Education
Key Improvements
Good-response rate jumped to 82.5% — up from under 10% at launch, in just 90 days of AgentGuard-monitored tuning
89% of case volume moved through — without a human needing to solve it from scratch
Production-first optimization — live shadowing let the team iterate in real conditions instead of a prolonged UAT cycle
In 90 days, GBP’s Agentforce agent went from under 10% to 82.5% good responses, using Diabsolut’s AgentGuard optimization model.
The Challenge
George Brown Polytechnic’s student support team was handling a high volume of low-complexity inquiries manually, creating unnecessary strain on staff capacity, inflating cost per case, and introducing risk of poor student experience during peak periods. With a growing student population, the college needed a scalable solution to deliver consistent, timely support without overburdening its team.
Manual responses for low-complexity cases led to higher-than-necessary average handle time, higher cost per case, and a risk of a poor support experience when the support team had low capacity.
Pain Points
- Repetitive, routine questions consumed a disproportionate share of agent time, pushing average handle time higher than it needed to be
- Manual handling of those same routine cases kept cost per case elevated
- Thin coverage during peak periods put the quality of the student experience at risk
Requirements
- Deflect low-complexity inquiries automatically through email-to-case, drawing on the college’s existing Knowledge Articles
- Reduce the cost per case without sacrificing support quality
- Give agents room to focus on the complex, high-value interactions that need a human
- Guarantee round-the-clock coverage that doesn’t depend on staffing levels
The Solution
Diabsolut implemented Salesforce Agentforce for Email-to-Case at George Brown Polytechnic, configuring an AI agent to automatically handle inbound student inquiries by leveraging GBP’s existing Knowledge Articles. Diabsolut’s proprietary AgentGuard monitoring and optimization model was used to continuously measure, tune, and improve the quality of agent responses throughout the engagement.
Rather than waiting out a prolonged UAT cycle, the team made the strategic decision to go live in production — with the agent shadowing service reps by documenting suggested responses on each case for rep review and feedback. This human-in-the-loop approach enabled rapid iteration and measurable improvement in real-world conditions.
Services & Products Implemented
The Results
By bringing Agentforce to George Brown Polytechnic’s student support operations, Diabsolut has helped the institution move from reactive, manual case handling to an always-on, AI-powered service model — one that scales with student demand, reduces operational costs, and keeps human agents focused on what matters most. What’s next: expanding the agent’s scope to additional inquiry types and channels as confidence in AI response quality continues to grow.
Agent Quality & Optimization
- Agent response quality: using the AgentGuard monitoring and optimization model, GBP’s Agentforce agent went from a good-response rate of under 10% to 82.5% in 90 days
- Production-first optimization: rather than staying blocked in UAT, Diabsolut moved the agent to a live shadowing model, enabling faster real-world iteration
- Knowledge base & guardrail optimization: filled Knowledge Article gaps, added missing data categories, and implemented stronger guardrails to optimize for cost and scope
Student Experience & Support Capacity
- Always-on support: students now receive immediate, AI-generated responses to low-complexity inquiries at any time
- Agent capacity freed: support staff can now focus exclusively on complex, high-value student cases
Operational Impact & Case Deflection
- GBP launched externally on April 3, 2026. Case volume was tracked April 18 through July 20, 2026 (6,433 cases) once the agent went fully autonomous:
- Of those 6,433 cases, the agent closed 38% on its own, correctly routed 51% straight to the right team (outside its scope by design), and escalated 11% to a human agent
- Combining the cases it closed and the cases it routed correctly, 89% of volume moved through without requiring a human to diagnose or triage it first
- Reduction in average handle time and improvement in the cost per case metrics are still being finalized by the client – stay tuned for updates