NightDesk now knows which customers call back

When I started building NightDesk, the pitch was simple: answer the 11pm call so your engineer doesn't have to. That's still the core. But after a few months of building out the feature set, I've realized the more interesting story is what happens after the call.

Every call NightDesk handles is a data point. Who called. Why. Whether the AI resolved it or escalated. How long it took. Whether the same customer called three times this week about the same thing.

An answering service throws that away. NightDesk now keeps it — and surfaces it.

What's new

Caller history. NightDesk now remembers who called and what happened last time. When a repeat caller rings in, the triage agent knows their recent history before they say a word. A customer who escalated twice in the last week gets lower thresholds for escalation on the third call — because the pattern says "this one needs a human."

Escalation heatmap. GET /admin/tenants/{id}/escalation-heatmap returns a 24×7 grid of when escalations happen. Most MSPs have a gut sense that Mondays are rough or that 2am calls are usually printers. Now there's data. If your Wednesday 10pm slot is a hotspot, maybe you rotate your best engineer onto that shift.

Customer health scores. A 0-100 score per customer, composed of resolution rate (did we actually fix things?), call volume vs median (are they calling more than usual?), escalation rate, and recurring issue penalty. Grade A-F. The /customer-health endpoint returns customers sorted worst-first — exactly what you want to see at your Monday morning standup.

Shift performance analytics. Per-shift resolution rates and escalation rates, broken down by AM/PM/Evening/Overnight with configurable timezone offset. Now you can answer: is your overnight resolution rate actually worse, or does it just feel worse?

Week-over-week call trends. ISO-week bucketing with trend signal (up/down/flat comparing last 3 weeks to the prior 3). Useful for spotting when a customer is in trouble before they call to complain about your response times.

Recurring issue detector. Surfaces (customer, issue_category) pairs that exceed a call threshold in a rolling window. This is the "early warning" endpoint — if a customer has called about "VPN connectivity" four times in 30 days, that's a systemic issue worth a proactive conversation, not a reactive ticket.

Why this matters for the sales pitch

The first objection I always get is "what if it misses something?" That's the wrong frame. The right frame is: what are you missing right now?

Most MSPs have call records in their phone system, ticket counts in their PSA, and absolutely no way to connect the two. They can tell you their SLA hit rate. They can't tell you which customers have a recurring problem that's generating 60% of their after-hours calls.

NightDesk connects those dots automatically. The analytics don't require extra setup — they're a byproduct of every call the AI handles. By the time a pilot MSP is two months in, they have a dataset that's genuinely useful for quarterly business reviews, staffing decisions, and proactive customer conversations.

The stack

All of this runs on the same Lambda + DDB infrastructure as the triage agent. No separate analytics database. The /admin/tenants/{id}/... endpoints are all auth-gated on a dashboard token that the MSP owner controls. Every endpoint returns clean JSON, so if you want to pull this into a BI tool or your own dashboard, you can.

Current test count: 1,455. Still pre-launch, but the feature set is there.

NightDesk is an MSP after-hours voice triage product I'm building. If you run an MSP and want early access, reach out at chester@cwfrazier.com.