Watchtower checks every AI-handled call against your own knowledge and rules, verifies every fail, and rolls the results into a single health score. No transcripts to read.
Traditional QA: listen to 5% of calls, hope the rest are fine. Problems surface as complaints.
No visibility into whether your agent is providing correct information or handling escalations properly.
Without metrics, you can't improve. Without trends, you can't spot problems early.
Every feature below is included on every plan. No add-ons, no feature gates.
Every call runs through pass/fail checks on resolution, accuracy, professionalism and escalation handling, plus expectations you add in plain language. A fail has to quote the transcript.
One number tells you how your agent is performing: the share of checks passed over the last 30 days, weighted toward this week.
Monday morning email with your score, trends, flagged calls, and areas for improvement.
Score drops below 70? You hear about it immediately, not in next week's report.
Watchtower starts scoring from your first call. No configuration needed — it's built into the platform and included on every plan.
Your agent answers in a voice and name you choose, included on every plan. Hear the voices →
80% of businesses use voice agents but only 21% are satisfied with the results. Watchtower shows you exactly where your agent excels and where it needs improvement.Source: Deepgram / Opus Research, 2025 State of Voice AI Report
Every AI-handled call is automatically scored after it ends
Scores roll up into your Agent Health Score (0-100)
Calls that fail a check are flagged for review with the quoted line and the reason
Weekly digest email keeps you informed without checking the dashboard
Alerts notify you immediately if quality drops
After each call, a scorer is given the transcript together with the knowledge, escalation rules, and lookup results your agent actually had on that call, and runs a fixed list of pass/fail checks in four categories: resolution, accuracy, professionalism, and escalation handling, plus any expectations you add in plain language. Pass is the default; a check can only fail if the scorer quotes the transcript line that shows it, and a second, independent pass verifies every fail before it counts. A call's score is the share of applicable checks passed, so a call handled correctly scores 100, including one where the caller wanted something you don't offer and the agent said so. Caller mood is reported but is not part of the score. Your org health score pools the last 30 days, weighting the last seven days fully and older calls progressively less.
A call is flagged when it fails a core check you would want to look at: an unsupported or contradictory fact, a gap the agent was not honest about, a need left unaddressed, a missed capture, a missed or unnecessary escalation, or behavior a caller would complain about. Smaller misses such as a rough close or a thin handoff explanation cost score but do not flag. Fails on your own expectations count in the score but never flag a call on their own. Every fail has to quote the transcript and survive a second-pass verification before it can flag anything, so a flag is evidence, not a hunch. Caller mood never flags a call: an unhappy caller is not by itself an agent failure. Flagged calls appear on the dashboard with the failed checks inline, and the weekly digest groups them by which check failed.
No. Calls under ten seconds, calls where the caller never spoke, near-silent dialer attempts, and calls the scorer identifies as robocalls or wrong-number hangups are filtered out before scoring. The filter is deliberately conservative, so a borderline call is included rather than dropped. If it gets one wrong in either direction, you can override the inclusion on the call detail page, and the override is reversible.
Not today, and it is worth being straight about the limits. The 90 / 75 / 60 bands are fixed platform-wide and cannot be customized per account. Scores are computed at the organization level, so there is no breakdown by which team member would have taken the call. Scoring reads the transcript, the tool log and the knowledge the agent had; it does not listen to the audio, so it does not judge sound quality. The digest email runs Monday with no other cadence. What you can change is what gets checked: add your own expectations in plain language, and switch off any core check that does not fit how you work. Watchtower is included on every plan at no extra charge.
White-glove setup
Custom API integrations, webhooks, CRM sync, escalation design — designed, wired, and tested on live calls. Included with any plan. No setup fees.
Call Allison, describe your business, and she'll build your agent. Or sign up and configure it yourself.