Post-call analytics and QA for voice AI
Know what your AI agent did on every call.
LensThe full-stack QA and quality management suite for voice AI agents.
Lens covers the whole quality job for your AI agents. It tests them before they go live, reviews and scores every call they take, and shows your team exactly what to fix. Quality, operations and compliance work from the same view of how your agents really perform.
01 / The QA gap
Your AI agent takes every call. Your QA team hears a few.
Contact-centre QA was built for human agents: sample a handful of calls a week, score them on a form, coach the person. AI agents run every call the same way, change whenever their instructions change, and fail without raising an error. A sampled review cannot keep up with them.
- Coverage
- A small sample, chosen by a reviewer with limited hours.
- Evidence
- A score on a form. The reason lives in the reviewer's head.
- Root cause
- "The bot was confusing." Nobody can say which instruction or system caused it.
- Fix
- A ticket to the vendor, then weeks of waiting to see if it worked.
- Coverage
- Every completed call is scored against your scorecard.
- Evidence
- Each finding points to the moment, quote or event that caused it.
- Root cause
- Lens separates agent mistakes, system problems and normal business outcomes.
- Fix
- A ready-to-use brief for your builders, then a simulated call to prove the fix.
02 / Call review
Replay any call, moment by moment.
Each call gets its own Lens page. Press play: the transcript, the agent's response times, interruptions and every step the agent took move together on one timeline. Share the page with a single link that only opens that call.
Try it
Example interface and call. Names, amounts and identifiers are illustrative.
Hear what the transcript misses
Slow replies, the caller talking over the agent and long silences rarely show up in text. Lens measures them from the call itself, so you can see a 3.7-second pause before the agent stated the amount owed.
Every outcome becomes a record
The agent records the result of each call as structured fields: promise to pay, amount, date, reason. They are delivered to your systems at hang-up and shown beside the call that produced them.
03 / Why the agent did it
Every decision in the call, in plain words.
Lens records each decision the agent makes: what it tried to do, what its rules allowed, and what happened next. Your team sees why the call went the way it did, without reading code.
Try it
Example call, the same one as in section 02. Builders can open the full technical record for any moment.
04 / QA auditor
Every call scored against your scorecard.
The Lens QA auditor scores each completed call on five measures and explains every deduction with the quote or event behind it and a suggested fix. Calls that need a person go to a review queue with their Lens link attached.
Try it
| Call | Outcome | Score | Verdict |
|---|
Example scores. Scorecard weights and checks are set per campaign.
05 / An opinionated scorecard
Start with rules learned on production calls.
You do not start from a blank form. Lens ships with checks drawn from running voice agents in production, and you add your own scripts, disclosures and compliance rules on top.
Every outcome is captured
The call's result and every fact the caller gave are recorded as data, not left in a summary.
No score without evidence
Each deduction cites a quote, a timestamp or an event. A score nobody can check is not reported.
Identity before account details
Balances and personal details are flagged if they are spoken before the caller is verified.
One clear goodbye
Repeated thank-yous and steps that end on an unanswered question are flagged as conversation quality issues.
Dropped calls are tracked
Calls that end early are recorded with what was collected up to that point, so nothing is lost.
Outcomes are not failures
A wrong number or a clear "not interested" is a business result. It is reported, but not held against the agent.
Who owns each finding
| Finding | What it means | Owner |
|---|---|---|
| Business outcome | Wrong number, not interested, already paid. The agent did its job; the list or the offer may need attention. | Operations |
| Agent quality | Slow replies, a skipped identity check, a missing field, an off-script statement. | Agent builder |
| System issue | A dropped line, a failed lookup, results not delivered to your CRM. | Platform and IT teams |
| Compliance risk | A required disclosure missed, account details shared before verification. | Compliance and QA |
06 / Campaign view
From single calls to the whole campaign.
Leaders see how far conversations get, where they stop and why, measured on the agent's real steps. Every number opens the calls behind it.
Try it
Example campaign data for illustration.
07 / Simulation lab
Test with difficult callers before real ones.
Before a new agent or a changed script goes live, run it against simulated callers: an angry customer, a caller who interrupts, a wrong number. Each simulated call gets its own Lens page and the same scorecard as a real one.
Try it
Example simulation. Simulated calls run faster than real time.
08 / The improvement loop
Find it, fix it, prove it.
Lens closes the loop between a bad call and a better agent. The debrief is written so a person or a coding assistant can act on it directly, and the fix is checked before it reaches customers.
Flag
A call scores below the bar, or a pattern repeats across calls.
Review
The reviewer opens the Lens link and confirms what happened.
Brief
One click copies a redacted debrief of the call's journey.
Fix
The builder, or their coding assistant, changes the agent's step.
Test
The same scenario is replayed with a simulated caller.
Confirm
Scores for the new version are tracked on live calls.
09 / Built for the contact centre
One record, read by every team that owns the call.
| Role | What Lens gives them | Starts in |
|---|---|---|
| Head of operations | Conversion through each step, reasons calls stop, and the minutes spent on bad numbers. | Campaign view |
| QA lead | A scored queue of calls that need a person, each with evidence and a link to the call. | QA auditor |
| Compliance | Checks for disclosures and identity verification on every call, with the moment each was met or missed. | Call review |
| Agent builder | Timelines, the step-by-step decision trail with exact JSON, the debrief and simulated callers to test changes. | Decision trail and simulation |
Redacted by default
The debrief leaves out phone numbers, contact identifiers and raw system traces.
Links scoped to one call
A Lens link opens only the call it was made for, and expires on a schedule you set.
Tracked by version
Every call records the agent version that handled it, so you can compare before and after a change.
People stay in charge
Reviewer notes sit on the call and are used to check the auditor's scores against your team's judgment.
Let's review
Bring us your hardest calls.
Share a set of real calls. We will score them, show you where they go wrong and what to change, on your own data.
dinodial.ai