Warranty

For a long time the expectations on both sides lived in conversation. It works well; we fix problems fast. Those sentences get said in a sales meeting while the contract records only the price and the term. Both are decent intentions and neither can be adjudicated.

So every outage produced the same argument. Six hours after the service came back, the customer calls it a breach and the vendor says no such promise was ever made. Both are holding the same contract, and because there is no standard in it, each is arguing from memory. Nothing in the document says whether six hours in a month is acceptable.

The warranty clause exists to write that standard down in numbers beforehand. In a product sold to individuals by card, the terms of service cover the role instead: there is nobody to negotiate with, so everyone accepts the sentences the company wrote. Then the first enterprise customer arrives and the clause shows up in full. Their legal reviewer sends over their own contract rather than accepting your terms, and near the front of it is a blank where what you promise is supposed to go.

The clause has three parts: what is promised, for how long, and what happens if the promise breaks.

1. State the promise. What percentage of the month the service will be available, how many hours until a first response on a severe incident. Alongside those sit promises that features behave as documented, that the code doesn't infringe anyone's rights, and that it contains no malicious code.
2. Set the duration. Some promises run for the whole term; others cover a fixed window, like 90 days from acceptance. Subscription services almost always use the former, because the customer is paying every month and expects the promise to be live every month.
3. Order the remedies. This sequence is the heart of the clause. Fix it first; if it isn't fixed inside a stated window, credit the fees for that period; if that still doesn't resolve it, let the customer terminate and refund the remainder. Without those steps, a small outage becomes a conversation about damages.
4. State the exclusions. A line saying no warranties are given beyond those stated is conventional, and free plans and beta features are usually carved out, because a promise attached to something nobody pays for has no way to be kept. Carve-outs work only if the product actually labels those features as beta where the customer can see it.
5. Connect it to limitation of liability. A separate clause caps the amount payable when a promise breaks, and the warranty operates inside that cap. Read the two apart and you will misjudge the size of the exposure.

Work out what the numbers weigh before agreeing to them. Monthly availability of 99.9 percent means 43 minutes of downtime is allowed in a month; 99.5 percent means three hours and 40 minutes. One decimal place changes how you deploy and whether someone carries a pager. Managing those numbers as a separate document is the service level agreement (SLA), and inside a contract the warranty clause usually points at it.

The clause gets opened in three situations.

1. Reviewing an enterprise contract. It's the first place the other side's legal reviewer looks. What has to be checked internally is a fact rather than a sentence: if actual availability over the past six months was 99.7 percent, writing 99.9 into the contract is breaching it in advance. Answering that question requires someone who can read the monitoring history, which is why engineering ends up in a contract review.
2. When you are the buyer. With payment processors and cloud providers, your company is the customer. A thin warranty on their side means nothing supports the promise you gave your own customers, so the numbers in both contracts belong side by side.
3. After an incident. Once service is restored, the clause gets opened and the remedy sequence followed. Calculating the credit from the clause rather than deciding it in the moment reduces the argument. This is also where it meets the termination clause, since repeated breaches become grounds for the customer to end the contract.

"The warranty says 99.9, but actual availability last quarter was 99.6."

The promise and reality disagree, so the clause needs revisiting.

"That's not a warranty breach, it's an indemnification matter."

Distinguishing a problem between us and the customer from one between the customer and a third party.

"Please add language excluding beta features from the warranty."

Declining to attach the same promise to something nobody is paying for.

The line that took longest when Pagedeck reviewed its first enterprise contract was this clause. The draft that arrived specified 99.95 percent monthly availability — 22 minutes of downtime a month. Theo pulled six months of records and the real figure was 99.7 percent.

Grace negotiated the number down to 99.5 percent and added a promise of a first response within two hours on severe incidents instead. What the customer was actually worried about was less the length of an outage than being unable to reach anyone during one. Trading a number the company couldn't hold for one it could, plus a commitment it could staff, closed the clause in a single round.

The first instinct is that a higher number earns trust. 99.99 percent in a contract looks like a serious company. But this figure isn't marketing copy; it's a promise that costs money when missed. It allows five minutes of downtime a month, and falling short means issuing credits every month. Half a year in, the customer expects the credit and the vendor repeats the same apology. Writing a number you can hold, and then holding it, is what builds trust.

The second is assuming a breach leads straight to unlimited damages. Breaking a promise sounds heavy, so the inference is natural. But most contracts stage the remedies and put a liability cap above them: fix, then credit, then termination, with a ceiling on what can be paid. What sets the size of the exposure is not how strong the warranty sentence is but the remedy sequence and the cap sitting next to it.