Knowledge, KEDB & Self-Service

Knowledge management is the quiet engine of shared services efficiency. A well-run knowledge base deflects demand (customers resolve without requesting) and accelerates resolution (agents resolve faster with known solutions). A poorly-run one — stale, inaccurate, unowned — is worse than none, because it confidently misleads. The discipline of knowledge, KEDB, and self-service is what turns knowledge from a by-product into a strategic asset.

The KEDB (Known Error Database) captures known problems and their workarounds — the issues the service desk sees repeatedly, with documented fixes. The knowledge base captures the broader knowledge customers and agents need. Self-service deflection uses both to let customers resolve without requesting, removing the ticket entirely. AI-assisted knowledge is making both generation and retrieval faster and more accurate.

This page covers the KEDB, the knowledge lifecycle, self-service deflection, AI-assisted knowledge, knowledge quality, and the knowledge culture that makes it compound.

The KEDB is the record of known errors — problems that have been identified, with their workarounds and (where known) their root causes and permanent fixes. When an agent receives a ticket matching a known error, they apply the workaround immediately, rather than diagnosing from scratch. The KEDB is the institutional memory that prevents every agent re-solving what the organisation has already solved.

The KEDB's value compounds with coverage and currency. A KEDB that captures the top recurring problems, with accurate workarounds, dramatically reduces resolution time and first-contact effort. A KEDB that is stale (workarounds no longer work) or sparse (common problems not captured) is a liability — agents trust it, apply an outdated workaround, and make the problem worse. Currency is non-negotiable; the KEDB must be maintained as problems evolve.

The KEDB links to problem management. Each recurring incident is a problem-management signal; the problem is diagnosed, the workaround documented in the KEDB, and the permanent fix tracked. The KEDB is the live view of what is known but not yet fixed; the problem backlog is what is being worked toward fix. Together they turn recurring incidents from repeated cost into documented knowledge.

Knowledge has a lifecycle: create, review, publish, use, maintain, retire. Each stage has an owner and a discipline. Knowledge created but not reviewed is unverified; reviewed but not used is shelfware; used but not maintained goes stale; maintained but not retired clutters the base with obsolete articles. The lifecycle is what keeps knowledge current and useful.

The create stage captures knowledge where it is produced — at the ticket, at the problem, at the project. Knowledge created after the fact, from memory, is less accurate than knowledge captured in the flow of work. The discipline is to make capture easy (templates, prompts at ticket close) and immediate (capture while the detail is fresh).

The maintain stage is the one most often skipped and most important. Knowledge decays — tools change, processes change, products change — and an article that was accurate at creation is wrong a year later. The maintenance discipline is a review cadence (each article reviewed by its owner at a defined interval) and a usage-triggered review (an article with high use but low helpfulness ratings is reviewed immediately). Without maintenance, the knowledge base becomes a graveyard that misleads.

Deflection is the customer resolving without raising a ticket — finding the answer in the knowledge base and applying it. The cheapest ticket is the one never raised. Deflection reduces L1 volume, improves customer experience (the customer gets an immediate answer), and frees agents for the work that genuinely needs them. A mature shared services organisation deflects a meaningful share of demand.

Deflection succeeds when the knowledge is findable and accurate. Findable means the customer can search in natural language and get the right article; accurate means the article solves the problem. Either failing breaks deflection — an unfindable answer might as well not exist; an inaccurate answer that frustrates drives the customer to call, having wasted their time. The discipline is to optimise both, measured by deflection rate (what share of would-be tickets were deflected) and by helpfulness ratings.

Deflection and self-service are related but distinct. Self-service is the customer requesting through the portal (the request is processed); deflection is the customer resolving without requesting (the request never exists). Both reduce L1 load, but deflection removes the demand entirely — the more valuable reduction. The mature shared services organisation deflects what the customer can self-resolve, self-serves what they can request, and leaves L1 for what genuinely needs a human.

GenAI is transforming knowledge management in two ways: generation (drafting articles from tickets and problems) and retrieval (surfacing the right knowledge at the point of use, in natural language). Both reduce the cost of knowledge capture and the friction of knowledge use — the two frictions that have historically limited knowledge programmes.

Generation lets knowledge be captured in the flow of work: an agent closes a ticket, and GenAI drafts a knowledge article from the ticket and the resolution, for the knowledge owner to review and publish. This dramatically lowers the capture cost — historically, agents did not write articles because the effort was high and the benefit diffuse. GenAI flips this: capture is low-effort, and the knowledge base grows from real work.

Retrieval lets knowledge be used at the point of need: a customer or agent asks a question in natural language, and GenAI retrieves and synthesises the answer from the knowledge base. This improves both deflection (customers find answers that search missed) and resolution (agents get the relevant knowledge without leaving the ticket). The governance — retrieval bounded to verified knowledge, sources cited, human review for consequential answers — keeps AI-assisted knowledge trustworthy.

Knowledge quality is measured, not assumed. The metrics that matter: helpfulness rating (did the article help the user?), accuracy (is it correct, verified by sampling), currency (is it within its review cadence?), and use (is it being found and used?). Each is a signal; together they tell whether the knowledge base is an asset or a graveyard.

The helpfulness rating is the most powerful. After an article is used, the user is asked 'was this helpful?' The aggregate rating, and the trend, reveal which articles earn their place and which mislead. Low-rated articles are reviewed and fixed or retired; high-rated articles are the base's valuable core. The rating is the quality signal that users provide and that drives maintenance.

Accuracy is verified by sampling. An article can be highly rated and subtly wrong (the user thought it helped but applied a fix that masked the problem). Periodic sampling by subject experts catches what ratings miss and maintains the base's trustworthiness. Accuracy without helpfulness is unused; helpfulness without accuracy is misleading; both are needed.

Knowledge programmes succeed or fail on culture. A culture that values knowledge — where agents capture, maintain, and use it as part of the work — produces a compounding asset. A culture that treats knowledge as extra work, or as someone else's job, produces a graveyard. The culture is the foundation; the system is the instrument.

The culture is shaped by recognition and time. Agents recognised for knowledge contributions (and given time to make them) contribute; agents measured only on ticket closure do not. The leadership that values knowledge in its metrics and recognition gets a knowledge culture; the leadership that values only throughput does not. Culture follows what is measured and recognised.

The culture also values maintenance. An organisation that celebrates new articles but never reviews them produces growth without quality; one that celebrates currency and helpfulness produces quality that compounds. Maintenance is not glamorous, but it is what keeps knowledge trustworthy — and the culture that values it is what makes maintenance happen.