A large financial-services client needed a high-impact AI pilot that could improve a real Product Training workflow while operating within corporate and federal guardrails.
I chose learning intake: a repetitive, high-friction process where a guided consultation could gather better information, suggest a starting point, and route a clearer request without replacing instructional-design judgment.
The Product Training team needed a credible use of AI inside a guardrailed environment. Learning intake offered a focused opportunity to reduce friction, improve request quality, and demonstrate value without handing decisions to the technology.
I identified the workflow opportunity, deconstructed the existing intake process, and designed a governed Copilot approach around a short needs analysis, transparent recommendations, and human confirmation.
The end-to-end workflow, guided interview, controlled decision logic, confirmation points, modality and non-training cues, and structured handoff to the instructional-design team.
Instructional designers begin with clearer information and spend less time chasing requestors, while the organization gains a practical AI pilot that respects its guardrails and keeps final judgment with people.
A guided interview captures the business need, desired behavior, audience, complexity, risk, practice requirements, tracking needs, timing, and existing resources.
The consultation is grounded in approved guidance so recommendations reflect the organization's learning standards rather than an open-ended model response.
Transparent rules identify likely modalities. AI explains why the cues point toward a recommendation; it does not invent an unconstrained solution.
The SME reviews the captured request and preliminary recommendation, corrects missing context, and confirms the record before routing.
A Power Automate or agent flow sends the confirmed fields to an Excel table or SharePoint List with a request ID and timestamp.
The instructional-design team reviews the request, follows up where needed, and owns the final modality and solution decision.
Set up the Copilot Studio agent in Teams with a focused learning-intake purpose and clear boundaries.
Build a guided conversation that gathers the inputs needed for triage and flags missing or ambiguous details.
Connect approved learning standards, modality definitions, intake criteria, and escalation guidance.
Translate modality cues into transparent rules and let AI explain the preliminary recommendation.
Present the request summary and recommendation for correction and explicit confirmation.
Create a request ID and timestamp, then write structured fields to a SharePoint List or Excel table.
Track ownership, status, urgency, follow-up, final ID decision, and notes; notify the instructional-design team.
Test clear, ambiguous, blended, and non-training cases; refine questions, rules, explanations, and escalation points.
The rules surface a likely starting point without making the final instructional-design decision.
Best when the task is performed in the flow of work, steps are stable, recall is not essential, and people need fast performance support.
Useful for a new tool or process with a short path to first success, especially when learners can continue exploring after a concise orientation.
A likely fit when delivery must scale, foundational knowledge is consistent, self-paced access matters, or completion and knowledge checks must be tracked.
Stronger when learners need discussion, coached practice, feedback, collaboration, nuanced judgment, or facilitated application to real scenarios.
Appropriate when the gap comes from unclear expectations, missing tools, process friction, access, incentives, staffing, or another condition training cannot solve.
The agent triages the request; the instructional designer owns the final recommendation. As intake volume and workflow complexity grow, a SharePoint List provides a stronger operational foundation than Excel.
The result is a governed front-end consultation—not a chatbot that merely fills a form: guided discovery → transparent recommendation → human confirmation → structured routing → ID judgment.