The Ontracko blog
SaaS reliability, outage trends, and recovering the SLA credits your vendors owe you — every number drawn from live monitoring data. New here? Start with what is an SLA credit or the SLA glossary.
Does OpenAI or Anthropic owe you an SLA credit when the API goes down?
OpenAI prints a 99.9% uptime SLA on its Scale Tier and Fast mode pages — for Enterprise customers only — but publishes no credit schedule, no uptime definition and no claim window. Anthropic publishes nothing at all and disclaims uninterrupted service by name. Here is what each one actually owes you.
Why do SLA credit claims get denied? The six exclusions that void a claim
A real outage is not the same thing as claimable downtime. Six exclusion families — plan tier, scheduled maintenance, your architecture, minimum-duration floors, third-party carve-outs and your own configuration — decide whether a vendor pays. Here is how each one works, with the vendors that use it.
DigitalOcean SLA credit: how the 99.99% Droplet guarantee pays out (and how to claim it)
DigitalOcean's CPU Droplet SLA commits to 99.99% monthly uptime per Droplet and pays a 100% service credit below it — no tiers. Here is what counts as downtime, what the credit covers, the two-billing-cycle deadline and the exact email to send.
Does Salesforce pay SLA credits? What its agreement actually commits to
Salesforce publishes no uptime percentage, no credit schedule and no filing window — its agreement promises only commercially reasonable efforts. Here is where your Salesforce SLA actually lives, and what to do after an outage.
How to claim an SLA credit from Twilio (and why your carrier fees don't count)
Twilio pays a flat 10% credit when its APIs miss 99.95% in a month — but the 10% is calculated on your Twilio API fees only, with carrier surcharges, taxes and third-party charges stripped out first. Here's the threshold, the 30-day deadline that runs from month-end, and the exact subject line the ticket needs.
How to claim a Snowflake SLA credit (and why it isn't money)
Snowflake's SLA pays you in Snowflake credits, not dollars — 1x, 3x or 7x your Average Daily Snowflake Credits, applied to next month's usage in the affected region. Here's the credit table, the 21-day deadline that runs from month-end, and how to file.
How long do you have to claim an SLA credit? Deadlines by vendor
Every vendor puts a clock on your SLA credit — 15 days at Atlassian, 21 at Snowflake, 60 at AWS — and they don't all start counting from the same moment. Here's the deadline table, the three different clocks, and how to work out your own date.
Wall-clock uptime vs request success rate: how your SLA is actually measured
Two vendors can both promise 99.9% and mean completely different things. One counts minutes the service was down; the other counts failed requests as a share of total requests. Getting this wrong is the cleanest way to have an SLA credit claim denied.
How to claim an SLA credit from GitHub Enterprise Cloud
GitHub Enterprise Cloud owes a service credit when it misses its 99.9% uptime SLA — but the claim window is quarterly, not monthly, and starts after the quarter closes. Here's the credit schedule and how to file.
How to claim an SLA credit from MongoDB Atlas (99.995% uptime)
MongoDB Atlas commits to 99.995% uptime on M10+ dedicated clusters — about two minutes of allowed downtime a month. Miss it and you're owed 10% to 100% of that cluster's fee, but only if you log a ticket within 24 hours. Here's the credit table and the two-stage claim.
What does a 99.9% SLA actually allow? Uptime, downtime, and your credit
A 99.9% SLA allows about 43 minutes of downtime a month; 99.99% allows about 4. Here's the full uptime-to-downtime cheat sheet, when downtime becomes a breach, and how to calculate the SLA credit you're owed.
How to claim an SLA credit from Microsoft Azure
Azure owes you a service credit when Virtual Machines, SQL Database, App Service and other resources miss their SLA. Here's the exact credit schedule, the deadline, the evidence you need, and how to file the support request.
How to claim an SLA credit from Atlassian (Jira, Confluence, Bitbucket)
Atlassian Premium and Enterprise owe a service credit when Jira, Confluence, or Bitbucket miss their uptime SLA — but you have only until the 15th of the next month to file. Here's the credit schedule and how to claim.
Which SaaS vendors actually pay SLA credits?
Not every vendor with an SLA pays out — and seven we checked pay nothing at all. Here's a side-by-side of the uptime commitments, credit percentages and claim deadlines for the vendors that do owe credits, plus the ones whose contract has no credit behind the number.
What is an SLA credit, and how do you claim it?
A plain-English definition of SLA credits: what they are, when a vendor owes you one, how much you can claim, and the step-by-step process to get the money back.
SLA credit vs service credit: what's the difference?
SLA credit and service credit are often used interchangeably — but the wording in your contract matters. Here's what each term means and what you're actually owed after an outage.
How to claim an SLA credit from Google Cloud
Google Cloud gives you just 30 days to claim an SLA credit — the shortest window of the major clouds. Here's the credit schedule for Compute Engine, Cloud Storage, BigQuery and more, and how to file before it closes.
How to claim an SLA credit from Cloudflare
Cloudflare's Enterprise SLA commits to 100% uptime and pays a formula-based credit. The stage that kills most claims is the notice of intent, due within 5 business days — and Workers, R2 and Queues are governed by their own separate SLAs.
How to claim an SLA credit from AWS
AWS owes you a service credit when EC2, S3, RDS and other services miss their SLA. Here's the exact credit schedule, the per-service case subject AWS requires, the real filing deadline, and how to file the case.
Which of the 144 SaaS vendors we monitor had the most incidents (last 60 days)
The SaaS and cloud vendors with the most public incidents over the last 60 days, scored from their own status feeds — with the reliability caveats spelled out.