Skip to content
Blog

Scattered company knowledge: connect documents, applications and expertise

Scattered company knowledge rarely means there is no documentation. More often, an answer requires combining a Drive proposal, a conversation, a project’s current status and something a senior employee remembers. Each part lives somewhere different. For an agency, consultancy or software company with 20 to 200 people, the challenge is making that context accessible without stopping work to migrate every file.

In this guide

The symptom: plenty of files, but no complete answer

Someone finds the client onboarding procedure but cannot tell what changes when an external supplier is involved. Another colleague locates the commercial proposal without realising its scope was amended afterwards. Both have information. Neither yet has all the context required to act.

Knowledge can be scattered across locations, versions and people’s experience. Search helps with location. Conflicting versions require a decision about which agreement still applies. Undocumented expertise requires somebody to explain the reasoning. Buying more storage will not address those three situations in the same way.

Microsoft’s 2023 Work Trend Index identified information searching and communication overload as workplace difficulties. That provides general context, not a measure of your company’s losses. Understanding those requires observing your own processes and the time they consume.

Inventory answers, not just applications

Choose a process that crosses several sources and has a clear owner: preparing a proposal, onboarding a client or resolving a recurring incident. Ask the department for questions that actually arise during the work. A small sample can reveal gaps before the project expands.

For each question, note which source supports the answer, who can confirm that it is current and which part only one person understands. If the answer is that everything is in Notion, ask for the specific page and an exception it does not explain. This separates the location of a file from the usefulness of its contents.

A template for one process; examples are hypothetical
Work questionAvailable sourceMissing context
What does this service include?Proposal and project recordLatest scope change
How is this incident resolved?Technical procedureWhen to apply the exception
Who approves the change?Team agreementDeputy and current status

Distinguish documents, business data and human judgement

A contract, an opportunity’s status and a senior employee’s explanation serve different purposes. If they appear as one paragraph without provenance, the team cannot tell what is an approved reference and what needs confirmation. Bringing knowledge together means preserving those distinctions.

  • Documents and agreements: describe processes, decisions and terms. They need an origin, an owner and review when things change.
  • Business applications: provide status, amounts or activity for a period. A number needs a date, relevant filters and authorised access.
  • People’s expertise: explains exceptions and when a rule stops being appropriate. It has to be expressed, captured and validated.

Connecting an application cannot capture an explanation nobody has provided. A recorded demonstration, a discussion of a difficult case or an answer to an unresolved question can help. The tacit knowledge guide explores capture methods. Here, the task is to place each contribution within an answer another person can use, with the right source and context attached.

Agree what happens when two sources contradict each other

Centralising contradictory documents can make the wrong information easier to reach. Before testing a question, establish which source should govern that kind of answer. A recent message does not automatically override a contractual term. An approved manual may need review if the actual process has changed.

Give the decision to the process owner. Record which content remains valid, which version is replaced and which exception applies only to one case. If no agreement exists, the answer should expose the discrepancy and request confirmation. Silently selecting the most convincing text hides a problem the team needs to resolve.

Also define what happens to a broken link or deleted document. If supporting material is no longer available, the person consulting it needs to know. Search and source maintenance should support that review. A citation by itself is not evidence that the information remains current.

A shared knowledge system still needs access boundaries

Operations may need the billing procedure without access to a personal employment record. An agent preparing a proposal may use an approved catalogue without seeing all financial information. Define access around the work that must be done, rather than the convenience of connecting an entire account.

Before adding a source, specify which content enters, which department it belongs to and who will be able to consult it. Then test a question with an authorised person and the same question with somebody who should not have access. A permission list in a presentation is not a substitute for that test.

In Savia, sources are selected and their department assignments reviewed. Company and department permissions do not automatically reproduce every individual permission in the original application. The security and data isolation page and data processing agreement explain the controls and providers involved. Access design belongs in the initial scope, not as a final adjustment once sensitive sources have already been added.

Test with questions that might fail

Easy demonstration questions tell you little. Include an answer spread across two documents, an exception explained by a senior employee, a fact that has changed and a question whose answer is absent. The sample should resemble the department’s actual work, including ambiguity.

Someone who knows the process defines the expected supporting sources. Someone who needs to use the process asks the question and explains what they would do next. This second step reveals answers that look correct while omitting the condition that prevents a mistake.

  • Does the answer find the right source and let the reader open it?
  • Does it preserve the exception or limit that changes the decision?
  • Does it distinguish an unconfirmed explanation from an approved agreement?
  • Does it recognise questions with insufficient supporting information?
  • Does it respect the person’s access permissions?

Keep failures as a work list. Updating a source and repeating a question verifies a specific correction. The number of connected files alone does not demonstrate that the knowledge is usable.

Start without organising the whole company first

Management can approve one process and a person responsible for validation. Inventory, source connections and testing are organised with that department and the implementation provider. The buyer does not need to become the administrator of every application or personally write the manuals.

An initial scope might contain a set of work questions, authorised sources and the missing explanations needed to answer them. This is not a commitment that the whole company will be ready within days. It provides evidence of the actual work involved and informs the next scope.

Define what would justify expansion: better verified answers, fewer repeated searches or a handover that works using the assembled information. If the real problem is that nobody has authority to approve an exception, access to more documents will not fix that lack of authority.

Use the knowledge loss diagnostic to examine dependency costs before prioritising. For a specific departure, see the senior employee replacement cost calculation.

The same company context for people, AI tools and agents

Once knowledge is connected, the benefit can extend beyond another search screen. AI tools or agents your team already uses can receive that context to prepare an analysis, compare options or develop a proposal with company agreements in view.

Savia provides a digital brain: it brings sources and contributions together, returns answers with provenance and helps collect missing explanations. The company can enable access from its AI tools and agents. Business data queries are read-only, and knowledge proposed by an external AI requires human review before publication.

There is no need to export a new copy of everything for every question. There is still a need to review access, understand which provider receives the necessary information and check freshness. Context supports a decision. It does not make every agent answer correct, or turn a suggested action into an approved company decision.

Prevent the new knowledge system from becoming outdated

Assign a reference person to each process and a way to communicate changes. When commercial terms change, a procedure is retired or a new exception appears, the update needs to reach the sources people consult. Connecting a tool does not remove that responsibility.

Review failed questions, contradictory answers and contributions awaiting approval. Focus on the knowledge that affects work rather than asking everyone to rewrite every document each month. A repeated question may reveal a missing explanation or show that finding an existing answer is still unnecessarily difficult.

Review handovers as well. If a senior employee moves to a different department, identify who can validate the knowledge that previously depended on them. That practice turns connected knowledge into a resource owned by the organisation instead of another collection only its original creator understands. Expansion can then follow evidence of usefulness, rather than the volume of content imported.

Sources and method

A practical inventory and evaluation proposal from Savia IA. The table contains hypothetical examples. Microsoft’s report provides context for the problem, not a company cost estimate or evidence of Savia outcomes.

Related reading

The cost of losing a senior employee: calculate your company’s exposure

Separate recruitment spending, internal time and knowledge that has to be rebuilt. A worked example with explicit assumptions and no invented salary multipliers.

Employee onboarding: provide context without repeating every senior explanation

A 30-day plan for checking access, context and autonomy, with a knowledge template for each task and a way to measure the support your experts provide.