Resolving customer service tickets end to end in 3 hours instead of 48

The challenge

This national sporting goods provider ran its entire customer service function out of Microsoft 365, with email as the primary case management channel. Every ticket was triaged manually, with agents updating spreadsheet records and writing responses by hand, creating bottlenecks at exactly the point customers were waiting on a reply.

Customer data sat scattered across multiple databases and sources with almost no automation between them, so agents were navigating several systems per case just to piece together the full picture before they could respond.

Routine, repetitive cases were eating the team's capacity, leaving little room for the complex escalations that actually needed human judgement. Response quality also varied by agent, with no guarantee that company policy was being applied consistently from one ticket to the next.

The client wanted agentic AI workflows, self-directed AI systems able to run the full customer service lifecycle from intake to closure, without losing human oversight of the outcomes.

Our approach

1. 01

   Establishing the Microsoft foundations

   We started by building the SharePoint architecture and standardised Power Apps intake forms the rest of the system would run on, giving every ticket a consistent structure from the moment it entered the system.

2. 02

   Building the AI-driven knowledge base

   We restructured the client's company policies, FAQs and historical tickets into a clean, structured corpus in SharePoint that the AI agents could reason over, rather than leaving that knowledge scattered across documents and inboxes.

3. 03

   Deploying the chained agent system

   We built a chained system of separate Classifier, Resolution, Validation and Escalation agents working in sequence via Power Automate, with Relevance.ai orchestrating the agentic workflows end to end, and rolled it out with staged automation rather than switching everything on at once.

4. 04

   Building in human-in-the-loop training

   Staff reviewed and corrected AI outputs during validation, with those corrections fed back into the system as continuous training, so accuracy and autonomy improved progressively rather than being fixed at launch.

5. 05

   Scaling automation once validated

   With the system validated through staged rollout, we scaled automation up to handle the majority of tickets unassisted, reserving human attention for the complex escalations that genuinely needed it.

The outcome

Average resolution time has dropped by 94%, from 48 hours down to 3 hours, with 67% of tickets now resolved end to end with no human intervention at all.

Every automated response aligns fully with company policy and brand voice, replacing the inconsistency that came from responses varying agent to agent, and agents have seen a 75% efficiency gain on the routine tickets that used to consume most of their time.

The system now handles thousands of cases concurrently with no additional headcount, and staff have been redeployed from routine ticket handling onto complex cases and strategic work. The client is now extending the same pattern to HR, procurement and IT support, alongside proactive monitoring agents that flag issues before customers report them and autonomous escalation routing based on specialist expertise matching.