Knowledge Management Solutions

Forrester runs two adjacent categories that people routinely confuse. Cognitive search platforms find knowledge. Knowledge management solutions produce it, curate it, and keep it correct.

That distinction decides which one you actually need, because they solve opposite halves of the problem.

Search assumes the knowledge exists somewhere and is accurate, and works on the difficulty of surfacing it. Knowledge management assumes nothing. Its questions are whether the article was ever written, whether the person who wrote it knew what they were talking about, whether it is still true after three product releases, and who is responsible when it is not.

Generative AI has transformed the first problem and barely touched the second. That asymmetry is the most useful thing to understand about this category right now.

What these platforms cover

Knowledge management spans the lifecycle of an organisation's documented know-how.

Authoring, with templates and structure so that articles are consistent rather than idiosyncratic. Review and approval, so that something inaccurate does not become authoritative. Publishing to the audiences and channels that need it, whether an agent desktop, a self-service portal, or an internal workspace. Feedback loops so that people using the knowledge can report when it fails. Governance covering ownership, review cadence, and retirement. And analytics on whether any of it is working.

Forrester's assessment of Atlassian describes strengths in a project-based approach to knowledge creation and collaboration, pulling knowledge from a wide set of sources into a centralised workspace, with integrated asynchronous video singled out as innovative.

Video is worth pausing on. A meaningful share of organisational knowledge has always resisted documentation because writing it down is slow and the person who knows it is busy. A recorded walkthrough captures it in the time it takes to do the thing once, which lowers the contribution barrier considerably. It also creates a searchability problem, since video is opaque without transcription and indexing.

Inside The Forrester Wave: Knowledge Management Solutions, Q4 2024

Published on 2 December 2024 and authored by Julie Mohr with Stephanie Balaouras, Sarah Morana, and Kara Hartig, the evaluation scored eleven providers.

Atlassian placed as a Leader, positioned by Forrester for customers looking to transform traditional silos into an enterprise-wide collaborative practice with an AI-first strategy.

KMS Lighthouse also placed among the Leaders, positioned around enterprise-level knowledge for midsize and large organisations.

USU was among the eleven evaluated, and its own reading of the result is the most interesting counter-positioning in the market.

Forrester's headline industry finding was that generative AI is changing knowledge management before our eyes, and its strategic recommendation was to pay close attention to still-evolving metrics and analytics in this category.

Both deserve unpacking, and the second is the one buyers skip.

A vendor competing on restraint

USU's stated interpretation of its placement is that it focuses on quality rather than uncontrolled use of AI, taking a targeted approach where other providers exploit the full potential of generative capability, with a quality guarantee covering correctness, completeness, usability, and repeatability.

That is a vendor positioning explicitly against the direction of its own market, and in knowledge management it is a defensible argument rather than a laggard's excuse.

The reason is what knowledge management is for. A knowledge base exists to be authoritative. When an agent reads an article to a customer, or an engineer follows a runbook during an incident, or a compliance officer relies on a documented procedure, the value depends entirely on the content being right.

Generative capability applied to authoring produces plausible articles quickly. Plausible is precisely the wrong target for a system of record. An article that reads well and is subtly wrong is more dangerous than no article, because it will be trusted and it will not be questioned.

The counter-argument is equally real. Most organisations have knowledge gaps because writing documentation is unrewarded work that nobody has time for, and a tool that drafts an article from a resolved ticket or a recorded session closes gaps that would otherwise stay open indefinitely.

The resolution is not either position but where the human sits. Generated drafts reviewed by someone accountable is a productivity gain. Generated articles published without review is a quality problem that compounds silently, because nobody discovers the error until someone acts on it.

Repeatability in USU's list is the underrated word. A knowledge system that gives the same answer to the same question every time is auditable. One that generates a fresh answer each time is not, and in regulated contexts that difference matters more than fluency.

Knowledge decays

The failure mode in this category is not that knowledge is missing. It is that it is stale, and staleness is invisible.

An article is written when a process is introduced. It is accurate that day. Then the product changes, a policy is revised, a system is replaced, a regulation shifts, and the article stays exactly as it was. It remains searchable, remains authoritative-looking, and is now wrong.

Nobody notices, because the failure surfaces at the point of use rather than at the point of storage. An agent follows an outdated procedure, a customer receives incorrect guidance, and the article that caused it is still sitting at the top of the search results.

The organisational reason is ownership. Articles are written by whoever solved the problem, and that person moves teams, changes role, or leaves. Review cadences exist in policy and lapse in practice, because reviewing four hundred articles is a quarter of somebody's job that nobody is funded for.

This is why governance capability matters more in evaluations than it appears to. A platform that tracks ownership, enforces review dates, flags articles whose underlying systems have changed, and retires content nobody has used in two years is doing the work that determines whether the knowledge base is an asset or a liability.

The related discipline that most organisations skip is retirement. Deleting knowledge feels destructive, so bases grow monotonically, and a search returning eleven articles of which two are current is worse than one returning two.

The metrics problem Forrester flagged

Forrester's recommendation to watch still-evolving metrics and analytics points at a genuine weakness, and it is worth being specific about why the existing measures fail.

Article views measure traffic, not usefulness. A heavily viewed article might be excellent, or it might be so unclear that people read it three times.

Deflection measures avoided contacts, which is a cost metric rather than a quality one. It counts the person who found an answer and the person who gave up and left equally.

Search success rates measure whether something was clicked, not whether it resolved anything.

Feedback ratings suffer from response bias, since the people who bother to rate are disproportionately those who were frustrated.

What the category needs and does not yet have is outcome measurement: did following this article solve the problem, did the customer come back, did the agent handling time drop, did the same question recur. Those require connecting knowledge usage to downstream events in other systems, which is why the analytics are still evolving.

The practical position for a buyer is to treat vendor-reported knowledge metrics with the same scepticism you would apply to any self-reported measure, and to instrument the outcome yourself where the data exists.

Knowledge management became AI infrastructure

The change that raises the stakes on everything above is that knowledge bases are now the grounding source for AI systems.

A retrieval-augmented assistant answering employee or customer questions pulls from the knowledge base. Every property of that content becomes a property of the assistant's answers.

An outdated article produces a confidently outdated answer. A contradictory pair of articles produces an answer that picks one arbitrarily. A gap produces an answer the model fills from its training data, which is where a large share of enterprise AI hallucination originates.

The difference from the pre-AI situation is scale and detectability. A person reading a stale article might notice something looks wrong, check the date, or ask a colleague. A generated answer strips the article's context, presents a synthesised statement with no visible age, and delivers it with uniform confidence.

So the unglamorous curation work that organisations have deferred for twenty years has become the determinant of whether their AI programme produces reliable output. That is an uncomfortable dependency, and it is the strongest argument available for funding knowledge governance properly.

It also explains why the boundary with cognitive search is blurring commercially while remaining clear conceptually. Search vendors need good content to retrieve. Knowledge management vendors need good retrieval to deliver. Both are now selling into the same AI-driven budget, and buyers should be clear which half of the problem they actually have.

If your organisation cannot find knowledge that exists and is correct, that is a search problem. If what surfaces is wrong, contradictory, or absent, no search platform will fix it, and buying one produces a very efficient way of retrieving inaccurate content.

What this asks of an organisation

The consistent finding across this category, over decades, is that knowledge management is a practice with tooling rather than a tool with a practice.

The platforms have improved substantially. Authoring is easier, structure is better supported, governance is more automatable, and generative assistance genuinely lowers the cost of getting something written down.

None of that supplies the thing that actually determines success, which is whether the organisation treats documented knowledge as work that counts. In most places it does not. Writing an article is what you do after the real work, it is not measured, and it competes against everything that is.

Organisations that resolve this do something concrete rather than cultural: knowledge contribution is part of resolution workflow rather than a separate task, ownership is assigned to roles rather than individuals, and review is scheduled rather than aspirational.

The ones that do not end up with an expensive system containing a large volume of content of uncertain age and unclear accuracy, which is now also feeding an AI assistant that presents it with total confidence.

Forrester does not endorse any vendor named here, and tier placement should not be read as a recommendation to buy.