The ask, evaluate and audit the SharePoint Online (SPO) environment before the Microsoft 365 Copilot rollout, Microsoft‘s artificial intelligence (AI) assistant for the M365 apps. The top-most concern of the business leaders is whether or not their content is permissioned properly. Though, a security audit could easily help identify SPO sites with lax sharing and/ or overly exposed content. An often-overlooked concern and harder question to answer, “Is the SPO data healthy enough for Copilot?”
Microsoft 365 Copilot, like any AI, thrives on data. Unfortunately, not all data is good data. As the saying goes, “Garbage in, garbage out.” If AI ingests bad data, then it can generate bad prompt responses. This bad “data in” can sometimes be categorized as ROT:
- Redundant
- Obsolete
- Trivial
What makes data redundant? Simply, there are file duplicates stored in multiple SPO folders, libraries, and/ or site collections. As a consequence, this results in there no longer being a single source of truth for the data, which could also complicate citing of sources.
Obsolete data is data that has outlived its usefulness. This content should ideally be purged because Copilot could ingest it and provide incorrect responses to prompts. Imagine asking Copilot to summarize recent tax codes and it includes answers for the 2004 tax codes simply because those files were never deleted.
Lastly, trivial data. This is data stored in SPO but offers no business value. Not because it is redundant or obsolete, but because it doesn’t directly contribute to the organization’s overall productivity. Vacation photos are examples of trivial data, at least in terms of business value. These are great to share with colleagues, but should they be ingestible by Copilot? Should personal medical records be stored in SPO and ingestible by Copilot?
Conclusion:
Microsoft 365 Copilot, as with any AI, thrives on data, but the quality of the data directly impacts prompt responses. Redundant data could make citing sources tricky, obsolete data run the risk of providing incorrect prompt responses, and trivial data serves no business purpose and shouldn’t be ingested by the AI. Cleaning up as much ROT as possible makes the data more healthy and ready for proper AI adoption.
“A democracy cannot thrive where power remains unchecked and justice is reserved for a select few. Ignoring these cries and failing to respond to this movement is simply not an option — for peace cannot exist where justice is not served.”
John Lewis
#BlackLivesMatter
console.log('Charles'); 