68% Ticket Deflection: How We Achieved It with Workday AI Agents

When we first quoted 68% ticket deflection to prospective clients, the reaction was skepticism. Most organizations running Workday AMS have tried chatbots before -- FAQ bots that answer three questions well and frustrate users on everything else. Those bots achieve 15-20% deflection at best. Our number seemed implausible. This article explains exactly how we achieve 68% deflection, what makes our approach different from failed chatbot deployments, and what the remaining 32% of tickets look like.

---

Why Previous Chatbot Attempts Failed

Most Workday chatbot deployments fail for the same reasons. They rely on keyword matching or intent classification with a small set of predefined responses. They cannot access real-time Workday data to provide personalized answers. They have no context about your specific Workday configuration. And they cannot take actions -- they can only point users to documentation that may or may not answer their actual question.

The result is a chatbot that can answer generic questions (What is Workday? How do I log in?) but fails on the questions that actually generate support tickets (Why was my bonus calculated this way? Why is my time-off balance wrong? Why did my expense report get rejected?). These specific, personalized questions account for 70-80% of support ticket volume -- and they are exactly the questions that FAQ-based chatbots cannot answer.

---

What Makes 68% Possible

Our approach differs from failed chatbot deployments in four fundamental ways:

1. Real-time Workday data access. Our AI agents query your Workday tenant in real time to answer questions with actual data. When an employee asks why their bonus was calculated a certain way, the agent pulls their compensation data, checks the bonus plan rules, runs the calculation logic, and explains the specific result. This is not a canned answer -- it is a personalized explanation based on their actual data.

2. Tenant-specific context engineering. Our agents are loaded with context about your specific Workday configuration -- your business processes, your custom fields, your integration mappings, your organizational policies. When an employee asks about their time-off balance, the agent understands your specific accrual rules, carryover policies, and plan configurations. The answer is specific to your organization, not generic Workday documentation.

3. Action capability. Our agents do not just answer questions -- they take actions. Need to submit a time-off request? The agent can initiate the Workday business process. Need to update your emergency contact? The agent can walk you through it step by step, or in some configurations, make the update directly. This action capability eliminates the category of tickets where someone knows what they want to do but needs help doing it.

4. Intelligent escalation. When the agent cannot resolve a request -- because it requires human judgment, involves a complex configuration change, or the employee is frustrated and wants to talk to a person -- it escalates seamlessly. The escalation includes full context: what the employee asked, what the agent determined, what it tried, and why it is escalating. The human consultant picks up with complete context instead of starting from scratch.

---

The Ticket Deflection Breakdown

Our 68% deflection rate breaks down across ticket categories:

Information requests (90% deflection). Questions about pay statements, benefits details, time-off balances, policy information, and how-to guidance. These account for roughly 35% of total ticket volume. Our agents deflect approximately 90% of these because they can access real-time data and provide personalized, accurate answers.

Simple actions (80% deflection). Requests to update personal information, submit time-off requests, check approval status, and perform routine transactions. These account for roughly 25% of total ticket volume. Our agents deflect approximately 80% by guiding users through the action or performing it on their behalf.

Troubleshooting (55% deflection). Issues where something appears wrong -- incorrect pay, unexpected deductions, missing data, failed submissions. These account for roughly 25% of total ticket volume. Our agents deflect approximately 55% by diagnosing the root cause and either explaining the correct behavior or identifying and resolving the actual issue.

Configuration and complex changes (15% deflection). Requests that require Workday configuration changes, complex investigations, or policy decisions. These account for roughly 15% of total ticket volume. Our agents deflect only about 15% of these (the simpler ones that match known patterns), routing the rest to human consultants with full diagnostic context.

The blended result: (0.35 x 0.90) + (0.25 x 0.80) + (0.25 x 0.55) + (0.15 x 0.15) = 0.315 + 0.20 + 0.1375 + 0.0225 = 0.675, or approximately 68%.

---

What the Remaining 32% Looks Like

The tickets that reach human consultants after AI deflection are fundamentally different from pre-AI ticket queues. They are harder, more complex, and require genuine human expertise. This is intentional -- AI handles the routine so that humans focus on the complex.

The remaining 32% typically includes configuration change requests that require architectural decisions, complex troubleshooting where root causes span multiple systems, policy questions that require business judgment rather than factual answers, new requirements that have not been encountered before, and situations where the employee explicitly requests human assistance.

Consultants working on these tickets report higher job satisfaction because they are solving interesting problems rather than answering the same questions repeatedly. This also reduces consultant turnover, which is a significant hidden cost in traditional AMS models.

---

Implementation Timeline

Weeks 1-2: Knowledge base construction. We ingest your Workday documentation, extract configuration details from your tenant, and build the tenant-specific knowledge base that powers accurate responses.

Weeks 3-4: Agent configuration and training. We configure the AI agents for your specific environment, set up Workday API integrations for real-time data access, and validate accuracy against historical ticket data.

Weeks 5-6: Controlled deployment. We deploy to a subset of users (typically one business unit or location) and monitor accuracy, deflection rates, and user satisfaction. We iterate rapidly based on real usage patterns.

Weeks 7-8: Full deployment and optimization. We roll out to all users, continue monitoring and optimizing, and establish the ongoing feedback loop that drives continuous improvement in deflection rates.

---

Sustaining and Improving Deflection

68% is not a ceiling -- it is where most clients start after initial deployment. Deflection rates typically improve to 72-75% over the following six months as the system learns from escalated tickets and expands its resolution capabilities. The improvement comes from three sources: new knowledge added from resolved tickets, expanded action capabilities as confidence grows, and refined retrieval that better matches user queries to relevant context.

---

Key Takeaways

• 68% ticket deflection is achievable with the right architecture -- real-time data access, tenant-specific context, action capability, and intelligent escalation.
• Previous chatbot failures occurred because FAQ-based systems cannot answer the personalized, data-dependent questions that generate most tickets.
• Deflection rates vary by category: 90% for information requests, 80% for simple actions, 55% for troubleshooting, 15% for complex changes.
• The remaining 32% of tickets are genuinely complex and benefit from focused human expertise.
• Implementation takes 6-8 weeks with results visible from week 5.

---

AssistNow achieves 68% ticket deflection for Workday support through our Assistly AI platform. Contact us to see a demo with your actual ticket categories.