Definition
Knowledge Quality Decay Index (KQDI) measures the ratio of knowledge article references in re-opened or recurring tickets versus total article references.
The Measurement Gap
The AI-powered knowledge management system had been running for 14 months. Article count: 4,200. Average article age: 8 months. Usage rate per article: 3.2 views per month. The knowledge manager's quarterly review showed every metric trending in the right direction: more articles, higher utilization, lower knowledge gap scores on incoming ticket categories.
What the metrics didn't show was what was inside the articles.
The AI system generated and updated articles based on patterns in resolved tickets. When a ticket on "VPN authentication fails after password reset" was resolved, the resolution note read: "Cleared browser cache, reconnected VPN client, resolved." That note, a workaround, not a fix, became the basis for the AI's knowledge article on VPN post-reset authentication issues. The root cause, a race condition in the authentication token refresh sequence, was never documented because the ticket was marked resolved before anyone investigated it.
Over the following 6 months, that article was referenced 847 times. Of those, the research team later estimated, roughly 40% were genuine solutions: cases where clearing the cache actually fixed the problem. The other 60% were users following the workaround, sometimes successfully, often not. When it didn't work, they reopened their ticket or contacted a human agent.
The knowledge base metrics looked excellent throughout. Article usage was up. Ticket deflection was up. The AI system was generating 12 new articles per week. Nobody had visibility into whether the articles were propagating solutions or propagating workarounds.
This failure mode is specifically enabled by AI-driven knowledge management. In a human-curated knowledge base, each article requires a human to deliberately write and review it. The effort required means only articles someone believes are genuinely useful get created. In an AI-driven system, the article creation barrier drops to zero: the AI generates articles continuously from ticket resolution patterns, and the quality of the source data determines article quality. More articles doesn't mean better articles. It often means more authoritative-looking documentation of the wrong answers.
Existing knowledge management metrics don't see this. Knowledge base article count measures volume, not accuracy. Article usage measures whether users are consulting the knowledge base, not whether the articles they're consulting are correct. Knowledge gap score measures coverage, not quality. CSAT surveys measure whether an interaction felt helpful, not whether the solution worked.
The measurement gap: there is no metric that tells you whether AI-curated knowledge articles propagate root-cause fixes or workaround patches, which means knowledge base quality can degrade while every volume and utilization metric improves.
Formula
Knowledge Quality Decay Index (KQDI) measures the ratio of knowledge article references in re-opened or recurring tickets versus total article references. When a user consults a knowledge article and then reopens their ticket, opens a duplicate, or contacts a human agent on the same issue, that article reference is a quality failure signal. KQDI aggregates these signals across the knowledge base to measure whether AI-curated content is improving resolution quality or propagating ineffective guidance. KQDI is Integration-Required. The data exists across two systems (knowledge base article usage logs and ticket recurrence data from ITSM) but they need to be joined on article reference ID and ticket outcome. Most organizations don't have this join built today.
In Plain Language
for every time a knowledge article was consulted during a service desk interaction, track what happened afterward. Did the ticket close without follow-on? Or did the user reopen, open a duplicate, or contact a human agent on the same issue? The ratio of "article referenced before a resolution failure" to "total article references" is KQDI. A rising KQDI means the knowledge base is increasingly populated with content that doesn't actually resolve issues.
Formula
KQDI = Article_references_before_resolution_failure / Total_article_references × 100
Where:
- Article_references_before_resolution_failure = instances where a knowledge article was consulted during ticket resolution, and the ticket subsequently reopened, generated a duplicate, or generated a channel-switch within N days
- Total_article_references = all knowledge article references during the measurement period
- N = recurrence window, consistent with TRR (14 days recommended)
Worked Example
A service desk tracks knowledge article usage via the ITSM platform's article linking feature. In Q3, knowledge articles were referenced in 8,200 ticket interactions: either the AI system pulled an article as its resolution basis, or the user accessed an article through the knowledge base portal. Checking each article reference against the 14-day recurrence window:
- 6,150 references: the associated ticket closed without reopen, duplicate, or channel-switch. The article appears to have supported a genuine resolution
- 2,050 references: the associated ticket generated at least one resolution failure within 14 days. The user reopened, opened a duplicate, or contacted a human agent
KQDI = 2,050 / 8,200 × 100 = 25%
One in four knowledge article references precedes a resolution failure. Drilling down by article: 12 articles account for 48% of the failure-associated references. These 12 articles are causing a disproportionate share of the knowledge quality problem: they're consulted repeatedly and not helping.
Sign Convention
Lower KQDI is better. Rising KQDI means the proportion of article references that precede resolution failures is increasing: the knowledge base is becoming less useful over time, even as it grows. The most actionable KQDI analysis is per-article: the aggregate tells you there's a problem; the per-article breakdown tells you exactly where it is.
Edge Cases
Article reference attribution: if the AI system pulls articles internally to generate responses without surfacing article IDs to the ITSM system, the reference isn't logged. The ITSM system needs to record which article ID was the basis for the AI's resolution response, not just that an AI response was given. This is a logging configuration requirement. Multiple articles referenced in one ticket: if the AI consulted three articles before responding and the ticket was subsequently reopened, all three references count as failures. Track primary article (the one used to generate the resolution) separately from supplementary references where possible. New articles with insufficient history: articles with fewer than 20 references in the measurement period should be excluded from per-article KQDI ranking. The sample is too small to distinguish quality from noise. Cross-references: Upstream, KQDI feeds TRR (True Resolution Rate, GMI-IT-007). When knowledge quality is low, TRR declines: the AI is resolving fewer tickets genuinely because it's drawing on low-quality knowledge to generate responses. Downstream, KQDI affects first-contact resolution rate, which is downstream of actual resolution quality. Maturity: Integration-Required. Article usage logs exist in knowledge base platforms. Ticket recurrence data exists in ITSM. The integration is joining them: recording which article was referenced during which ticket interaction, and checking whether that ticket generated a follow-on event. This requires the ITSM system to capture article reference data in a structured, queryable form.
Proposed formula, pending validation pilots.