Luca Barberis

How wide is the spread of service performance?

№ 202 · SLA distribution

01

Definition

MetricDistribution of service performance against the service level agreement, read as percentiles rather than a mean

Unit% of units or contacts inside SLA, plus p50 and p90 of service time

Histogram of SLA hit time per ticket or order. Right-skewed. Pull ticket open and close timestamps. Long tail tells you about the worst-case experience customers see. Example: 88% delivered on time; 8% 1-day late; 4% above 1-day late (focus on the 4%).

SLA
Service level agreement. Contracted response or resolution time. Measured as percent of tickets resolved inside the promised window.
02

Benchmarks

Bottom 30%MedianTop 30%
below 80% inside SLA with a fat right tail, p90 above 6 times p5085% to 92% inside SLA, p90 around 3 to 4 times p5095% or more inside SLA, with p90 under 2 times p50

Blended service operations covering customer contact channels and physical fulfilment, 2024 to 2026. Contact channel anchors from Zendesk, Tidio and Verint panels; fulfilment anchors from the WERC DC Measures survey and retailer on-time in-full compliance programmes. The spread ratio between p90 and p50 is the transferable figure; absolute times are channel-specific. · Two definitional issues dominate. First, mean versus percentile: a small number of very late units inflates the mean and simultaneously hides the typical experience, which is why platform vendors recommend reporting median alongside p90 rather than an average. Second, what counts as the clock start and stop. First response time normally excludes automated acknowledgements, and on-time shipping measures leaving the dock rather than arriving with the customer, so the same operation can publish very different numbers honestly. The p90 to p50 ratio bands above are a synthesis across the cited channel data rather than a single published distribution, so treat them as a working rule. Absolute attainment bands are better evidenced.

By category
CategoryBottom 30%MedianTop 30%
Live chat first responseover 5 minutesabout 1 minute 35 secondsunder 40 seconds
Email first responseover 24 hours7 to 12 hoursunder 4 hours
Social first responseover 12 hoursnot separately publishedunder 1 hour
Outbound shipment on-time (dock)below 90%95% to 98%99.2% or better
03

When it looks bad

The mean sits comfortably inside SLA while the histogram carries a long right tail, so the published number stays green and a stable minority of customers has a bad experience every single week.

Median first response 42 minutes against a 4 hour SLA, but p90 is 31 hours and 9% of tickets breach. Those breaching tickets carry most of the refund requests.

04

What to do about it

  • Report p50 and p90 side by side and retire the average from the operating review. A small number of very late units inflates the mean while simultaneously hiding the typical experience, which is why support platform vendors recommend tracking both (Lorikeet, 2026).
  • Convert the SLA from an average to a percentile commitment, the 80% within 20 seconds convention, so the operating target and the customer promise are the same object rather than two loosely related numbers (Verint Workforce Management Study, 2024).
  • Automate first touch on the highest volume intent to pull the left side of the distribution and free agents for the tail. AI now handles roughly 74% of initial chat interactions with first response under 3 seconds, which changes the shape of the histogram rather than just its centre (Tidio, 2025).
  • Give the breach tail its own queue and its own owner. Best-in-class warehouse on-time shipping is 99.2% and the distance from 95 to 99 is almost entirely tail management, not average speed, so it will not be closed by making the median faster (WERC DC Measures, 2024).
05

Sources

  1. Warehousing Education and Research Council, reported by Hyster-YaleBenchmarking and improving distribution center metrics, from the 2024 DC Measures Report · 2024 · annual survey of distribution and logistics professionals; best-in-class is the top 20% of respondents Best-in-class on-time shipments 99.2% or better, decreased from 99.5% the prior year and described as a continued challenge. On-time shipments ranked fourth among the most-tracked DC metrics. scg-lm.s3.amazonaws.com ↗
  2. Verint, reported by HelpableLive Chat Response Time Benchmarks 2026 · 2026 · Verint Workforce Management Study 2024 The contact centre convention is 80% of chats answered within 20 seconds, an explicit percentile commitment rather than an average target. gethelpable.com ↗
  3. Tidio, reported by HelpableTidio Customer Service Benchmark Report 2025, cited in Live Chat Response Time Benchmarks 2026 · 2025 · Tidio customer service platform data across industries Median chat first response time 1 minute 35 seconds across all industries, SaaS 1 minute 22 seconds, ecommerce 1 minute 48 seconds. AI handles 74% of initial chat interactions without human involvement, and for those first response falls under 3 seconds. gethelpable.com ↗
  4. ZendeskZendesk 2025 CX Trends Report · 2025 · n=10,000+ global consumers and business leaders Response time is reported as the strongest single predictor of chat satisfaction. Every additional minute of wait reduces CSAT by roughly 2 to 3 points, so the right tail of the distribution carries a disproportionate satisfaction cost. zendesk.com ↗
  5. Zendesk, reported by RinglyCustomer service response time benchmarks for 2026 · 2026 · Zendesk CX Trends 2026 plus HubSpot State of Service data 63% of customers rank speed of response as the number one factor in a support experience, ahead of speed of resolution at 57% and channel availability at 49%. 88% expect faster responses than a year earlier and 53% say waiting too long is the single most frustrating part of dealing with a business. ringly.io ↗
  6. LorikeetFirst Response Time Benchmarks: How Fast Should Your Team Be in 2026? · 2026 · cross-platform support benchmark synthesis Top-performing teams run under 40 seconds on live chat, under 1 hour on social and under 4 hours on email, against an industry average email first response of 7 to 12 hours. Explicitly recommends reporting median plus 90th percentile rather than the mean because a small number of delayed tickets inflates the average. lorikeetcx.ai ↗
  7. MetricHQOn-Time In-Full (OTIF) · 2025 · supply chain metric reference, sample not disclosed Fast-moving consumer goods typically target 95 to 98% OTIF to major retailers, industrial manufacturing accepts 88 to 92%, healthcare demands 98%+. Immature organisations typically discover true OTIF is 15 to 20 points lower than believed once tracking is comprehensive. metrichq.org ↗
  8. Red Stag Fulfillment, citing Walmart supplier termsOn-Time In-Full (OTIF): Meaning, Benchmarks, Best Practices · 2025 · retailer supplier compliance programmes Walmart charges suppliers 3% of the cost of goods sold for each OTIF failure. Partial success scores zero, so an order that is on time but 10% short counts as a full miss. redstagfulfillment.com ↗
  9. MRPeasyOn-Time In-Full (OTIF): A Complete Guide · 2025 · industry synthesis, sample not disclosed OTIF around 95 to 99% is considered excellent in most industries and a poor score falls below 85%. mrpeasy.com ↗