Notes on resolution
Get The Resolution Report, our monthly email, in your inbox.
Introducing Pazau Agent 3
Procedures that test themselves against your past conversations before they go live, multi-step Actions with human approvals, and full parity on voice. Plus a precise definition of the only number we think matters: resolution rate.
Voice agents under 800ms: our latency budget, line by line
Humans expect a reply within about 700 milliseconds in conversation. Here's where every millisecond goes in a Pazau Voice turn, and the tricks that got our p95 voice-to-voice latency under 800ms.
Evaluating agents with simulation: how we test a Procedure before it talks to a customer
Starting today, Procedure changes in Pazau can be replayed against your real historical conversations before they are published. Here's how the simulator samples, replays, mocks Actions, and grades, and how we know the grades are right.
The real cost per contact (and why most teams undercount it by 40%)
Most support orgs divide agent salaries by ticket volume and call it cost per contact. The fully loaded number is much larger, and it changes how you should evaluate AI pricing.
Why we never train on customer data
It's a one-line promise on our homepage and a clause in our contracts. Here's the architecture behind it: zero data retention, redaction before inference, per-tenant isolation, regional residency, and what 'opt-in improvement' actually means.
One year after our Series C: what $85M bought
In March 2025 we raised an $85M Series C led by Northgate Partners. A year later, here's where the money went, what we got right, what we got wrong, and what comes next.
EU data residency is generally available
Every conversation, every model call, every Trace and every backup for EU workspaces now stays in the EU. Dublin is primary, Frankfurt is failover, and your contract is with Pazau Ireland under Irish law.
How Copperline Outfitters ran peak season with zero seasonal hires
Conversation volume hit 3.4× baseline between Black Friday and Christmas. Pazau Agent resolved 71% of it, and Copperline didn't hire a single temp. Here's what they built and what broke.
Procedures, not prompts
A 14,000-token system prompt is not an operating manual. Why Pazau Agent runs on structured, natural-language Procedures, and what that changes about reliability, auditability and who gets to edit the agent.
Resolution, not deflection
Deflection rate is the most popular metric in AI support, and it's mostly a measure of how well you hide the exit. Here's the number we bill on instead, and why we think the whole industry should switch.