Fortune 500 Customer Service Provider

Generative AI conversational assistant for 5179 customer support agents

Read primary source

Maturity stage

Production

Use-case type

Augmentation

Function

Customer Service/Support

Company size

Enterprise

Evidence

14% productivity increase; 34% improvement for novice workers

ROI / outcome figure

14% avg productivity gain; 34% for novices

Deep dive

The setup

An NBER-published field study at a Fortune 500 customer-service provider deployed a GenAI conversational assistant to 5,179 agents handling support cases.

What happened

Productivity rose 14% on average. Novice agents - those in their first months on the floor - improved 34%. Tenured agents barely moved. The intervention compressed the experience curve: AI made beginners perform like mid-tenured staff.

Root cause

Why this worked, when so many GenAI deployments do not, comes down to scope discipline. The use case was tightly bounded (one task, one channel, measurable resolution time), the workflow was already digital, and the KPIs were instrumented before rollout. Every condition for IS success per DeLone & McLean was in place.

Takeaway for teams considering similar work

The novice-vs-tenured split is the most replicated finding in the GenAI productivity literature - it shows up in BCG's consulting trial, GitHub Copilot studies, and Microsoft 365 Copilot TEI. If a deployment is failing to show productivity gains, the first question is whether the experience-distribution skews tenured.

Pattern fit

Anchors most of the AI-assist time-saved figures in the Task Mapper, especially for Q&A and explanatory work.

Why this case is cited as evidence

IMeasurement Maturity
Are there established KPIs, baselines, and evaluation cadences?