# Dispatch efficiency: verify arrival variance against a 15-minute threshold

Chase Pierce · October 13, 2026

> Learn how to verify arrival variance against a 15-minute dispatch efficiency threshold and compare like-for-like totals before committing to a live option.

| Takeaway | Detail |
| --- | --- |
| Use a 15-minute response variance threshold for dispatch efficiency. | The thesis defines the guide around Beyond time to arrival: a dispatch efficiency metric using a 15-minute response variance threshold. |
| Verify the live, complete option before committing. | The reader rule requires comparing like-for-like totals and terms before any commitment. |
| Add a 15-minute cost review to the existing monthly architecture review. | The EKS cost monitoring stack added a 15-minute cost review; automated enforcement above a cost threshold was scrapped after engineers gamed it by splitting workloads across namespaces. |
| Focus on the 15-minute window of “easy time” to prioritize efficiency. | The 15-minute city guide says focusing on this 15-minute window of “easy time” puts us into the mindset of prioritizing efficiency. |

This guide delivers a verify-before-you-commit method for dispatch efficiency using a 15-minute response variance threshold. It shows how to compare live, complete options on like-for-like totals and terms before committing.

![Dispatch efficiency](https://static.mm-ais.com/article-images-ai/dispatch-efficiency-verify-arrival-varia-ai-31b1f467.jpg)

## How It Works

At its core, the metric works by measuring the spread of arrival times around a fixed reference point rather than the average alone. You record each dispatch's actual time to arrival, subtract the scheduled or promised time, and keep the signed difference. A dispatch that lands early produces a negative deviation; one that lands late produces a positive deviation. The variance is then computed across the full set of deviations, not just the late ones, because a system that swings wildly in both directions is unstable even when its mean looks acceptable. The 15-minute figure functions as the tolerance band: deviations inside that band count as on-target, and the metric's job is to report how much of the distribution falls outside it and by how much.

Three terms carry the mechanism. *Time to arrival* is the elapsed interval between dispatch release and the unit's arrival at the committed location, measured on one clock and one timezone. *Response variance* is the dispersion of those intervals around the reference value — the signed deviations described above, summarized so that both direction and magnitude survive. *Tolerance threshold* is the boundary you set for acceptable deviation; here it is 15 minutes, meaning any arrival more than 15 minutes from its reference value is counted as a miss regardless of whether it was early or late. The Wikipedia entry on efficiency frames the underlying idea well: efficiency is the measurable ability to avoid wasting time, effort, and materials while performing a task. Applied here, the waste being avoided is time lost to arrivals that miss their promised window, and the 15-minute band is how you measure it.

| Term | What it measures | How to check it |
| --- | --- | --- |
| Time to arrival | Elapsed dispatch-to-arrival interval | One clock, one timezone, timestamped at both ends |
| Response variance | Signed spread of arrivals around the reference | Subtract reference from actual; keep the sign |
| Tolerance threshold | Acceptable deviation boundary | Flag any deviation beyond 15 minutes, early or late |

The practical check follows from the definition. Before you commit to a variance figure, confirm that every dispatch in the sample has both timestamps recorded, that the reference value is the same one used at dispatch time, and that early and late deviations are both retained. If early arrivals have been discarded, the variance is understated and the metric flatters the operation. Recompute one row by hand: if a unit was promised at a given time and arrived 20 minutes after it, the deviation is +20 and it sits outside the 15-minute band; if it arrived 10 minutes before, the deviation is −10 and it sits inside. That single arithmetic step is the whole test in miniature.

Interpretation depends on what the variance is compared against. A tight variance with a small share of out-of-band arrivals indicates a predictable process; a wide variance with the same mean indicates one that cannot be relied on for planning. The 15-minute band is a design choice, not a natural constant, so state it explicitly whenever you report the metric and never let it drift between reporting periods. The same discipline applies to the reference value: if it changes, the variance changes with it, and any comparison across periods becomes meaningless.

One caution on enforcement. A cost-monitoring team writing for DEV Community tried automated rejection of deployments above a threshold and found engineers gamed it by splitting workloads across namespaces, so the rule was scrapped. The lesson transfers directly: a variance threshold that triggers automatic penalties invites the same kind of splitting, and the metric stops measuring what it was built to measure. Use the variance figure to inform review, and verify the raw timestamps yourself before acting on any summary.

![How It Works — Dispatch efficiency](https://static.mm-ais.com/article-images-ai/dispatch-efficiency-verify-arrival-varia-ai-0c1ce747.jpg)

## Key Factors to Consider

When you evaluate a dispatch efficiency metric built on a 15-minute response variance threshold, three criteria do most of the work: the completeness of the timing record, the symmetry of the comparison, and the stability of the threshold itself. Start with completeness. A variance figure is only as good as the sample behind it, so before you commit, confirm that every dispatch in the review window has a recorded actual arrival time and a recorded promised time. If either field is missing for even a handful of records, the variance you compute describes the subset that was logged, not the operation. Ask for the count of dispatches in the period and the count of dispatches with both timestamps; if those two numbers differ, treat the metric as provisional until the gap is explained.

Symmetry is the second criterion, and it is where most comparisons quietly break. Two dashboards can report the same threshold and still measure different things if one uses signed differences and the other uses absolute values, or if one anchors to the scheduled time while the other anchors to a later revised promise. Before you compare, verify that both sides use the same reference point, the same sign convention, and the same inclusion rule for early arrivals. A dispatch that arrives ahead of schedule produces a negative signed difference; whether that counts as good performance or simply as variance is a definitional choice, and the two sides of your comparison must make it the same way.

The third criterion is threshold stability. A 15-minute variance threshold is a policy line, not a law of nature, and it can be moved. Confirm how long the current threshold has been in force, who can change it, and whether it changed inside the period you are reviewing. If the line moved mid-period, split the data at the change date rather than averaging across it.

| Number to verify | Why it matters |
| --- | --- |
| Dispatches in period vs. dispatches with both timestamps | Sample completeness |
| Reference point used (scheduled vs. revised promise) | Like-for-like comparison |
| Signed vs. absolute difference convention | Direction of variance |
| Date the 15-minute threshold took effect | Period integrity |
| Count of dispatches inside vs. outside the threshold | Distribution, not just average |

The numbers that matter most are the ones that describe distribution rather than central tendency. An average variance can look healthy while a meaningful share of dispatches sit far outside the threshold, so ask for the count inside and the count outside, plus the largest single deviation. The goal, per the efficiency definition cited earlier in this guide, is to avoid wasting time and effort while performing the dispatch task, and a threshold metric only serves that goal when you can see how much of the operation actually clears the line rather than a single average that hides the tail.

## Common Mistakes

The first pitfall is comparing a partial timing record against a complete one and calling the difference efficiency. Suppose you track a single week of dispatches and find that 12 of 40 arrivals landed inside the 15-minute window, a 30% hit rate. A vendor or an internal dashboard then shows a 45% hit rate for the same period. Before you commit to that gap as real, verify the denominators match: 12 ÷ 40 = 30%, but if the comparison figure came from only 20 logged dispatches with 9 inside the window, that is 9 ÷ 20 = 45% on half the sample. The honest like-for-like check is to recompute both rates on the same set of dispatches, then compare the signed differences (actual time to arrival minus promised time) for each one. If the two records do not cover the same dispatches, you are measuring record-keeping, not dispatch efficiency.

The second pitfall is treating the 15-minute threshold as a fixed property of the operation when it is really a review cadence you chose. A 15-minute cost review added to an existing monthly architecture review is a monitoring habit, not a physical constant, and the same logic applies here: the threshold is a decision rule, and decision rules get gamed. In one documented case, automated enforcement that rejected deployments above a cost threshold was defeated by engineers splitting workloads across namespaces, so the rule survived while the behavior it was meant to catch did not. The same failure mode applies to arrival variance. If dispatchers learn that anything past 15 minutes triggers scrutiny, they will reclassify, split, or round the record rather than fix the underlying delay. Your check is to audit a sample of raw timestamps against the reported variance, not the summary.

A third, quieter mistake is letting the threshold drift while you are not looking. The metric only means something if the same rule is applied to every dispatch in the comparison, so before you commit, confirm the threshold value, the reference point, and the inclusion rules are identical on both sides of the comparison. If one side counts a dispatch as on-time at 15 minutes and the other at 14, the totals are not comparable and no amount of arithmetic will rescue them.

Run one test before you commit: pull the raw timing records for both sides, recompute the signed differences yourself, and confirm the threshold and inclusion rules match. If they do not, fix the record before you fix the operation.

## Insider Tactics

The most reliable insider move is to run the variance calculation on a rolling window rather than a single reporting period, because a one-off snapshot can be gamed by a good week. Pull the last several dispatch cycles, compute the signed difference between actual and promised arrival for each one, and watch whether the spread is narrowing or just shifting. A metric that looks efficient in aggregate can hide a widening tail, and the tail is where the operational risk lives. If your review cadence is monthly, add a short interim check so a drifting threshold gets caught before it compounds; the EKS cost-monitoring pattern described in the Common Mistakes section — a short review attached to an already-running monthly meeting — is a useful cadence to copy even when the underlying metric is time rather than dollars.

Timing matters more than most teams admit. The moment to verify is immediately before you commit to a threshold, not after the first month of data. Set the reference point, confirm the record is complete for every dispatch in the window, and only then lock the threshold. If you lock first and backfill later, you are comparing a partial record against a complete one, which is the fastest way to manufacture a false efficiency gain. The efficiency definition cited earlier in this guide makes the point directly: efficiency is the ability to avoid wasting time and effort, and an incomplete timing record makes any variance figure meaningless no matter how clean the summary looks.

Use a like-for-like pairing rule. Before you compare two dispatch cohorts, confirm four things match: the same reference point, the same window length, the same inclusion criteria, and the same treatment of late or missing entries. If any one of those differs, the comparison is not a comparison. Write the four checks into your template so the reviewer cannot skip them under deadline pressure.

One non-obvious tactic is to test the threshold against a deliberately adversarial case before you adopt it. Take the single worst dispatch in your record, apply the threshold, and ask whether the metric would have flagged it. If the answer is no, the threshold is too loose for the decision you are trying to support. The Wikipedia entry on efficiency makes the same point in general terms: it is a measurable ratio of useful output to total use, and a ratio that cannot separate good from bad is not doing work.

| Pre-commit check | What to verify |
| --- | --- |
| Record completeness | Every dispatch in the window has an actual and a promised time |
| Reference point | Identical across both cohorts being compared |
| Window length | Same number of cycles on each side |
| Adversarial test | Worst case is flagged by the threshold |

Finally, treat the threshold itself as a living parameter, not a fixed law. Recompute the spread at each review and ask whether the current threshold still separates efficient dispatches from inefficient ones. If it no longer does, adjust it and document why. A threshold that never moves is either perfectly calibrated or quietly obsolete, and the second case is far more common.

## Comparison

Put the two options on one line and the decision usually makes itself. Take a dispatch operation that logs 200 arrivals in a month. Option A, the current vendor, records every arrival and shows a mean signed difference of +4 minutes with a spread that puts 18 arrivals outside the 15-minute threshold. Option B, the challenger, reports a mean of +2 minutes but only logs 140 of the 200 arrivals. On the surface B looks tighter. Recompute on the same base: B's 140 logged arrivals represent 70% of the total, so its apparent advantage rests on 60 unlogged dispatches that could sit anywhere. Until B produces a complete timing record, the like-for-like comparison is A's 200 against B's 140, and A wins on verifiability even though its mean is worse.

When the records are equally complete, the winner flips to whichever option keeps more arrivals inside the threshold. Suppose both log all 200. A has 18 outside; B has 11 outside. B wins on the variance count, and the gap is 7 dispatches — small enough that you should check whether the threshold itself was applied the same way in both datasets. If A counted a 16-minute arrival as outside and B rounded it to 15 and counted it inside, the 7-dispatch gap is an artifact of rounding, not efficiency. Ask each vendor for the raw signed differences and re-tally the count yourself before you commit.

Each option wins under a different condition, and the condition is the point. The complete-record option wins when you cannot audit the incomplete one — a partial log is a claim, not a measurement. The tighter-variance option wins when both logs are complete and the threshold was applied identically. The cheaper option wins only after you have confirmed that its completeness and its threshold treatment match the alternative; a lower mean built on missing data is not a lower mean, it is a smaller sample.

| Check | Option A | Option B |
| --- | --- | --- |
| Arrivals logged | 200 of 200 | 140 of 200 |
| Mean signed difference | +4 min | +2 min |
| Arrivals outside 15-min threshold | 18 | Not verifiable |
| Like-for-like verdict | Wins on completeness | Loses until log is complete |

Run the same table with both logs complete and the verdict can reverse: 200 logged each, A at 18 outside, B at 11 outside, and B takes it. The rule that survives both scenarios is simple — never declare a winner until the two columns cover the same arrivals and the same threshold arithmetic. Verify the live, complete option first; compare totals and terms second; commit third.

## What to do next

| Step | Action | Why it matters |
| --- | --- | --- |
| 1 | Pull your current dispatch logs and calculate response variance against the 15-minute threshold for the last full month. | You can't enforce a threshold you haven't measured; this establishes your baseline before any change. |
| 2 | Add a 15-minute cost review to your existing monthly architecture review, mirroring the cadence used in the EKS cost monitoring stack. | Attaching the review to an already-running meeting keeps it alive; standalone reviews get skipped. |
| 3 | Verify the live, complete dispatch option — full totals and terms — before committing to any tooling or process change. | Comparing like-for-like totals prevents a partial picture from driving the decision. |
| 4 | Reject any automated enforcement above a cost or variance threshold; the EKS rollout showed engineers gamed it by splitting workloads across namespaces. | Enforcement that can be gamed produces compliance theater, not efficiency. |
| 5 | Focus improvement work on the 15-minute window of "easy time," as framed in the 15-minute city guide, to prioritize efficiency where it compounds. | Targeting the window where variance is easiest to recover builds the mindset of prioritizing efficiency. |
| 6 | Re-run the variance calculation after one cycle and compare like-for-like against your baseline before locking the threshold into policy. | Confirming the delta on identical terms ensures the 15-minute threshold is earning its place. |

## Frequently Asked Questions

**Should the dispatch score be based on average arrival time alone?**

The metric measures the spread of arrival times around a fixed reference point rather than the average alone.

**What calculation keeps early and late arrivals distinguishable?**

Record each dispatch's actual time to arrival, subtract the scheduled or promised time, and keep the signed difference.

**If a dispatch arrives early, what sign does its deviation carry?**

A dispatch that lands early produces a negative deviation.

**What must be true of an option before I commit to it?**

The guide requires verifying the live, complete option before committing.

**How often should cost reviews run relative to the monthly architecture review?**

Add the 15-minute cost review to the existing monthly architecture review.

**Why did the EKS cost monitoring stack drop automated threshold enforcement?**

Automated enforcement above a cost threshold was scrapped after engineers gamed it by splitting workloads across namespaces.

## Quick answers

| What threshold should be used for dispatch efficiency? | Use a 15-minute response variance threshold for dispatch efficiency. |
| --- | --- |
| What does the metric measure at its core? | At its core, the metric works by measuring the spread of arrival times around a fixed reference point rather than the average alone. |
| How is each dispatch's deviation recorded? | You record each dispatch's actual time to arrival, subtract the scheduled or promised time, and keep the signed difference. |
| What should be verified before committing? | Verify the live, complete option before committing. |
| What method does this guide deliver? | This guide delivers a verify-before-you-commit method for dispatch efficiency using a 15-minute response variance threshold. |

Also worth reading: **Work order closeout with voice: 18 to 3-4.2 minutes, dispatch or skip**: [Work order closeout with voice:](https://technician.dev/blog/work-order-closeout-with-voice-18-to-3-42-minutes-dispatch-or-skip.php) · **Warranty checks for field service: 28-second verify vs blind dispatch**: [Warranty checks for field service:](https://technician.dev/blog/warranty-checks-for-field-service-28-second-verify-vs-blind-dispatch.php) · **The AI dispatch metrics that actually move the needle**: [AI dispatch metrics that actually](https://technician.dev/blog/the_ai_dispatch_metrics_that_actually_move_the_needle.php)

### Related reading

- [2026 Dispatch: 15% Override Rate Resets AI Confidence Threshold](https://technician.dev/blog/2026-dispatch-15-override-rate-resets-ai-confidence-threshold.php)
- [Warranty checks for field service: 28-second verify vs blind dispatch](https://technician.dev/blog/warranty-checks-for-field-service-28-second-verify-vs-blind-dispatch.php)
- [AI Diagnostics: Why 80% Release Threshold Beats 95%](https://technician.dev/blog/ai-diagnostics-why-80-release-threshold-beats-95.php)
- [Auto Assign Field Technicians: 80% Confidence Triggers Dispatch in 2026](https://technician.dev/blog/auto-assign-field-technicians-80-confidence-triggers-dispatch-in-2026.php)
- [Work order closeout with voice: 18 to 3-4.2 minutes, dispatch or skip](https://technician.dev/blog/work-order-closeout-with-voice-18-to-3-42-minutes-dispatch-or-skip.php)
- [2026 Dispatch Math: 3-SKU Universal Kit vs Closest Tech $47](https://technician.dev/blog/2026-dispatch-math-3-sku-universal-kit-vs-closest-tech-47.php)

### Latest

- [When to trust AI triage over a dispatcher gut call: 2026 split at 3 signals vs 1](https://technician.dev/blog/when-to-trust-ai-triage-over-a-dispatcher-gut-call-2026-split-at-3-signals-vs-1.php)
- [Auto Assign Field Technicians: 80% Confidence Triggers Dispatch in 2026](https://technician.dev/blog/auto-assign-field-technicians-80-confidence-triggers-dispatch-in-2026.php)
- [Predictive Skill Matching: How to Verify First-Time-Fix and Travel-Time Claims](https://technician.dev/blog/predictive-skill-matching-how-to-verify-first-time-fix-and-travel-time-claims.php)

Canonical: https://technician.dev/blog/dispatch-efficiency-verify-arrival-variance-against-a-15-minute-threshold.php
Markdown: https://technician.dev/blog/dispatch-efficiency-verify-arrival-variance-against-a-15-minute-threshold.php/index.md