Measure cross-silo knowledge transfer by identifying which other groups need a particular practice, what they can do with it and where the evidence stops. Record the direction of transfer, the receiving group's conditions and a checked example where possible. Views, messages and meeting attendance can help locate activity, but they cannot establish that another department can use the knowledge.
For a knowledge or AI enablement lead in a company with several departments, the useful result is a decision about a boundary. Does a team need access, a clearer explanation, a local adaptation or help from an experienced colleague? Start with a few consequential practices and their relevant recipients, rather than a score for every employee.
The tools and workflows guide covers the wider arrangements that support this work. This guide focuses on the evidence needed to see where useful knowledge travels and where it does not yet have a usable route.
Choose the boundaries that matter to the work
Not every group needs every practice. A finance team may have no use for a laboratory's sample-intake method. Some information also needs to remain restricted. Treating every absent connection as a failure would reward indiscriminate sharing and make the measurement less useful.
Choose a knowledge item and ask who needs it to complete a specific job. Identify the source group, the intended receiving groups and the condition that makes each recipient relevant. Include shifts, sites or functions that might be missed by a department-level list.
Suppose Sam Taylor coordinates shared guidance at a construction-materials testing business. The North laboratory has a reviewed method for using an approved AI tool to draft an internal discrepancy note when a sample label and its request form disagree. A coordinator checks the note against both records and refers the discrepancy through the existing process. The tool does not resolve the identity of the sample or authorize testing.
South laboratory handles similar intake work. A separate sample-receiving team prepares records before laboratory intake, while East laboratory works from another approved document system. The night shift also encounters discrepancies. These are relevant receiving groups for different reasons. Their needs should shape the review before Sam asks for examples of success.
APQC's guidance on knowledge silos recommends identifying knowledge likely to benefit other groups and targeting its sharing. It also warns that overwhelming people with marginally relevant material can defeat the purpose. This is professional practice guidance, not a tested AI measurement scheme.
Keep the starting record small:
- The knowledge item: Identify the current instructions and the check they are meant to preserve.
- The receiving group: Name the site, shift or function that has a relevant task.
- The reason for transfer: Explain what that group should be able to do differently.
- The boundary: Note permitted access, important local differences and the responsible contact.
Use the library-entry guide if the source material lacks those conditions. A measurement exercise cannot repair an ambiguous practice merely by counting its readers.
Separate contact from usable knowledge
A connection between departments is a possible route for knowledge. It is not an inventory of what moved along that route. Two teams may exchange messages because a process keeps failing. A quiet relationship may work well because an agreed method is already embedded in the job.
A study by Longqi Yang and colleagues, published online in 2021, examined collaboration data from 61,182 US Microsoft employees around the 2020 shift to remote work. Its analysis found more siloed collaboration networks. The authors could not directly measure productivity or innovation, and the study concerns one firm over a short period. Its network measures should not become a benchmark for successful AI knowledge transfer.
For your own review, state what each observation establishes. Different evidence can support different next actions.
| Observation | What you can reasonably record |
|---|---|
Another team opened the guidance | Exposure to the material; understanding and use remain unverified |
A recipient describes using it | Reported use, with the task and circumstances attached |
A recipient shows a permitted, checked output | Observed use on that example, including any help or correction |
A team changes its instructions after trying it | A documented adaptation; review whether it preserves the important checks |
No usable response or example is available | An unknown result, with the reason and follow-up route recorded |
A cross-department meeting can still be useful without proving transfer. In an October 2026 LinkedIn account, learning and development manager Edoardo S. Pellizzari described bringing together shipboard trainers from security, public health, safety and environmental compliance. They identified opportunities to coordinate induction topics and connect key messages. His account describes a discussion and proposed improvements, rather than measured implementation results.
The useful follow-up would be to inspect the changed induction arrangements and the receiving trainers' experience. Reporting the meeting as completed is accurate. Reporting that knowledge now flows successfully between all four functions would require further evidence.
Keep the direction and the local change visible
Return to Sam's discrepancy-note method. South laboratory can show a reviewed note prepared from its own permitted sample records. The sample-receiving team adapts the method to compare its receiving reference with the laboratory reference. During a walkthrough, that team identifies a missing cross-reference check and returns the proposed improvement to North for review.
East cannot open the approved example from its document system. Nobody has yet followed up with the night shift. A library dashboard could combine all of this into activity around one popular item. Sam needs the differences to choose what to do next.
One source practice leaves four different transfer records
A local example was reviewed
South shows a checked discrepancy note from its own permitted records. This establishes observed use on that example.
An adaptation returns a question
A walkthrough exposes a missing cross-reference check. Receiving sends the proposed improvement to North for review.
The access route needs repair
East cannot open the approved example through its document system. An authorized retrieval route remains unresolved.
Use remains unobserved
No follow-up example has been obtained. The record does not establish successful use, failed use or refusal.
The South walkthrough supports a claim about the observed task. It does not show that everyone at South can now perform it. The receiving team's adaptation is another kind of contribution: knowledge can change as it crosses a boundary, and useful feedback can travel back. North should review that proposed check before incorporating it into the maintained method.
Record the link between the source version and the local version. Keep the receiving team's reason for changing it, the example reviewed and any unresolved condition. Do not require identical prompt wording as evidence of transfer. In this case, a different reference system makes a local change necessary.
Look beyond the most visible team when the record is incomplete. East's access problem calls for an authorized route to the approved material. The night shift's missing observation calls for follow-up at a suitable opportunity. Neither establishes that the employees refused to share or learn.
For a deeper walkthrough of an individual recipient's use and later opportunities, follow the practice-transfer observation guide. Keep that detailed observation separate from the broader question of which relevant groups your review has covered.
Report coverage before summarizing success
Choose the receiving groups and review period before collecting the most attractive stories. Include relevant groups that tried and stopped, could not obtain access or have not yet encountered the task. Explain how groups were selected and whose work was actually observed.
In Sam's example, the review covers four selected receiving groups. There are reviewed examples from South and sample receiving, an access barrier at East and no use evidence from the night shift. That is a useful summary. Calling it “50% adoption” would hide what was measured and imply more than two walkthroughs can establish.
Keep the counts at the same level. State which unit you counted:
- People: Four attendees establish participation by four people.
- Groups: Four departments establish a set of organizational units to investigate.
- Work examples: Four reviewed notes establish the size of the reviewed example set.
These are not interchangeable denominators. A large department can contain several shifts with different access and practices. One representative cannot automatically speak for them all.
If the next review finds evidence from the night shift, distinguish improved observation coverage from newly acquired capability. The shift may already have used the method before Sam asked. Likewise, a revised department structure can change the count of boundaries without changing anyone's practice. Keep the definitions stable or explain the change.
An account of friction may be more useful than an unexplained red mark on a chart. In an August 2025 Reddit discussion, a Reddit user who described designing and repairing delivery workflows said apparent team underperformance could reflect problems elsewhere, such as sales commitments or unclear prioritization. This is an unverified individual account. It is a reason to investigate the surrounding process before treating a missing transfer as a team's deficiency.
If leaders need an organization-wide estimate, agree a sampling and evaluation design that fits that claim. Selected walkthroughs can diagnose obstacles. They do not establish a representative diffusion rate or the causal effect of a knowledge programme.
Collect enough to repair the boundary
Explain the purpose of the review, who can see the observations and how they will be used. For a knowledge-improvement exercise, a reference to a permitted example and a short account of the difficulty may be sufficient. Copying private prompts, customer documents or message histories into a progress report adds information the decision may not need.
The materials-testing business can review an approved, sanitized discrepancy example. Its report can say that East lacks a permitted retrieval route without publishing an employee's access history. Keep identifiable working notes restricted, agree a retention period and consider whether a small shift remains recognizable in an aggregate.
Do not quietly reuse a learning record to rank employees or assess performance. That would change the purpose and the implications of collection. The privacy and adoption-measurement guide covers the wider collection, access and reporting decisions.
Use the evidence to agree a specific repair and a review point:
- Name the unresolved boundary. For East, the relevant material is unavailable through its approved system.
- Assign the responsible action. Ask the knowledge and access owners to provide a permitted route, preserving existing restrictions.
- Check the affected task. Have a recipient try the current guidance on a suitable example after the repair.
- Update the transfer record. Preserve the earlier observation, record the new evidence and decide what still needs attention.
The receiving team's proposed cross-reference check needs a different action: review it, then use the guidance-maintenance process to update the source and relevant local versions. The night shift needs an opportunity for observation. Adding another all-hands demonstration would not answer either question by itself.
Bring those decisions to the knowledge review. The record has done its job when an owner can explain which boundary needs attention, what evidence supports the repair and how the next review will tell whether it helped.
Questions about measuring cross-silo knowledge transfer
Which metrics show that knowledge is spreading across departments?
Use measures tied to a defined knowledge item, relevant receiving groups and a stated period. Report the groups covered, the evidence of use, important adaptations and unresolved barriers. Keep reported use separate from reviewed examples and unknown results. Views and message counts describe activity; they cannot establish usable transfer alone. Start with the coverage discussion before combining observations into a percentage.
Can organizational network analysis prove that knowledge transferred?
Organizational network analysis can reveal patterns of contact or reported advice relationships, depending on its data and method. Those patterns help locate possible routes and gaps, but do not by themselves show what a recipient understood or used. Pair a relevant network question with task evidence and local context. The contact and evidence distinctions explain why more connections should not automatically count as better transfer.
What denominator should we use for a transfer rate?
First define the claim and the unit you can observe. For a review of one practice, list the relevant receiving groups and explain their selection. Report how many supplied which kind of evidence, including blocked access, no opportunity and missing follow-up. Do not divide reviewed outputs by employee headcount or call a selected sample company-wide adoption. The four-group example shows a useful summary without an unsupported success rate.
Does a local adaptation count as knowledge transfer?
A local adaptation can support transfer when its relationship to the source is clear and the receiving group can explain and demonstrate its use. Check whether the change preserves important conditions and quality checks. Merely copying or renaming a file establishes less. Record the source version, local difference and reviewed example, then return useful improvements to the owner through the directional transfer record.
How can we measure transfer without monitoring everyone's messages?
Begin with selected, relevant source-to-recipient relationships, short conversations and permitted examples of work. Explain the purpose, limit identifiable notes and report missing evidence honestly. This can identify obstacles without collecting everyone's communication history, although it does not provide a representative organization-wide rate. Use the collection and repair guidance to connect each requested observation to a decision someone can act on.
