The workflow before
Exelon serves 10.7 million customers through six utilities. What customers said lived in separate channels: call center transcripts, emails, web chats, news coverage and social media. Marketing relied on lagging surveys and manual reading, so leadership had hypotheses about what drove reputation but no data to test them, and issues were spotted late.
What we built
We built an NLP application that pulls every customer touchpoint into one data lake: call center transcripts, email and chat logs, transaction data, earned media and social posts from Twitter, Facebook and LinkedIn.
Models classify each item by topic, score sentiment with 87% accuracy, extract entities and cluster related conversations, so the team sees which issues (billing, outages, service, rates) are moving sentiment in which territory.
Results land in Elasticsearch with Kibana dashboards for executives and the communications team, plus alerts when sentiment shifts, cutting the time to spot and respond to a reputation issue by 40%.
Before and after
| Measure | Before | After |
|---|---|---|
| Time to spot and respond to an issue | Baseline | 40% faster |
| Sentiment read | Lagging surveys and manual reading | Every channel classified, 87% accuracy |
| Topic tagging precision | Manual | 92% |
| Marketing campaign effectiveness | Baseline | 20% better targeting |
What's next
The next step is L3: an agent that drafts the response or outreach when an alert fires and routes it to communications for approval.
