Have AI propose a source image for a difficult procedure step
A proposed AI media assistant searches an approved asset library for a filter-cartridge orientation step. The step’s part revision and task relationship help it propose a view of the sealing edge, inlet, and latch rather than a similar-looking obsolete part. Service engineering verifies the source image; writers annotate it and technicians test the revised procedure.
See the business case- Business goal
- Reduce costs
- Business metric
- Self-service rate
The task and the failure point
AI asset retrieval should find a view that supports the cartridge orientation task, not just an image named “filter.” The maintained step-to-part-revision relationship and approval metadata let reviewers validate the proposed source before annotation.
Adding media everywhere increases maintenance work and can hide the next step. The useful question is where a user must recognize a part, orientation, state, or motion that words alone do not convey well. A media management workflow selects, validates, and versions an asset for that decision point.
The content repairPlace an approved annotated image at the orientation step, with equivalent text: sealing edge toward the inlet; slide until the latch clicks. Service engineering must verify the actual revision B part geometry.
AI asset proposal from an unscoped library
“Use this filter-cartridge image.” A filename or visual resemblance does not establish revision B, inlet direction, rights, or approval.
AI proposal with task and revision evidence
“Candidate view for the revision B insertion step,” with its library ID, source, revision, and approval record. Service engineering checks that the sealing edge, inlet, and latch are visible before writers label the verified view.
A comparison worth testing
Run the same orientation-step request with the same AI model, prompt, and library snapshot, first without task/revision relationships and then with them. Review whether the proposed existing asset matches revision B and shows the required landmarks. Separately compare technician task completion using the verified image and equivalent text against text alone.
The step record below gives AI a retrieval request tied to cartridge revision B and the orientation decision. It also tells the reviewer what the proposed image must show before its labels and equivalent text can be approved.
What the content must carry
Step: Insert the cartridge
Place the sealing edge toward the housing inlet. Slide the cartridge until the latch clicks. The adjacent annotated image labels the inlet, sealing edge, and latch; its alt text conveys the same orientation. Asset owner: service engineering. Applies to: cartridge revision B. Replace or reapprove when the housing or cartridge changes.
From content work to self-service rate
A useful AI asset proposal becomes an approved, accessible visual that helps a technician install the cartridge correctly. Record asset rejection and review effort alongside Media freshness coverage and Media accessibility pass rate; observed task completion is needed before assessing Self-service rate.
How to measure it
Save AI asset IDs, proposed reasons, model, prompt, and library metadata snapshot. Engineering scores revision fit and landmark visibility, including obsolete lookalikes. Run matched text-only/enriched task sessions with approved imagery; count completed installations, errors, and help requests rather than asset views.
Start with a bounded task
Try AI asset retrieval for one observed high-friction step. Review the proposed source IDs with engineering, add labels only to a verified view, and test the finished procedure with technicians.
See the implementation steps Assess your AI readinessWhich step forces readers to guess what they should see?
Intuitive Stack can help connect AI asset proposals to procedure steps, part revisions, and approval evidence, then test the verified visual with technicians.