Service level agreements and service credits in SaaS contracts

Reference article · 5 min read

A vendor's public uptime page is a marketing document. It can be rewritten without notice, it defines availability in terms that favor the vendor, and it rarely provides a remedy an agency can use. A link to that page is not a service level commitment.

An enforceable service level agreement is a dated contract exhibit with four working parts: a definition of the service precise enough to measure, reporting the agency can verify, remedies that scale with the harm, and an exit if failure becomes chronic. Each part is negotiable.

Define availability so it can be measured

• Scope. A platform can be up while the payment module, the public portal, or the nightly interface to finance is down. Define availability for each service the agency depends on, including integrations and batch jobs.
• Degraded is not available. Add a response-time threshold so a system too slow to use counts as downtime.
• Measurement point. Ask where the monitoring probe sits and how often it runs. A check every five minutes can miss short outages, so the reported percentage overstates availability.
• Start and stop of the clock. Downtime should begin at the earlier of vendor detection or the agency's first report, and end when service is restored for users, not when the ticket closes.

Exclusions and maintenance windows

The exclusion list is where availability commitments quietly disappear. Carve-outs typically cover maintenance, force majeure, carrier problems, agency-side networks, and third-party services. Three deserve negotiation.

• Cap scheduled maintenance. Limit the hours per month, require notice measured in days, and set windows in the agency's time zone outside business hours.
• Constrain emergency maintenance. Allow it, but require prompt notice and a written follow-up. An unlimited exception swallows the commitment.
• Do not excuse the vendor's own suppliers. The vendor chose its cloud host and its subprocessors. An outage there is a vendor outage.

Measurement period, reporting, and evidence

• Use a monthly period. Quarterly and annual periods absorb a long outage into a number that still reads as compliant.
• Translate the percentage into minutes. A commitment of 99.9 percent in a thirty-day month permits roughly forty-three minutes of downtime. At 99.5 percent the allowance is more than three hours.
• Require proactive reporting. The vendor should publish the calculation every month whether or not a target was missed. Reporting on request only means the agency must detect the problem first.
• Preserve evidence rights. Require access to monitoring data and incident records, and allow the agency's own monitoring as evidence when the vendor produces none.

Support response and resolution by severity

• Write severity definitions into the contract, with examples from agency operations, and state who assigns severity and how disputes are resolved. If the vendor classifies unilaterally, most incidents become low priority.
• Separate response from restoration. An acknowledgment is not work. Set targets for initial response, for a workaround, for permanent resolution, and for status updates.
• Match coverage hours to the service. Dispatch, utility outage management, and detention systems need round-the-clock coverage with a reachable human. A permitting system may not.
• Escalate by title and elapsed time, so an unresolved severity one incident reaches an executive.

Service credits, and why they are a weak remedy

Credits are usually a tiered percentage of the monthly fee for the affected service, rising as availability falls. The mechanics matter more than the percentages.

• Claim process. Many agreements require a written claim within a short window. Assign that duty to a named contract administrator, or negotiate automatic credits.
• Cap. Total credits are almost always capped at a portion of one month's fee, so know the ceiling before relying on the remedy.
• Sole and exclusive remedy. Most vendor forms make credits the agency's only recourse, which converts a serious outage into a small discount. Carve out termination rights, data loss, and security incidents at minimum.
• Credit versus cash. A credit against future invoices is worthless to an agency that is leaving. Require a cash refund for prepaid fees and for credits outstanding at termination.

A credit signals that something went wrong. It does not compensate for what the failure cost: a day of stalled benefits processing costs far more than a day of fees.

Chronic failure, root cause, and exit

Require a written root cause analysis within a set number of business days after a severity one incident or a missed target, with corrective actions and dates. Define chronic failure numerically: misses in consecutive months, a set number in a rolling year, or repeated severity one incidents. It should trigger a corrective action plan, then a right to terminate for cause with no termination charge, a prorated refund of prepaid fees, and transition assistance.

Attach the agreement as a dated exhibit that changes only by written amendment. A clause letting the vendor revise terms at a web address is not a commitment.

What to ask and what to require

Ask during evaluation how availability is defined and where it is measured, whether maintenance counts against uptime, who assigns severity, whether credits are automatic or claimed and what caps apply, and whether reports arrive without a request.

Require in the contract per-service availability definitions, a monthly measurement period, capped maintenance windows, proactive reporting with evidence rights, a severity table with response and restoration targets, automatic credits that are not the exclusive remedy, root cause analysis obligations, and a chronic failure termination right.

Terms vary by state and agency policy, and counsel should review the final language. HQT works with state and local agencies on cloud and software service agreements, including the service level terms that go with them.

Frequently asked questions