Case studies
What people actually call me about
Most engagements start the same way — a platform that grew faster than its architecture, and a floor of agents that cannot stop dialing while it gets fixed.
Building a multi-tenant Press-1 platform that holds at 5,000 concurrent attempts
Press-1 is unforgiving infrastructure: nearly all the work happens on calls that never reach a human. This one carries 5,000+ concurrent attempts, detects machines with its own AMD, and bills to six decimal places per tenant.
- Asterisk
- In-house AMD
- Multi-tenant billing
- SIP
- Carrier routing
- 5,000+
- Concurrent attempts sustained
- 6 dp
- Billing precision
- Removed
- AMD licensing cost
- Full
- Tenant isolation
Moving a 1,200-agent floor off self-hosted VICIdial without dropping a call
Cutover of the largest single floor I have run — 1,200 concurrent agents — from ageing on-premise hardware to a managed cluster, executed live between shifts.
- VICIdial
- Asterisk
- MariaDB cluster
- Multi-carrier SIP
- HAProxy
- 0 min
- Downtime during cutover
- 1,200
- Concurrent agents sustained
- −38%
- Answer-to-agent latency
- 0
- Unplanned outages since
Putting a SIP layer in front of the dialers so maintenance stops costing calls
Every patch meant a maintenance window, and every maintenance window meant lost dialing. Moving signalling off the dialers made nodes disposable and upgrades boring.
- OpenSIPS
- Kamailio
- VICIdial
- Asterisk
- Dispatcher failover
- 0
- Maintenance windows needed
- 0
- Calls dropped per node drain
- Tested
- Carrier failover
- 2 wks → 1 day
- Time to patch the cluster
Bringing an abandon rate back under the line without losing contact volume
A predictive dialer tuned for raw throughput had pushed the abandon rate into territory the client’s compliance counsel would not sign off on. Fixed the pacing without surrendering connects.
- VICIdial
- Custom AMD tuning
- DNC scrubbing
- Scheduled reporting
- 2.6% → 0.9%
- Abandon rate
- +11%
- Connected calls
- −72%
- AMD false positives
- 0
- Compliance exceptions
When the CRM and the dialer disagreed about the same lead
Two systems, two versions of the truth, and a sales team that had learned to trust neither. The fix was reconciliation, not another integration.
- vTiger
- SugarCRM
- VICIdial
- Event-driven sync
- −94%
- Records disagreeing
- 100%
- Call logging coverage
- −45%
- Rep time on data entry
- Retired
- Shadow spreadsheets
Taking a US-only outbound team into eleven countries
Local presence, carrier routing and compliance rules for an outbound team expanding beyond North America — without standing up a call centre in each market.
- Adoptiv
- Multi-carrier routing
- Local presence numbers
- Regional compliance rules
- 11
- Markets live
- +2.4×
- Answer rate, non-US
- −29%
- Cost per connect
- 0
- Regional platforms needed
Replacing 2% manual call sampling with 100% AI auditing
A QA team listening to a fraction of a percent of calls now reviews all of them — with humans spending their time on the calls that actually warrant it.
- VICIfast
- AI transcription
- AI quality auditing
- Sentiment analysis
- 2% → 100%
- Call coverage
- 3 wks → 1 day
- Time to coaching
- −80%
- QA hours on routine scoring
- +22pts
- Compliance phrase adherence