A 45-agent Medicare insurance call center came to us doing 42,000 outbound attempts per day with a 3.2% connect rate. Fourteen days later they were at 7.1%. That's not a typo — connect rate more than doubled, daily sales went from 114 to 235, and agent connects jumped from 30 to 65 per day.
I want to walk through what we actually changed and in what order, because the sequence matters more than people realize. Three categories of optimization drove the result, and each one had a different magnitude of impact.
The Starting Point
The center had been on VICIdial for three years with a self-hosted setup managed by an in-house IT generalist. Their numbers looked acceptable if you didn't dig:
| Metric | Value |
|---|---|
| Daily outbound attempts | ~42,000 |
| Daily live connects | ~1,344 |
| Connects per agent per day | ~30 |
| Average talk time per connect | 4.2 minutes |
| Sales conversion rate (of connects) | 8.5% |
| Daily sales | ~114 |
| Revenue per sale | $320 (first-year commission) |
| Daily revenue | ~$36,480 |
The numbers looked "okay" in isolation. The problem was they were leaving massive amounts of money on the table — agents spending more time waiting than talking, answering machine detection killing live calls, and DID reputation in the gutter.
The Diagnosis: Three Things Were Broken Simultaneously
When we audited their VICIdial setup, we found the same trio of problems we see in roughly 80% of underperforming centers:
1. AMD was killing live humans (estimated 35% of lost connects). Their answering machine detection was running default parameters that weren't tuned for their demographic — Medicare-eligible adults, age 65+. Older people take longer to pick up the phone. They pause more when they answer. The system was misclassifying about 9.3% of all calls as answering machines when they were actually live humans. At 42,000 daily attempts, that's approximately 3,900 wasted contacts per day.
We confirmed this by looking at call duration patterns: calls dispositioned as "Answering Machine" that lasted only 1-4 seconds. If it were truly a voicemail greeting, you'd expect the call to last longer. Short-duration AMD dispositions are the signature of false positives — the system hung up on a live person before they finished saying "Hello?"
2. Every single outbound DID was flagged as spam (estimated 40% of lost connects). Twenty local numbers, all in service for over a year without rotation, no health monitoring, no daily call caps. We checked every one against the major carrier analytics databases — Hiya, Nomorobo, First Orion, TNS Call Guardian. The worst DIDs had answer rates below 2%. The best were at 5.8%. A clean DID in their market should hit 8-12%.
3. Fixed-ratio dialing was wasting leads (estimated 25% of lost connects). They were running RATIO mode at a fixed 3.0 dial level. At shift start when all 45 agents log in simultaneously, this creates a massive spike — 7.8% drop rate in the first hour, well above the 3% TCPA ceiling. Later in the day when agents are spread across various states (talking, paused, wrapping up), the same 3:1 ratio was too conservative, leaving agents waiting 15-25 seconds between calls.
What We Changed (And What Moved the Needle Most)
We tackled these in sequence over 14 days, measuring the impact of each category before moving to the next. This is important: if you change everything at once, you can't attribute improvement to specific changes, and you risk multiple changes interacting in unexpected ways.
Days 1-3: AMD Tuning — Moderate Impact
We adjusted the answering machine detection parameters for their elderly demographic. The key changes: longer initial silence tolerance (3200ms vs. 2500ms default) to give seniors more time to pick up, longer greeting tolerance (2000ms vs. 1500ms), wider pause tolerance between words (75ms vs. 50ms), and an extra word allowance (4 vs. 3) because some greetings are longer than three words.
We also added a post-detection verification step. Instead of immediately dropping AMD-flagged calls, the system listened for an additional 1.5 seconds for live speech patterns before disconnecting. This caught about 60% of false positives with minimal impact on agent efficiency.
Day 3 results:
| Metric | Before | Day 3 |
|---|---|---|
| Connect rate | 3.2% | 4.1% |
| AMD false positive rate | 9.3% | 3.2% |
| AMD accuracy | ~85% | 93.1% |
| Agent connects/day | 30 | 38 |
Good improvement — but not the big move by itself. AMD tuning added about 0.9 percentage points to the connect rate.
Days 4-7: DID Replacement — The Biggest Single Jump
This was where the numbers actually moved. We replaced all 20 burned DIDs with 60 fresh numbers (3x the pool size) and implemented intelligent rotation:
- Round-robin rotation across the full pool
- Cap of 30 calls per DID per hour, 150 per day (carrier spam algorithms flag numbers making more than ~200 calls/day)
- Automated daily health monitoring script flagging any DID with answer rate below 4%
- Numbers spread across multiple underlying carriers to avoid single-carrier flagging
Day 7 results:
| Metric | Before | Day 3 | Day 7 |
|---|---|---|---|
| Connect rate | 3.2% | 4.1% | 5.9% |
| DID average answer rate | 4.1% | 4.1% | 9.2% |
| Agent connects/day | 30 | 38 | 54 |
| Daily sales | 114 | 137 | 195 |
The DID change alone pushed connect rates from 4.1% to 5.9% — a 44% improvement. This single change was responsible for more improvement than AMD and predictive algorithm tuning combined. In our experience across 100+ centers, DID reputation is the number-one silent killer of connect rates. Most operations don't monitor it at all.
Days 8-14: Predictive Algorithm Optimization — The Final Push
With AMD and DIDs sorted, we switched from RATIO to ADAPT_TAPERED:
- Auto-adjusted dial levels starting from 1.0, with maximum of 8.0
- 3% maximum drop ceiling, 2% target
- Tapered ramp-up preventing the drop rate spike at shift start
- Adaptive intensity of 30 (conservative, can increase later)
We also bumped the hopper from 200 to 500 leads and configured intelligent lead recycling — different retry intervals for no-answers (2 hours), busy (1 hour), AMD-classified calls (4 hours), and drops (30 minutes).
Final Day 14 results:
| Metric | Before | Day 14 | Change |
|---|---|---|---|
| Connect rate | 3.2% | 7.1% | +122% |
| AMD accuracy | ~85% | 94.3% | +9.3 pts |
| Drop rate | 4.1% | 1.7% | -59% |
| Agent wait time (avg) | 18 sec | 7 sec | -61% |
| Agent connects/day | 30 | 65 | +117% |
| Daily sales | 114 | 235 | +106% |
| Agent utilization | 61% | 78% | +17 pts |
The ROI Math
Their incremental cost was $830/month (ViciStack's managed service at $6,750/month for 45 agents versus their previous $6,000/month in self-hosting and IT time).
The revenue delta: 121 additional daily sales at $320 per sale. Over 22 working days, that's $851,840 per month in additional revenue for $830 in additional cost. Even if you assume only half the improvement came from us, the ROI is over 500:1.
Revenue per agent per day went from $800 to $1,664. At $150/agent/month, ViciStack's cost represents 0.4% of what each agent generates.
What This Means for Your Center
The specific numbers here are from a Medicare insurance operation, but the three pillars — AMD tuning, DID management, predictive algorithm optimization — drive improvement in every vertical. The magnitudes vary:
| Vertical | Pre-Optimization Connect Rate | Post-Optimization Connect Rate |
|---|---|---|
| Insurance (Medicare, ACA) | 2.5-4.0% | 6.0-8.5% |
| Solar/Home Improvement | 3.0-4.5% | 6.5-9.0% |
| Debt Settlement/Financial | 2.0-3.5% | 5.5-7.5% |
| Real Estate | 3.5-5.0% | 7.0-10.0% |
| B2B Appointment Setting | 4.0-6.0% | 8.0-12.0% |
The Lesson Most Centers Miss
Everyone wants to jump straight to predictive algorithm tuning. It's the exciting part — changing dial levels, tweaking adaptive settings, adjusting hopper parameters. But in this case, it accounted for the smallest portion of improvement.
The unsexy work — checking DID reputation, replacing burned numbers, setting up rotation caps, tuning AMD parameters for your specific demographic — that's where the money is. DID replacement alone drove 44% of the total improvement.
If your connect rates are stuck below 5% and you haven't checked your DID answer rates recently, start there. Pull the per-DID answer rate query from your vicidial_log data, compare each number against your campaign average, and replace anything below 4%. You can do this yourself with VICIdial's CID Group Rotation feature and a fresh batch of local numbers from your SIP provider.
For the full day-by-day metrics timeline and every specific configuration change, see the complete case study at ViciStack.
Originally published at https://vicistack.com/blog/vicidial-roi-case-study/
Top comments (0)