How long do SaaS outages actually last?

Every vendor posts when an incident opened and when it closed. Nobody adds them up. We do, every day, across 804 status pages — 15,172 incidents that carry both timestamps.

15,172incidents with a start AND a resolve timethe measurable set

2.3 hmedian incident lengthhalf are fixed sooner

28.1 h90th percentile1 in 10 runs longer than this

76vendors with 20+ resolved in 90 daysranked below

Slowest vendors to resolve, last 90 days — median length of every incident they opened and closed in the window (minimum 20 measured). · Source: each vendor's own status page, read daily by this project · as of 2026-09-24

Median time to resolve, last 90 days

Filter by vendor

Vendor	Median	90th pct	Measured
Exact	27.4 h	13.0 days	85
Qualys	26.3 h	5.3 days	57
Luno	23.9 h	4.7 days	23
Ionos Cloud	22.2 h	6.5 days	30
Akamai Edge DNS	17.1 h	3.6 days	26
Datto	10.4 h	7.1 days	69
F5	9.2 h	24.8 h	28
ThousandEyes	6.3 h	14.9 days	21
Twilio	5.9 h	46.2 h	160
Kraken	5.8 h	4.8 days	73
KnowBe4	5.7 h	26.4 h	21
MeridianLink	5.3 h	24.1 h	34
Vonage	5.1 h	17.8 h	21
Flexera	5.0 h	3.1 days	30
Imperva	5.0 h	25.4 h	25
IPVanish	5.0 h	21.8 h	38
Barracuda	4.8 h	24.7 h	20
Ivanti Cloud	4.3 h	3.2 days	41
QuickNode	4.2 h	19.0 h	66
Granicus	4.0 h	44.2 h	50
Bandwidth	3.9 h	21.6 h	93
Digitalocean	3.9 h	21.2 h	30
Plaid	3.5 h	22.0 h	22
Gemini	3.4 h	29.0 h	24
Vonage API	3.4 h	31.1 h	65
Linode	3.3 h	4.3 days	25
Sentry	3.2 h	6.3 h	30
Harness	3.1 h	24.4 h	36
Zenoti	3.0 h	20.9 h	31
Webroot	3.0 h	2.5 days	50
Voximplant	2.8 h	4.7 days	24
The Linux Foundation	2.6 h	5.4 days	26
Genetec	2.6 h	4.5 h	27
Shippo	2.5 h	13.8 h	68
Tailscale	2.4 h	7.7 h	23
Elastic	2.3 h	4.1 days	26
Sinch	2.3 h	16.7 h	70
Temporal	2.2 h	10.7 h	32
Dbt Cloud	2.0 h	6.3 h	25
Circle	2.0 h	3.1 days	44
Supabase	2.0 h	24.5 h	57
WatchGuard	2.0 h	17.3 h	21
IONOS	119 min	26.0 h	47
Snowflake	116 min	6.0 h	21
Cisco Umbrella	109 min	11.2 h	25
Grafana Labs	107 min	8.6 h	64
Coinbase Prime	105 min	18.6 h	44
Array	103 min	4.8 h	32
Cycode	100 min	7.6 h	27
Cisco Systems	97 min	7.1 h	43
BitGo	89 min	7.6 h	22
Flyio	85 min	7.5 h	51
Expo	83 min	8.4 h	33
Circle CI	82 min	6.1 h	29
Alchemy	81 min	19.9 h	57
OpenAI	81 min	9.4 h	53
Whatnot	80 min	7.2 h	35
GitHub	79 min	5.3 h	60
Radware Cloud Waf	78 min	4.7 h	40
Coinbase	78 min	27.1 h	71
Scaleway	77 min	3.8 days	75
Amplitude	76 min	4.4 h	23
Cloudflare	71 min	10.1 h	107
Discord	62 min	2.4 h	21
Anthropic	61 min	3.4 h	56
Caseware Cloud	59 min	4.8 h	24
Vercel	57 min	2.1 h	32
Hostinger	54 min	3.8 h	85
Fireblocks	47 min	3.1 h	26
GoDaddy	46 min	3.0 h	38
Zoom	44 min	12.2 h	64
Ledger	32 min	21.9 h	55
Kore.ai	19 min	3.1 h	27
Visma	11 min	6.3 h	66
HeroCoders	9 min	113 min	26
Liveramp	8 min	85 min	44

Only vendors with at least 20 incidents opened AND resolved in the last 90 days appear — a median over three incidents reads like a fact and is not one. The full per-incident table is the free outage-duration dataset (CC BY 4.0).

What is excluded, and why

Of every incident on record, these are dropped rather than estimated:

Rows	Excluded because
656	scheduled maintenance windows (a planned window is not an outage)
146	incidents still open, or with no resolve time posted
45	resolutions we inferred because the incident simply vanished
235	incidents the vendor posted as resolved BEFORE they started

Read this before quoting a number. Duration is the gap between the vendor's OWN posted timestamps. It is not measured downtime, not uptime, and not an SLA figure. A vendor that posts every small blip and closes it honestly will look "slower" here than one that posts nothing at all — this measures how vendors communicate as much as how fast they fix. Statuspage-hosted vendors back-fill their history, so they dominate the measurable set; platforms that expose only currently-open incidents contribute almost nothing, which is why we do not publish a per-platform comparison.

Watch your own vendors — free MIT template