Catching 340+ Silent Call Failures During Student Enrollment Peaks
A leading EdTech platform's student support calls were silently failing during enrollment peaks — with no dashboard surfacing it. Sherlock traced the root cause in hours.
15h/wk
Debug time saved
340+
Issues detected
99.2%
Uptime achieved
TL;DR — What Sherlock found
- 1
A leading EdTech platform's student support calls were silently failing during enrollment peaks — no dashboard surfaced it, and individual provider logs showed no errors.
- 2
Sherlock correlated Twilio routing events with ElevenLabs session logs, identifying a misconfigured routing rule invisible in either provider's UI.
- 3
The team fixed the routing in hours, not weeks. Uptime went from 87% to 99.2% and 15+ hours of weekly manual debugging disappeared.
The Problem
Students were complaining about dropped calls, but the ops team couldn't pinpoint the cause across Twilio and ElevenLabs logs. Manual debugging took days.
The Process
Sherlock correlated call events across multiple providers, identified a misconfigured routing rule, and surfaced a pattern of failures invisible in individual dashboards.
The Solution
With Sherlock's cross-service timeline, the team fixed the routing issue in hours instead of weeks — and set up alerts to catch similar problems instantly.
Results
- ✓15h+ saved per week on debugging
- ✓340+ silent failures detected automatically
- ✓Uptime improved from 87% to 99.2%
Use Cases
Questions the team asked Sherlock
- SC“Why are student calls dropping during enrollment?”
- SC“Which Twilio routing rules are causing failures?”
- SC“Show me all failed ElevenLabs sessions in the last 24 hours”
- SC“What percentage of calls failed this week vs last week?”
- SC“Find calls where the AI agent disconnected before completing”
Deep Dive
The full story
When student support calls began failing silently during enrollment peaks, the EdTech platform's operations team faced a familiar problem: two separate provider dashboards showing no obvious errors, but students reporting dropped connections and incomplete sessions. The gap between what the dashboards showed and what students experienced pointed to a cross-system issue — the kind that individual provider logs are structurally unable to surface.
Sherlock's investigation started with a cross-provider timeline: Twilio call routing events correlated with ElevenLabs session logs across the same time windows. Within minutes, a pattern emerged — a subset of calls was being routed through a configuration that triggered a graceful disconnect at the Twilio layer before ElevenLabs could establish its session. The routing rule had been misconfigured months earlier, but because it only manifested during high-concurrency enrollment periods, it had never appeared during routine testing.
Once the root cause was identified, the engineering fix took less than two hours. More importantly, the operations team gained something they hadn't had before: a joint timeline showing exactly how calls move through both Twilio and ElevenLabs, and the ability to ask Sherlock directly when anomalies appear. Uptime climbed from 87% to 99.2% — and the 15+ hours per week previously spent manually correlating logs disappeared entirely.
The broader lesson from this investigation: in multi-provider voice AI architectures, failures at the integration layer are common and nearly invisible in individual dashboards. The only reliable way to detect them is cross-provider correlation — which is exactly what Sherlock was built to do.
FAQ
Frequently asked questions
How does Sherlock detect silent call failures that don't appear in provider dashboards?
- Sherlock correlates events across multiple providers simultaneously — Twilio routing events, ElevenLabs session logs, and any other connected source. Failures that only appear at the intersection of two providers' event streams are invisible in individual dashboards but immediately visible in Sherlock's cross-provider timeline. You can query for anomalies in plain English from Slack and get a sourced answer in seconds.
What is cross-provider correlation and why does it matter for voice AI?
- Cross-provider correlation means matching events from different systems — for example, a Twilio routing event and an ElevenLabs session event — by call ID and timestamp, then analyzing them together. It matters because most voice AI failures happen at the boundary between systems, not inside any single system. Without correlation, those failures appear as unexplained gaps in each provider's logs.
How quickly can Sherlock investigate a call failure incident?
- Most investigations take minutes, not hours or days. Once connected to your providers, Sherlock can query the relevant time window, correlate events, and return a root cause hypothesis in under five minutes. Complex multi-provider incidents that previously took days of manual analysis are typically resolved in a single Slack conversation.
Does Sherlock work with telephony providers other than Twilio and ElevenLabs?
- Yes. Sherlock connects to a growing list of telephony and voice AI providers. If your provider is not yet integrated, Sherlock can often work with exported call data or logs. Check the integrations page for the current list of supported providers.
How do I set up alerts for call failures in Sherlock?
- Alerts are configured from the Sherlock Slack app. You specify the condition — for example, 'alert me if the failure rate exceeds 2% in any 30-minute window' — and Sherlock monitors your call data continuously, posting to the Slack channel of your choice when the threshold is crossed. No code or webhook configuration is required.
What does an improvement from 87% to 99.2% uptime actually mean in practice?
- At 87% uptime, roughly 1 in 8 calls was failing — meaning 1 in 8 students reaching out for support during enrollment had their call silently dropped. At 99.2%, fewer than 1 in 100 calls fails, and the remaining failures are detected and alerted on automatically. The practical difference is the gap between systematic student frustration and a reliable, trustworthy support channel.
Ready to stop guessing?
Let Sherlock investigate your voice calls. Find failures, cut costs, and get answers — all from Slack.