Customer Success AI agent · Voice of customer
Voice-of-Customer Synthesis AI Agent
Establish what a recording was permitted to be used for before reading it, let each theme rest only on the words that survive that test, and hand the quote to the insight lead.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Insight lead
- Consent first
- After launch
- Agent Care
What this agent does
Draws the theme, never publishes the quote
In
- 01
A call is read, and whether it could lawfully be recorded was settled before the agent reached it.
- 02
A recording leans on 18 U.S.C. 2511(2)(d), which bars liability rather than granting a right.
Reason
- 03
A call comes from an all-party state, while Massachusetts asks only that nobody recorded in secret.
- 04
A tone score off the agent, not the caller, is what Article 5(1)(f) has prohibited since February 2025.
- 05
A caller sentiment read is not prohibited; Annex III point 1(c) makes it high-risk from December 2027.
Decide
- 06
A transcript tells speaker A from speaker B, and joining that label to a name is what is under challenge.
- 07
A recording kept for quality assurance is read for insight, and the new purpose is tested, not assumed.
Out
- 08
A theme is drawn, and it names the calls and the words under it rather than standing on its own.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
Reading, clustering and attribution belong to the agent. Publication belongs to a named insight lead, who puts the quote in front of the business and answers for it.
Example workflow
One theme, recordings to publication
- 1Feedback receivedSupport calls, tickets, survey verbatims, community threads or in-app messages
- 2Consent basis establishedThe jurisdiction the call came from, the disclosure made, the purpose it was recorded for and the date
- 3Themes drawn from the wordsThe theme, the quotes under it, the sources they came from and confidence
- 4Controls appliedConsent checks, purpose checks, attribution checks and drafting confidence
No human action required
Stages 1 to 4 run unaided, and nothing is published at any of them — the agent is reading, and the insight lead lane opens at the attribution gate.
Goes to the named insight lead.
Adds a privacy lead read first.
Insight lead review
The theme is held with its quotes, their consent basis and the sources they came from.
Publish · Add source · Send to privacy review
Published — by the named insight lead
- 6Insight and reporting records updatedOnly where write access and records policy allow it
- 7Outcome evaluatedAttribution accuracy, consent coverage, lead corrections and what review found
Corrections
Each insight lead correction is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Publishing a customer quote outside the company.
- Deciding that a recording was lawfully made.
- Approving a recording for a purpose it was not made for.
- Signing off a theme for a board pack.
Automation boundary
Agent acts unaided
- ✓Read calls, tickets, surveys and community threads.
- ✓Carry the consent basis forward beside each quote the theme holds.
- ✓Draw the theme and name the words that produced it.
- ✓Flag a quote whose recording or purpose will not support its use.
Nothing is published except by a named insight lead, inside the agreed boundaries.
- Judging whether a customer can be named at all.
- Enabling a score of an agent tone, stress or empathy.
- Choosing what the company does about a theme.
- Changes to the consent, purpose or attribution rules.
Example output
One theme, annotated
This serves an insight team who may have to say where a customer sentence in a board pack came from; below is one theme exactly as the agent leaves it.
Theme record · single theme
Illustrative example- Theme
- Onboarding friction, mid-market accounts
- Recorded as
- Drawn across calls and tickets
- Source mix
- Consent basis recorded
- Evidence of record
- Call transcript, 6 May 2026
- Confidence
- Held unpublished
- Held for
- The named insight lead
As received
Drawn from the transcripts and tickets on file, and it claims nothing the words do not say.
What the record holds
- Call transcript
- Support ticket
- Survey verbatim
Why no publication here
Publishing a customer sentence is a judgement the insight lead makes.
What the score decides
Below the configured threshold the theme gets a privacy read before the lead sees it.
Action
PublishAdd sourceSend to privacy reviewValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every theme
From the words that produced it
Where the words came from
Attention measurement and the marketing analytics agents measure campaigns; this reads what customers actually said, and its first question is whether it was allowed to.
The consent came first
Move the state a call came from, or the purpose it was recorded for, and the same sentence stops being usable, with nobody having lied.
What was checked, and not found
Checked across the statutes surveyed: nobody at the company signs anything. The BIPA written release is executed by the subject, the retention policy is made public rather than filed, and Illinois has no regulator to receive it.
Build an evidence trail
The quote, the recording it came from and the consent behind it stay together.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Contact centre platforms
- Genesys Cloud
- NICE CXone
- Five9
- Talkdesk
- Amazon Connect
Support desks and tickets
- Zendesk
- Freshdesk
- Intercom
- Salesforce Service Cloud
Survey and feedback tools
- Qualtrics
- Medallia
- SurveyMonkey
- Delighted
- Typeform open text
Agent
Voice-of-customer synthesis
- Reads the feedback
- Draws the themes
- Holds for the lead
Community and review boards
- Discourse
- Khoros
- Higher Logic
- App store and review-site posts
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six strainers between the model and the quote
Six strainers set in a line, the last the finest. What still passes is drawn in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a theme may stand
- L3
- Leaves the publication to a named lead
- L4 – L5
- Keep the quote and the recording behind it
- L6
- Quotes nothing and reports the gap when signals degrade
How Nestack evaluates it
Evaluate the whole reading — not only the theme that comes out.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the quote a board pack carries
- E1Final-output evaluationDid the theme record the words and the calls it was actually drawn from?
- E2Step-level evaluationDid the agent read the right sources, the right period and the live consent record?
- E3Tool evaluationDid it read and write the correct theme and the correct quote?
- E4Confidence calibrationDo low-confidence themes actually attract more lead corrections?
- E5Slice evaluationHow does performance change across specific feedback channels?
- E6Business outcomeHow many themes needed a correction before the lead published?
Floor — the words a theme rests on
Failure modes
Where each failure originates in the agent
Seven failure modes, each placed where the stage first uncovers it.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- OE-03
Consent record missing
The call is read with no basis recorded for it.
Stage gathers
The calls, the tickets, the surveys and the posts
02 · Reasoning
2 modes
- OE-04
Theme asserted, not sourced
A theme appears with no words under it.
- OE-06
Barred purpose read as allowed
A recording is mined beyond its purpose.
Stage proposes
The theme, the quotes and the consent basis
03 · Tool / write
2 modes
- OE-02
Thin theme passed forward
A theme moves on without the privacy read.
- OE-05
Quote bound to wrong customer
The words are filed against another account.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- OE-01
Published, source unrecorded
The record shows a quote but not the call under it.
Stage returns
The theme a pack carries and a board reads
05 · Change / Version
1 mode
- OE-07
Silent consent drift
A recording policy changes while stored quotes keep the old one.
Stage tracks
Model, prompt, consent rules and source dates
- Sev-1 · a quote published without consent
- Sev-2 · a wrong quote reaches the board
- Sev-3 · source degrades, theme held back
Affected slices
All-party audio absorbs the corrections
A theme-level attribution figure can read clean while calls from all-party states carry most of the corrections. Nestack reports the correction rate by theme, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| All-party state call audio | 12.2% | 3.7× | Review | |
| Multi-channel merged themes | 8.7% | 2.6× | Review | |
| Community and review threads | 5.4% | 1.6× | Watch | |
| Single-channel survey themes | 2.2% | 0.7× | Normal |
Bar: correction-rate lift vs. single-channel baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
What an unconsented quote costs
A cycle ends when the quote pulled from an unconsented call is a regression case. That suite is what the next theme issued is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Correction rate rises on calls from all-party states.
- 02
Diagnose
The customer sentence in the board pack, pulled from a call recorded in a state that wanted both sides to agree, is worked backwards until one cause is left standing.
- 03
Improve
Number the change; the quotes that forced it are attached to it.
- 04
Verify
One quote case still red is enough to hold the release back.
- 05
Learn
It stays a standing test, and the consent rules are amended alongside it.
Learn → Detect
The return edge. The next theme is measured against a suite one case longer.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, theme synthesis, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Feedback-corpus and automation-boundary scoping.Week 1
- 02Call, ticket, survey and community sources.Week 1, Week 2
- 03Consent-basis and attribution-coverage mapping.Week 1, Week 2
- 04Call, ticket and survey ingestion.Week 2
- 05Quote, source and consent binding.Week 2, Week 3
- 06Attribution scoring and review routing.Week 3
- 07Insight lead publication workflow.Week 3, Week 4
- 08Feedback-platform integration.Week 5
- 09Consent and attribution cases.Week 4, Week 5
- 10Guardrails and attribution controls.Week 4
- 11Quote-trail instrumentation.Week 5, Week 6
- 12Deployment, documentation and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne channel, one quarter | ProductionProduction insight workflow | AdvancedMultiple channels / regions |
|---|---|---|---|
| Introduced at Pilot | |||
| Theme synthesis to your corpus | ✓ | ✓ | ✓ |
| Named insight lead publication | ✓ | ✓ | ✓ |
| Feedback-corpus baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by theme | — | ✓ | ✓ |
| Insight review workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Call-recording integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-source theme merging | — | — | ✓ |
| Cross-channel insight packs | — | — | ✓ |
| Large feedback corpora | — | — | ✓ |
| Multi-state consent controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, feedback volume, approval controls and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01Your live feedback channels and what each was collected under
→Consent-basis capture and channel binding
Week 102Representative calls, tickets, surveys and threads
→Source binding, clustering logic and the theme baseline
Week 203Your reporting calendar and the leads it names
→Consent mapping, source binding and the automation boundary
Week 104Access to relevant APIs, feeds or exports
→Contact-centre, desk and survey source assessment, then integration setup
Week 205Quotes you would not want sourced
→Consent cases and failure-mode testing
Week 406What no quote may represent
→Attribution scoring, review routing, guardrails and release controls
Week 307A named insight lead who publishes the theme
→Release to the named insight lead, then pilot and production validation
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
Six weeks of measured work rather than six tidy columns, which is why the fifth carries a pair.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Feedback discovery, consent mapping and the automation boundary
W2Source integration and the feedback-corpus baseline
W3Theme synthesis, attribution logic and release controls
W4Evaluation suite, consent cases and failure-mode testing
W5Feedback-platform integration, pilot themes and targeted corrections
W6One feedback quarter run under the insight lead, then Agent Care handover
- Reading the band
- Each bar covers only the weeks its own work is named for, and week five is shared by design.
- At the end of W6
- Once the consent record validates, Agent Care picks the agent up.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Customer Success AI agent
Build a voice-of-customer agent around the consent your last quote never wrote down.
Show us one theme you reported last quarter and the calls its quotes came from. The recording came first, and whether it was lawful was settled before anybody read a word of it. What no quote can carry, we name before it reaches a board pack.
- Typical delivery
- 4–6 weeks
- Deployment
- Your feedback channels
- Insight lead
- Consent first
- After launch
- Agent Care available
Nestack Agents · Voice-of-customer synthesis
AGT-CX-12 · Agent Care available after launch