at a signup event and ends on a clock. Boundedness is inherited rather than chosen: the trial account carries the paid edition’s limits with the payment removed.
The sales-led pilot. The grant runs to a named team inside a named account, with your people in it. The customer selected that population, which means the customer decided what the evaluation would prove. Duration begins at a kickoff and drifts, because a kickoff is a meeting and meetings move. Boundedness lives in a scope document no system enforces.
The enterprise proof of concept. The grant is scoped to a stated question: does this work against our data, in our environment, at our volume. Deployment frequently moves toward the customer’s infrastructure or their production data. Duration answers to their procurement calendar rather than yours. Boundedness is the hard one, because running at real volume is the entire purpose.
Those differ on every grant dimension, not on length. Hand the same paper to all three and you convey three different things while calling them one.
A pilot in this sense means a product evaluation. A pricing pilot is the separate exercise of introducing new pricing to a subset of customers before committing to it.
What an Unmanaged Pilot Cost Us
I ran a software company through this in the late nineties. A large building-supply chain with locations across the central US wanted to try our new cloud edition, because it needed to be proven. Weeks became months. The rollout kept extending on the argument that they had to see the aggregated view before they could centralize their sales function. Usage grew, and our costs grew with it, storage being the expensive one then, along with a steady line of change requests. The team was thrilled about landing a customer that size. When conversion time came, they decided not to proceed.
A year had passed on a poorly managed free pilot. The lost sale was the smaller half of it. We had reprioritized the product roadmap around what we believed was an imminent close and a happy customer, so we lost the roadmap year too. That is the cost nobody books against an evaluation: not the compute, the direction.
The Evaluation Grant Is a Duration Decision With Two Riders
Nearly the whole public conversation about trials is a conversation about length, and length is the least productive place to spend the argument. Peer-reviewed field experiments find that a longer window raises how many people begin an evaluation at all, because length reads as vendor confidence. It does little for conversion inside the window: what extra time adds in product learning it subtracts in urgency.
Modelling work treats the trial as an instrument for reducing buyer uncertainty. Its value falls as a product’s worth comes more from how many other organizations are on it than from what one buyer can learn alone.
Rider one: assignment. Who inside the customer holds the grant, and whether the population that evaluates resembles the population that will buy. A trial held by one engineer proves the product works for one engineer; if the purchase is enterprise-wide, the deal has to prove it again in a room that engineer does not sit in. Spread the evaluation generously across departments and the failure inverts: the paid grant has to withdraw access people already have, which is a harder first commercial conversation than a boundary would have been.
Rider two: boundedness. Whether the evaluation has a ceiling, and on what. Time is one instrument, quantity another, capability a third. Nothing at all is the fourth, and across the patterns our library holds it is the one we see most. Time-capping is the default because it is the easiest to administer and the least informative boundary available: a calendar reports nothing about how much of the product was consumed. A quantity cap is the only one that reports anything about cost, which is why it stopped being optional once products started carrying meters. Where each boundary lands is architecture work specific to what you sell, and it belongs to the first of the three decisions behind every software price.
Where Does Your Pricing Architecture Actually Stand?
A few questions return your pricing architecture score and show which of your licensing, packaging, and pricing decisions needs attention first. Real diagnosis, not a mailing-list toll.
What the Grant Should Say About How It Ends
An evaluation grant is the only license grant most companies write that is designed to expire, and the expiry is the part nobody designs. Leave the ending unstated and one of three outcomes arrives.
Three Outcomes of an Unstated Ending
Silent continuation. The date passes and nothing happens, because nothing was configured to happen and nobody owned the date. Access continues, and if there is a meter behind the product you are paying to serve an unpaid account. A temporary grant becomes indefinite through an unread calendar entry.
Silent continuation is not confined to evaluations. The same failure runs inside paid accounts wherever a commitment is priced off an estimate nobody reconciles. A customer projects a volume, earns the rate that volume deserves, and consumes something well below it. If the systems only surface the gap at year end, the conversation repeats on renewal: the customer forecasts higher volume again, holds the better rate again, and falls short again. Both cases share one defect. A date or a number was written into an agreement, and no system and no person was made responsible for reading it back.
A renegotiation nobody scheduled. The date arrives, access stops or is threatened, and the commercial conversation that follows is timed by an expiry clock rather than by either party’s plan. Neither side chose the moment. A term nobody read chose it.
Production on an evaluation grant. The customer wired the product into something that runs, and the only paper covering that use contemplated evaluation. Availability, support obligations, what you may verify, what happens when it fails: all of it scoped to a trial, none of it to a system carrying real work. This one is the worst, because it stays invisible until something breaks.
Why the Ending Has to Be Decided in Advance
The ending has to be decided rather than discovered, for a behavioral reason. Peer-reviewed adoption research finds that buyers systematically overvalue an arrangement they already hold, and that the resistance strengthens the longer it runs. Every week an evaluation continues past its stated end, its terms become more of the customer’s reference point, until paying is the change being resisted.
The tooling vocabulary answers this with enforcement. In entitlement systems a trial is a runtime state, a flag with a date, and the systems that carry it are a mature software category. Enforcement is not the decision: a system ends an evaluation on the date it was configured with, executing a considered grant and an accidental one with the same indifference. When the only answer to what happens at the end of a trial is that a flag flips, the company holds an enforcement behavior rather than a grant.
Why AI Evaluations Broke the Old Trial Shape
For most of the industry’s history a trial cost the vendor a seat in a multi-tenant system and some support time: fixed, small, and independent of how hard the prospect pushed. That is why a trial could be a marketing decision at all.
What a Generative Evaluation Consumes
Generative capability changes what an evaluation consumes. Every prompt during the evaluation is LLM inference you buy, at a cost that scales with precisely the behavior the evaluation exists to produce. Your sales engineer tells the prospect to push it hard, to run it against real volume, to bring the ugliest case they have.
So the AI evaluation has a shape the old one never had: variable cost concentrated in one account that has signed nothing, rising in proportion to how well the evaluation is going. A free edition spreads that cost thinly across people who will never buy. An evaluation concentrates it in the one prospect you want, which makes another week easy to justify and a boundary hard to hold.
An uncapped AI evaluation is a cost commitment filed under marketing.
We do not have a clean number for how often this occurs, and the reason we do not is itself the finding. In the agreements we review, sales routinely invents both new agreement constructs and new trial parameters, and finance discovers them later with no way to entitle them, enforce them, or bill against them. It is most pronounced in large enterprise software sales, especially those that still contract predominantly on paper. The selling system holds a reference to a PDF, sometimes not even that, so the terms that govern an evaluation are not in any system that could count them.
The Unit the Evaluation Is Denominated In
The unit question follows immediately. When the paid product bills in credits, the evaluation grant has to state how many credits it conveys, and what a credit is becomes the prospect’s problem before they are a customer. Credits, points, and compute units are surrogate units rather than value metrics, because the buyer cannot count them without your conversion table. An evaluation exists to reduce a buyer’s uncertainty, and an allowance in a unit they cannot convert introduces a fresh one.
We watched this run its course. A software company minted a new currency of its own, one unit standing in for several tracked metrics at once. A three-month paid pilot began in earnest. A month in, the customer received a bill for six hundred thousand dollars in overage nobody had anticipated. The vendor moved to smooth it, spread it, make it go away. It did not matter by then. The customer was spooked, cancelled the pilot, and stopped taking calls.
Read what failed there. The overage was a symptom; the cause was that the customer could not forecast their own consumption in the vendor’s unit, because the unit was an aggregate whose conversion only the vendor could compute. Nothing about the pilot was unfair. It was unforecastable, which in an evaluation is worse: the exercise existed to remove uncertainty and instead demonstrated that the buyer could not predict what using the product would cost them. That is a conclusion, and they drew it.
Assignment stops being administrative the moment the thing consuming the product is not a person. AI agents can consume in an afternoon what a human population consumes across a whole window, so the grant question the pillar puts to paid agreements arrives during the evaluation, before there is an account to invoice.
Tests to Run Against Your Own Trial License Terms
Can you state in one sentence what your evaluation grants that your paid edition does not? If the answer is nothing, it is free for a while, the evaluation is a discount with an expiry date.
Does the population that evaluates resemble the population that will buy? If not, the evaluation is proving something about the wrong group.
What do your terms say happens on the day the evaluation ends, and who is responsible for noticing that day? If the answer is a system, the follow-up is who configured it, and against which decision.
If an evaluation consumed ten times the inference you modelled, what in the grant stops it? If the answer is a conversation, the boundedness rider is improvised by whoever is most invested in the deal closing.
If a prospect never converted and never stopped using the product, which of your systems would tell you?
Most companies answer the first without hesitating. The last one is where we see the pause, because answering it means naming a system that watches for an evaluation still running after its date, and that watching is usually assigned to a person rather than to anything that would raise its hand on its own. For a read on what your terms convey, talk to a pricing expert and describe the paper a prospect signs on day one.
An evaluation is the first license grant most customers ever hold from you, and it sets the reference point for every grant that follows. Deciding it deliberately costs one design conversation. Inheriting it from whatever the last deal produced costs a renegotiation you did not schedule, a cost line nobody owns, and a customer whose expectations were set by terms you never chose. Talk to an expert and describe what your trial conveys today.
