Summary

  • APNIC reports service-quality Top-3 results of 94% in 2022, 95% in 2024 and 93% in 2026. The earlier published benchmarks are Member-based; the 2026 recent-contact base includes both Members and Stakeholders.
  • The 93% figure is valid for its stated base and remains high. It is not, by itself, a like-for-like Member trend or a measurement of technical service performance. APNIC can preserve both readings by publishing a clearly defined Member-only bridge alongside a separate Stakeholder series.

A familiar line, with a different endpoint

The three numbers invite a simple reading: service quality improved by one point, then slipped by two. APNIC’s 2026 report itself describes results as broadly consistent with earlier surveys. Yet the line needs a footnote before it can carry that interpretation: the people behind its final point are not grouped in exactly the same way as those behind the earlier published service-quality scores.

APNIC reported a 94% Top-3 rating in 2022 among 980 Members who had used APNIC services. The 2024 report gave 95% among 817 Members who had used, interacted or contacted APNIC in the previous two years. In both cases, the published service-quality benchmark was for Members. The 2024 report also presented Stakeholder responses elsewhere, for experience dealing with APNIC and responsiveness; those are not the same service-quality row.

The 2026 question moved onto a wider footing. Q9 asked respondents to rate APNIC overall, and the report says it was put to all respondents. Its headline service-quality result, however, is calculated for a recent-contact group: 1,073 people who had used, interacted or contacted APNIC in the preceding two years. “Don’t know” answers are excluded from the Top-3 score. The report’s historical chart then places 93% beside 94% and 95%.

That sequence retains useful information, but the population label has changed. In the 2026 data appendix, Q9’s service-quality base is 1,072, divided into 780 Members and 292 Stakeholders. The one-response difference between the report’s 1,073 and the appendix’s 1,072 is visible in the published package; the documents do not explain it. It should be reconciled, not assigned a guessed cause.

The central point is narrow. The 2026 aggregate may be sound for its stated recent-contact base while still being a less direct continuation of the Member-only historical indicator. The published record does not provide the unrounded 2026 Member-only Top-3 result beside the prior Member benchmarks. Without it, readers cannot tell whether the apparent two-point movement from 2024 is also present within the Member series.

The added group is part of the result

Stakeholders are not a rounding detail. They account for 292 of the 1,072 responses in the appendix’s Q9 table. The published response-category distributions show why a reader should not infer the aggregate from the headline alone: 54% of Members and 49% of Stakeholders selected “excellent,” while 34% and 35%, respectively, selected “good.” These are displayed whole percentages, not unrounded subgroup Top-3 values. Adding “above average” shares and treating the sum as an exact group score would invent precision the table does not publish.

The split does not establish that Stakeholders caused the combined score to differ from the Member result. It does establish that the combined number represents more than one respondent group. The composition could matter to the aggregate, but the published tables do not quantify how much, and they do not show what the 2026 Member-only Top-3 would have been under a common definition.

Nor does the 93% figure settle whether service improved or worsened. The two-point difference from 2024 is a difference between published point estimates, not an explanation. No causal contribution can be assigned to the Stakeholder share from the rounded subgroup categories. The report’s one-point difference from 2022 is smaller still. Neither change proves a service failure, an operational improvement, or a change in technical reliability.

This is a perception measure. It tells APNIC something about how eligible respondents rated service quality; it does not directly measure registry availability, response times, case resolution, routing data accuracy or compliance with a service objective. High satisfaction and strong operational performance may coexist, but one is not a substitute for the other. The same distinction applies in the opposite direction: a lower survey score would warrant investigation, not an automatic diagnosis.

APNIC says the survey is evolving. Its 2026 report describes collecting operational feedback continuously through engagement channels so the biennial survey can focus more on strategic direction and future priorities. That is a reasonable reason to widen the respondent frame. It also makes a transparent series note more important. A change in who is asked, who is counted, or which answers enter a headline may be sensible; the reader still needs to know when a trend changes its footing.

Three stages sit behind the 93%

The public materials contain three different quantities that should not be compressed into one denominator. The survey as a whole recorded 1,385 responses. Q9 was asked of all respondents. The headline service-quality table then names a recent-contact base of 1,073 and excludes “Don’t know” answers from the positive Top-3 calculation. These statements can coexist: the audience for a question, the group used for a published analysis and the answers eligible for a particular percentage are separate stages.

That distinction is useful well beyond this one chart. A reader who sees only “93%” cannot tell whether it means 93 of every 100 people who opened the survey, 93 of every 100 people with recent APNIC contact, or 93 of the people who supplied a rating rather than selecting “Don’t know.” The report provides more context than the headline alone, but the context needs to travel with the value whenever the chart is reused.

The category table in Appendix B adds a second kind of evidence. It shows that both respondent groups supplied service-quality ratings and lets readers see the distribution of answers. It does not print a precise subgroup Top-3 percentage. The difference matters: category distributions answer “which response labels were selected?” while a Top-3 score answers “what share fell into the three highest labels?” Once those displayed category shares have been rounded, summing them cannot restore the exact group score.

Nor would a single number explain the quality of the sampling frame. A useful note should state who was eligible, how recent contact was defined, how Q9 responses were handled, and how “Don’t know” answers were treated. If the total invited population or response rate is available, it can be shown separately; it must not be inferred from the count of completed survey responses.

The older Member benchmarks also have different entry rules

The shared respondent category is Members, but the public descriptions of the 2022 and 2024 bases are not identical. The 2022 footnote says “Members who have used APNIC services only,” n=980. The 2024 service-quality row specifies n=817 Members who used, interacted or contacted APNIC in the previous two years. The 2024 questionnaire uses its own routing language, and the final report’s explicit table label is the safer guide to the published benchmark. A Member-only label is necessary for a careful comparison, but it does not by itself prove that every eligibility rule stayed fixed.

This point sharpens, rather than weakens, the case for a bridge. APNIC should not promise a perfectly continuous three-wave line if the underlying definitions cannot be reconciled. It can publish a 2026 Member-only Q9 under the current recent-contact rule, state how that rule differs from each historical base, and show the combined recent-contact result beside it. If the underlying records permit a second calculation that reproduces an earlier eligibility rule, it can be labelled as a reconstruction with its own denominator. If they do not, a visible break is the more accurate record.

The 2024 Stakeholder results do not fill the gap. They concern experience dealing with APNIC and responsiveness, with a separate n=198, rather than the same Member service-quality row. Keeping that distinction prevents a tempting but invalid shortcut: a reader cannot splice a different question into the Member series merely because it comes from the same survey wave.

The practical reporting unit is therefore not a lone percentage. It is a small, stable record attached to each wave: question identifier and wording; respondent group; eligibility and contact window; number asked and number answering; number included in the score; the treatment of “Don’t know”; the three categories included in Top 3; and the unrounded numerator and denominator. A one-line change note can explain when the group definition shifts. This lets future analysts compare like with like where possible and clearly mark where the published evidence does not support a direct comparison.

The question identifier also changed, from Q11 in 2022 to Q10 in 2024 and Q9 in 2026. A different number does not establish that the underlying idea changed, but it makes a crosswalk valuable. The question wording, scale, answer labels, respondent group and published base should travel together so a future reader can see what stayed comparable and what did not.

Preserve the Member line and show the wider view

APNIC does not need to choose between historical continuity and broader participation. It can publish two adjacent measures for 2026:

  1. A Member-only service-quality result, with the exact question wording, eligibility rule, recent-contact window, denominator, treatment of “Don’t know,” and any exclusions stated. If the historical definitions cannot be reconstructed exactly, the report should say so and label the closest available Member comparison.
  2. A separate Stakeholder result, defined on its own terms, followed by the combined recent-contact result. This keeps the expanded community signal visible rather than hiding it to preserve a Member trend.

The next chart could show all three series with a short break marker at 2026: the earlier Member benchmarks, the 2026 Member-only bridge, and the 2026 combined community measure. A compact table should give the number invited or eligible if available, the number answering Q9, the number rating the item, the number choosing “Don’t know,” and the Top-3 numerator and denominator before rounding. The n=1,073 versus n=1,072 discrepancy between the report and Appendix B should be explained in the same reporting note or corrected in a versioned update.

The recommendation is not to publish respondent-level records. Group-level results, explicit bases and a description of processing rules can support scrutiny without exposing individual answers. If the Member-only value cannot be computed on a comparable basis, the honest output is a labelled series break and a new baseline—not a reconstructed percentage derived from rounded category cells.

This distinction matters because an institution can use survey evidence in more than one way. A stable Member series can inform longitudinal questions about the relationship between APNIC and its Members. A broader community series can reveal how Stakeholders experience the same organization. Neither is a mandate from all account holders or the regional Internet community, and neither is a substitute for operational measures. Keeping both on the page makes their uses easier to judge.

APNIC’s 93% service-quality score is not weakened by a more precise label. It is a strong reported result for the group APNIC names. The unresolved question is whether it continues the same series as 94% and 95%. Publish the bridge, show the Stakeholder result separately, and let the reader see both continuity and change without asking one number to do two jobs.

Sources