Download the Case Study

When Customer Emotions Cost Millions: Fuel Company Turns to AI to
Transform Fuel Rewards Support

Managing millions of fuel rewards members across North America is about more than points and perks. It’s about protecting a multi-million-dollar revenue stream, where every negative interaction risks driving up operational costs, frustrating agents, and losing out to fierce competition. That’s why this Fortune 500 energy company zeroed in on the post-call experience. Post-call survey data showed that one in three escalated calls resulted in negative sentiment, with some members threatening to switch to a competing brand. Support agents were juggling incoming support requests, changing redemption rules, and new promotions. Meanwhile, the company’s traditional approach to agent training wasn’t scaling. The company turned to DATAMARK, its longtime BPO provider for contact center services, to execute an AIled shift: use advanced call summarization and emotion detection to help agents adjust their approach before escalation occurs.


By combining DataScribe’s live conversation intelligence with behavioral analysis, agents stood to gain something they’d always missed: a real-time emotional roadmap for every conversation.

A Confident Bet on AI-Driven Sentiment Analysis

Customers want assurance that the agent on the other end of a customer service call understands their needs. The client’s 30% escalation rate indicated a significant gap in this understanding. A high rate of agent burnout further underscored the operational costs of confrontational calls.

Solution testing revealed an opportunity to go beyond traditional sentiment analysis. DATAMARK’s implementation team conducted a two-week diagnostic phase. They analyzed call recordings, agent workflows, and post-call survey data.
The analysis revealed three critical friction points:

  • Post-Call Work Bottleneck: Support agents were spending 2-3 minutes after each call manually documenting customer issues, resolutions, and sentiment observations. With an average of 45-60 calls per agent per day, this resulted in 90-180 minutes of non-productive time.
  • Missed Escalation Signals: Quality monitoring covered only 3% of calls due to resource constraints. Post-call surveys showed that 30% of escalated interactions had exhibited early warning signs (shifts in tone, repeated questions, or frustration cues)that agents didn’t detect in time.
  • Cultural Communication Gaps: Offshore mix of Mexico and India-based support teams sometimes struggled with American English colloquialisms and emotional inflections. Agents reported feeling uncertain when customers used casual profanity or expressed enthusiasm.

Working closely with DATAMARK, the company greenlit an innovative pilot program. The solution centered around a combination of DATAMARK’s live-audio processing and AI-powered sentiment detection, courtesy of Microsoft Azure AI and Valence, two of DATAMARK’s trusted technology partners.

The goal of the pilot was to equip agents with:
Customers want assurance that the agent on the other end of a customer service call understands their needs. The client’s 30% escalation rate indicated a significant gap in this understanding. A high rate of agent burnout further underscored the operational costs of confrontational calls. A Confident Bet on AI-Driven Sentiment Analysis Actionable, real-time intelligence during service calls Personalized recommendations for de-escalation Scalable post-call coaching based on emotional patterns

Successful Pilot Charts Path for Post-Call Experience

Rather than a full-scale deployment, DATAMARK and the client agreed on a controlled six-month pilot with 150 agents handling fuel rewards inquiries— a high-volume, emotionally-charged segment where members often called frustrated about point discrepancies or redemption issues. The pilot implementation leveraged
a three-layer approach:

  • Audio capture & processing: DataScribe captures live audio locally on agent desktops. Real-time transcription processes conversations as they happen. Zero latency between customer speech and agent visibility.
  • Emotional intelligence layer: Valence AI analyzes tone, cadence, pace, and linguistic patterns. Contextual awareness separates confusion from frustration, excitement from agitation. Emotional score is determined on a dynamic scale:
  • Agent guidance system: Live coaching prompts appear based on emotional shifts. Suggested responses appear based on detected emotional states. Visual indicators show conversation trajectory.

In addition, the system would expand post-call intelligence to include automated summaries that highlight each customer’s emotional journey. It would now flag coaching opportunities for supervisors and provide trend analysis across customer segments. Supervisors would be coaching with emotional context—not just metrics—to help identify agent burnout risks and skills gaps.

This comprehensive approach transformed agents into emotional intelligence experts that could sense and respond to customers’ fundamental needs. The pilot evaluated these new capabilities across three measurement tiers:

Operations
AHT
Post-call time
Escalation rate

Quality
Call quality scores
First-call resolution rate
Post-call surveys (NPS)

Experience
Agent stress
Agent confidence
Agent attrition rate

Over the course of the six-month pilot, escalation rates dropped from 14% to 6%; average handle time for emotional calls decreased by 6% and agent confidence scores rose, with agents reporting feeling “equipped to handle any situation.”

AI Assistance that Benefits Humanity: Real-Time Emotional Intelligence at Scale

Based on these results, the client approved full deployment across all fuel rewards agents. What started as an experiment in emotional intelligence had proven itself as a critical competitive advantage in the fuel market.
The decision to green-light this pilot program marked a deliberate strategic choice: use emotional intelligence to empower agents,
not audit them. Close the gap between customer sentiment and agent response
through AI-enhanced human empathy.
Notably, offshore stigma dissolved as emotional intelligence helped agents bridge cultural gaps:


• Agents in Mexico and India achieved identical satisfaction scores to US teams


• Regional expressions and colloquialisms, no longer created confusion


• Customers stopped asking “where are you located?” when they felt understood

Today, the fuel company plans to extend emotional intelligence capabilities to chat, email, and social channels. Decision-makers from other business units have expressed interest in adopting emotional intelligence pilots as well after hearing of the pilot program’s success.
When executed properly, sentiment analysis becomes the connective tissue between agent performance and customer trust, empowering agents to steer customer experience in a variety of operational contexts.