Mortgage servicers handle millions of payoff quote requests every year. Most call centers treat them all the same—quick in, quick out, move to the next call. Efficiency metrics look great.
But here's the problem: Only 15-20% of those requests actually convert to payoffs. The other 80%? They're tire-kickers, rate shoppers, or people just curious about their balance. And you're spending the same amount of time on both.
The Real Problem (What Most People Miss)
Call centers optimize for the wrong metric. They focus on average handle time (AHT)—getting people off the phone quickly.
But that's backwards. You should want to spend more time with customers who are actually ready to refinance or pay off their loan. Those conversations drive revenue. The 3-minute call that converts is worth more than ten 90-second calls that go nowhere.
The real problem isn't efficiency—it's that we can't tell high-intent customers from low-intent ones until we've already wasted time treating them identically.
My Approach
Step 1: Build an intent detection model using NLP
- Analyze call transcripts to identify language patterns that signal true intent
- Train ML model on historical data: Which calls converted vs. which didn't?
- Score incoming calls in real-time on 0-100 intent scale
Why? Because intent reveals itself in how people ask questions. "I'm closing on my new house next Tuesday" signals different intent than "Just wondering what my payoff would be."
Step 2: Layer in contextual data beyond the call
- Check if customer recently applied for a new loan
- Look at web activity: Did they use the payoff calculator? How many times?
- Consider account history: Payment behavior, equity position, interest rate
Why? Words alone don't tell the full story. A customer with 40% equity and a 7% rate asking about payoff is very different from someone at 95% LTV with a 3% rate.
Step 3: Route calls dynamically based on intent score
- High intent (70-100): Route to specialized agents with authority to negotiate
- Medium intent (40-69): Standard process with option to escalate
- Low intent (0-39): Automated IVR or junior agents, optimize for speed
Why? Not all calls deserve the same treatment. Your best agents should focus on your most valuable opportunities. That's resource allocation 101.
Step 4: Create intervention playbooks by intent tier
- High intent: Offer rate match, expedited processing, dedicated support
- Medium intent: Educate on benefits of staying, address concerns
- Low intent: Provide quote quickly, minimal retention effort
Why? Different intent levels need different strategies. Don't waste retention dollars on people who were never leaving anyway.
Step 5: Measure what actually matters—conversion rate and revenue per call
- Track: Payoff conversion rate by intent tier
- Calculate: Revenue impact of retention vs. cost of intervention
- Monitor: Did high-intent customers get appropriate attention? Did low-intent calls get handled efficiently?
Why? AHT is a vanity metric. What matters is: Did you save the customers worth saving? And did you do it cost-effectively?
The Contrarian Insight
Call centers optimize for speed. They should optimize for intent.
One high-intent call where you save a customer is worth ten low-intent calls you handle quickly. But most contact centers can't tell the difference until it's too late.
The future of call center operations isn't faster service—it's smarter routing. AI doesn't replace agents. It helps them focus their time on the conversations that actually matter.
Next Steps (First 90 Days)
Month 1: Data Collection & Baseline
- Pull 6 months of call transcripts for payoff requests
- Tag which calls converted vs. didn't convert
- Interview top-performing agents: How do they identify serious buyers?
- Baseline metrics: Current conversion rate, AHT by outcome, retention success rate
Month 2: Build & Test Intent Model
- Train NLP model using GPT-4 or Claude with fine-tuning on your data
- Incorporate contextual signals (web activity, loan characteristics, etc.)
- Target: 75%+ accuracy in predicting conversion within first 30 seconds of call
- Build simple dashboard showing real-time intent scores
Month 3: Pilot Smart Routing
- Pilot with 20% of calls—route by intent score
- A/B test: Intent-based routing vs. standard FIFO routing
- Measure: Conversion rate lift, revenue per call, agent satisfaction
- Refine playbooks based on what agents learn from high-intent calls
Key Metrics I'd Track:
- Intent prediction accuracy (compare predicted vs. actual outcomes)
- Conversion rate by intent tier (should see clear separation)
- Average handle time by intent tier (high-intent should be longer—and that's good!)
- Revenue per call by routing method
- Customer satisfaction scores by intent tier
- Cost to serve per customer retained
Why This Approach Works
At LoanCare, we deployed real-time NLP routing in Amazon Connect that lifted conversions by 18 percentage points and reduced average handle time by 22%.
Wait—how did we improve both conversion AND efficiency? Shouldn't those trade off?
Here's the secret: We spent more time on high-intent customers and less time on low-intent ones. The average went down because we stopped wasting time on tire-kickers. But our retention rate on serious buyers went way up.
That's the power of smart routing. You're not optimizing for average performance across all calls. You're optimizing each call based on its actual value.