Agentic Pipeline for Gap Analysis
- Copilot Studio
- Power Automate
- AI Builder
- Encodian
- →Replaces a one to two day manual analysis per organisation
- →Survey to JSON to PowerPoint to PDF, routed for human approval
- →Cost modelled at roughly $1 to $3 per report depending on volume
Context
An internal team ran workplace-learning gap analysis by hand: read a 28-question survey from each partner organisation, then write up a gap analysis and recommendations report, one to two days of work per company. Under an institution-wide push to use generative AI to cut manual work, they asked our team to build an agentic system to do it. It is aimed at new partner organisations and is currently in the review and production-readiness phase, with the pilot yet to run a live case.
The pipeline
I built it with Copilot Studio, Power Automate, AI Builder, and Encodian. A survey submission is picked up by a Power Automate orchestrator and passed to a Copilot Studio agent, which returns a structured JSON gap analysis and set of recommendations. Encodian renders that into a formatted PowerPoint, which is converted to PDF and emailed to a reviewer for approval before anything reaches the end user.
Encoding the knowledge base
The core of the work was turning the team's reference material, its knowledge base and assessment rubrics, into system instructions the model could apply reliably. It was not a one-time setup: as reports came back for different industries I folded reviewer feedback into the prompt for that industry and used worked examples to pull the model toward the expected tone, structure, and level of detail.
Operating economics
I also ran a cost analysis. From average token use across ten sample reports and AI Builder pricing, the pipeline caps at about 125 reports a month under current settings. Cost per report falls with volume: over $3 at low volumes, around $2 to $3 at roughly 50 reports a month, and just over $1 at the 125-report ceiling. The next step is moving from pilot into full production.