Skip to main contentDaniel Nash

Case Study

Contact Center ChatGPT Pilot - AI Copilot for Agents

Designed and launched a 6-month ChatGPT Enterprise pilot for Guitar Center's contact center, driving a $2.7M annualized impact and proving the business case for scaling AI.

Annualized Impact
$2.7M+
Pilot Duration
6 months
Pilot Cohort
15 agents
AI Product ManagementChatGPT EnterpriseContact CenterExperimentationChange Management

Snapshot

Role
Senior Product Manager - Contact Center & Order Management
Company
The Guitar Center Company
Timeframe
June 2024 - November 2024
Scope
15 agents in pilot cohort with a matched control group
Objective
Validate whether AI copilots improve revenue per call, efficiency, and experience without sacrificing brand voice or data security.

Context & problem

Agents juggled multiple tools, manually drafted emails and call scripts, and spent valuable time preparing for every interaction. Leadership wanted to know if AI could serve as a safe, reliable copilot that improved performance without risking customer trust.

  • Help agents prepare before calls with faster access to relevant context.
  • Draft on-brand emails in minutes instead of grinding through templates and copy/paste loops.
  • Deliver measurable lift in Revenue Per Call, Items Per Transaction, and Average Order Value.
  • Respect Guitar Center's brand voice and data governance requirements.

My role

  • Framed the opportunity, success metrics, and business case for piloting AI inside the contact center.
  • Partnered with EIS and Legal to ensure security, privacy, and compliance were baked into the pilot from day one.
  • Facilitated co-design sessions with supervisors and agents to capture real workflows and pain points.
  • Defined and iterated on customGPT instructions, tone guidance, and workflow prompts.
  • Designed the experimental setup, including control group selection and measurement methodology.
  • Socialized pilot results and recommendations with senior leadership to green-light expansion.

Approach

The pilot focused on shipping working workflows quickly - balancing agent creativity with the guardrails needed for a retail contact center.

Brainstorming sessions -> focused customGPTs

  • Partnered with supervisors and frontline agents to surface their highest-friction jobs-to-be-done.
  • Converted the insights into purpose-built customGPTs, including a call-prep assistant and brand-safe email helpers.
  • One of the pilot customGPTs became the most-used customGPT inside the company, proving we solved a real agent problem.

Safe use of CRM data & security review

  • Collaborated with EIS to review ChatGPT Enterprise security and privacy posture.
  • Agreed on guardrails for pasting basic CRM context (no PCI or PII) and double-checking generated copy before sending.
  • Documented do/don't guidance so agents understood exactly how to work safely.

Agent workflow design: call scripts and emails

  • Defined repeatable workflows for pre-call prep and follow-up emails using structured prompts.
  • Enabled agents to turn CRM snippets into concise call plans that emphasized solution paths and relevant product tie-ins.
  • Accelerated outbound communications without automating call summaries - keeping humans in the loop where it mattered most.

Measurement & experiment design

Retail seasonality and promotional spikes make simple before/after comparisons unreliable. We designed a matched control methodology to keep the signal clean.

  • Formed a 15-agent ChatGPT test group and a matched control group with similar tenure, skill, and call mix.
  • Measured growth deltas between groups to control for seasonality instead of relying on simple before/after comparisons.
  • Primary metrics: Revenue Per Call, Items Per Transaction (IPT), Average Order Value (AOV).
  • Efficiency signals (handle time, after-call work) influenced the $2.7M annualized impact model - ChatGPT agents consistently outperformed control across each vector.

Key outcomes

$2.7M annualized impact

Pilot agents outpaced the control group across Revenue Per Call, IPT, and AOV while reducing time spent on prep and follow-up.

Proof that AI can work safely with CRM context

EIS and Legal signed off on using basic CRM data within ChatGPT Enterprise under the guardrails defined in the pilot.

CustomGPT adoption

One of the pilot customGPTs quickly became the most-used customGPT across the company, validating product-market fit for agent workflows.

Foundation for scaling

Results unlocked a decision to roll ChatGPT Enterprise out to the entire contact center (~350 agents) and set the stage for org-wide expansion.

What I'd do next

  • Turn the most successful customGPTs into task-specific mini-agents embedded directly in the agent desktop.
  • Expand the measurement framework to capture customer sentiment and first-contact resolution alongside revenue metrics.
  • Integrate with OMS and knowledge-base systems in a human-in-the-loop pattern that keeps agents accountable for final outputs.

Interested in building something similar?

Let's talk about your AI roadmap and how to launch or scale copilots responsibly.