AI adoption
Retire old instructions before adding AI.
Give an AI-assisted workflow a clear set of current instructions, an accountable owner and a way to handle conflicting information.
Consider a hypothetical service team with three versions of its cancellation process: a shared document, a saved email and a training slide. Each looks plausible. Adding an AI assistant to search all three does not decide which instruction the business has approved. That decision needs an owner.
Find the disagreement first
Choose one recurring question the team needs to answer. Ask two people to find the instruction they would use, without sharing their starting point. Compare the wording, conditions and dates. A difference is useful evidence of a problem to resolve before connecting more material to an AI workflow.
Keep this exercise narrow. You do not need to reorganise every shared folder before making progress. Start with information where a wrong answer could create a customer commitment, rework or an avoidable escalation.
Make approval visible
For each instruction in scope, identify the person authorised to confirm it. Record the approved location, the date it takes effect, the situations it covers and the event that should trigger another review. A recent upload date alone does not prove that the business instruction is current.
- Owner: who can resolve a disagreement?
- Scope: which service, team or situation does this cover?
- Status: approved, draft or superseded?
- Review trigger: what change would make this instruction unreliable?
Separate current guidance from history
Keep superseded material out of the collection used for current operational answers. Preserve historical records where needed, with a clear label and an appropriate location. The aim is to stop yesterday’s instruction competing with today’s, while retaining a traceable record of what changed.
In the hypothetical cancellation example, the owner confirms the current process and labels the previous version as superseded. Links in the team’s working checklist are updated too. Otherwise, people may continue to follow an old bookmark even when the AI collection has been corrected.
Define what happens when sources conflict
Where the tool supports source links, show the relevant approved instruction alongside the proposed answer. Tell staff what to do when the source is missing, conflicting or outside its stated scope. For consequential answers, hold the response and ask the named owner instead of selecting whichever version sounds most convincing.
Try that route with synthetic examples: an ordinary question, a question covered only by an old document and a question on which two sources disagree. Check whether the workflow points to the right source or stops for clarification. Do not assume a written instruction to the AI guarantees that behaviour.
Keep the collection owned
Add source updates to the same change process used to approve the business instruction. Name who updates the collection and related links, who checks the resulting answer and how staff hear about the change.
Rethink’s AI review and enablement can help define this manageable starting point, with ongoing adoption support scoped around the work your team needs to maintain.