Product AI agent · Research analysis
User-Research Analysis AI Agent
Hold each transcript inside the consent it was taken under: purpose tagged at intake, sensitive lines redacted, retention clocks running, and the coded themes released by a named research lead.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Research lead
- Consent-scoped
- After launch
- Agent Care
What this agent does
Codes the themes, and holds the words
In
- 01
A consent is taken, and Art. 5(1)(b) binds the corpus to the purpose it names.
- 02
A transcript lands, and the study purpose, the recording state and the retention date ride with it.
Reason
- 03
A recording starts, and Florida and Pennsylvania turn on expectation, not on secrecy.
- 04
A line reveals a diagnosis, and C-184/20 puts it inside Art. 9 whether or not it was sought.
- 05
A corpus is queried, and the Art. 6(4) compatibility factors are put to the question first.
Decide
- 06
A withdrawal arrives, and EDPB Opinion 28/2024 denies the model a blanket claim to anonymity.
- 07
A study runs under a Federalwide Assurance, and 45 CFR 46 attaches to work it never funded.
Out
- 08
A theme is coded, and no rule reaches that step, so the controls sit around it instead.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
Tagging, redaction and coding belong to the agent. Release belongs to a named research lead, and a wrong finding is withdrawn, re-coded and logged against the study.
Example workflow
One study, consent to finding
- 1Session material receivedRecordings, transcripts, moderator notes, diary entries or intercept responses
- 2Consent and purpose boundThe purpose the participant agreed to, the recording state, the jurisdiction and the retention date
- 3Redaction and coding runThe redacted transcript, the codes applied, the quotes carried and confidence
- 4Controls appliedPurpose checks, special-category checks, retention checks and coding confidence
No human action required
Stages 1 to 4 run unaided, and nothing leaves the repository at any of them — the agent is coding, and the researcher lane opens at the purpose gate.
Goes to the named research lead.
Adds a privacy counsel read first.
Researcher review
The theme is held with its quotes, its consent scope and the redactions made.
Release · Append quote · Send to privacy review
Released — by the named research lead
- 6Repository and report records updatedOnly where write access and retention policy allow it
- 7Outcome evaluatedPurpose fit, redaction misses, researcher corrections and what review found
Corrections
Each researcher correction is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Releasing a finding into a product decision.
- Setting the purpose a study is collected for.
- Deciding a corpus may answer a new question.
- Signing a participant consent form.
Automation boundary
Agent acts unaided
- ✓Bind each transcript to the consent record behind it.
- ✓Run the coding pass, and carry each quote back to its transcript.
- ✓Redact special-category lines as the corpus is indexed.
- ✓Hold each theme for the named research lead, with its scope attached.
Nothing leaves the repository except by a named research lead, inside the agreed boundaries.
- Judging whether a theme is supported at all.
- Inferring emotion from voice or facial video.
- Extracting a voiceprint or face geometry.
- Changes to consent, retention or purpose rules.
Example output
One coded theme, annotated
This serves a research team who may have to show, long after a study closed, what a participant agreed to; below is one theme exactly as the agent leaves it.
Theme output · single study
Illustrative example- Study
- Onboarding, mid-market admins
- Coded theme
- Setup stalls at the invite step
- Quote carried
- Purpose tag, onboarding
- Consent of record
- Consent form, 3 August 2026
- Confidence
- Held unreleased
- Held for
- The named research lead, by name
As received
Taken from the recorded session and the signed consent, and it claims nothing beyond them.
What the record holds
- Signed consent
- Redacted transcript
- Retention date
Why no release here
Taking a theme into a product decision is a judgement a lead makes.
What the score decides
Below the configured threshold a theme gets a privacy read before the lead sees it.
Action
ReleaseAppend quoteSend 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 transcript
From the consent it was taken under
Where the corpus is used
A research agent collapses a citation chain to a source; this one holds the words of a named participant under the consent they were given under.
The corpus outlived its purpose
Ask a repository a question nobody was recruited for, and the purpose recorded at collection stops covering it, with nobody having lied.
What was checked, and not found
No rule was found requiring a company to run user research, to run it competently or to report what it finds honestly, and Illinois has had no flat all-party recording prohibition since Public Act 98-1142, so what the agent carries is the consent, not a standard of practice.
Build an evidence trail
The transcript, the consent it was taken under and the researcher who coded it stay together.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Research repositories
- Dovetail
- Condens
- EnjoyHQ
- Coded studies, quotes and tags
Session recording
- Zoom
- Teams
- Lookback
- Session recordings and transcripts
Participant panels
- Respondent
- User Interviews
- Screeners, consents and incentive records
Agent
User-research analysis
- Reads the consent
- Codes the transcript
- Holds for the lead
Product and issue systems
- Jira
- Productboard
- Notion
- Findings, themes and the issues raised
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six filters between the corpus and the finding
Six filters down one line, the last the closest. What is retained 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
- Hands the release to a named research lead
- L4 – L5
- Keep the transcript and the consent behind it
- L6
- Withholds the theme when signals degrade
How Nestack evaluates it
Evaluate the whole analysis — not only the theme that comes out.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the finding a decision carries
- E1Final-output evaluationDid the theme record the consent scope it was coded inside?
- E2Step-level evaluationDid the agent read the right study, the right purpose and the live redaction rules?
- E3Tool evaluationDid it read and write the correct study and the correct theme?
- E4Confidence calibrationDo low-confidence themes actually attract more researcher corrections?
- E5Slice evaluationHow does performance change across specific study types?
- E6Business outcomeHow many themes needed a correction before the lead released them?
Floor — the corpus a finding rests on
Failure modes
Where each failure originates in the agent
Seven failure modes, each set where the stage first exposes it.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- QF-03
Superseded consent read
The consent read is not the one the participant signed.
Stage gathers
The consents, the sessions, the purposes and the codes
02 · Reasoning
2 modes
- QF-04
Theme asserted, not evidenced
A theme appears with no quote beneath it.
- QF-06
Closed purpose read as live
A shut study is worked as an open one.
Stage proposes
The consent scope, the redactions and the codes
03 · Tool / write
2 modes
- QF-02
Thin theme passed forward
A theme moves on without the privacy read.
- QF-05
Quote bound to wrong participant
The words are filed against another person.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- QF-01
Released, consent unrecorded
The record shows release but not the consent under it.
Stage returns
The finding a team acts on and a board reads
05 · Change / Version
1 mode
- QF-07
Silent purpose drift
A corpus widens while the stored purpose keeps the old one.
Stage tracks
Model, prompt, purpose rules and consent dates
- Sev-1 · a corpus read outside its purpose
- Sev-2 · a raw quote reaches a report
- Sev-3 · consent degrades, theme held back
Affected slices
Re-used corpora absorb the corrections
A study-level purpose figure can read clean while a small number of re-used legacy corpora carry most of the corrections. Nestack reports the correction rate by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Re-used legacy corpora | 11.4% | 3.7× | Review | |
| Employee-participant studies | 8.1% | 2.6× | Review | |
| Recorded video sessions | 5.1% | 1.7× | Watch | |
| Single-study text corpora | 2.5% | 0.8× | Normal |
Bar: correction-rate lift vs. single-study baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
What a widened corpus costs
A loop closes when the study reused outside its stated purpose is a standing case. That suite is what the next corpus indexed is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Correction rate rises on re-used legacy corpora.
- 02
Diagnose
The interviews taken to fix onboarding, now answering a question about price, are read back against the consent until one cause is left standing.
- 03
Improve
Number the change; the transcripts that drove it are filed beneath it.
- 04
Verify
A single red consent case stops the whole release.
- 05
Learn
One case joins the suite, one line joins the intake rules.
Learn → Detect
The return edge. The next corpus 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, transcript analysis, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Consent-scope mapping and automation-boundary work.Week 1
- 02Repository, recorder and panel sources.Week 1, Week 2
- 03Purpose-to-transcript and consent-coverage mapping.Week 1, Week 2
- 04Transcript ingestion.Week 2
- 05Consent, purpose and study binding.Week 2, Week 3
- 06Redaction scoring and review routing.Week 3
- 07Researcher release workflow.Week 3, Week 4
- 08Repository and tracker integration.Week 5
- 09Consent and purpose cases.Week 4, Week 5
- 10Guardrails and purge controls.Week 4
- 11Transcript-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 study, one round | ProductionProduction research workflow | AdvancedMultiple studies / regions |
|---|---|---|---|
| Introduced at Pilot | |||
| Coding to your study purposes | ✓ | ✓ | ✓ |
| Named researcher release | ✓ | ✓ | ✓ |
| Participant-consent baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by study purpose | — | ✓ | ✓ |
| Researcher review workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Repository-and-recorder integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-jurisdiction recording rules | — | — | ✓ |
| Cross-study theme packs | — | — | ✓ |
| Large transcript archives | — | — | ✓ |
| Multi-study retention controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, transcript 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 studies and the purpose each was collected for
→Consent capture and purpose versioning
Week 102Representative transcripts, recordings and notes
→Source binding, redaction rules and the coding baseline
Week 203Your retention schedule and the leads it names
→Consent mapping, purpose binding and the automation boundary
Week 104Access to relevant APIs, feeds or exports
→Repository, recorder and panel assessment, then integration setup
Week 205Transcripts you would not want subpoenaed
→Purpose cases and failure-mode testing
Week 406What no study may re-use
→Redaction scoring, review routing, guardrails and release controls
Week 307A named research lead who releases the finding
→Release to the named research 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
Two phases genuinely share week five here, and not one band was widened to make the column look tidy.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Consent discovery, purpose versioning and the automation boundary
W2Source integration and the participant-consent baseline
W3Coding workflow, redaction logic and release controls
W4Evaluation suite, purpose cases and failure-mode testing
W5Repository integration, pilot studies and targeted corrections
W6One study round run under the research lead, then Agent Care handover
- Reading the band
- A band covers the weeks its own work is named in, and week five carries two by design.
- At the end of W6
- When 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 · Product AI agent
Build a user-research agent around the consent your last study was taken under.
Show us one study you would re-open today and the form its participants signed. That form covered one purpose, for one study, for a stated term. If you build a regulated device, the record you must keep is the one the GDPR pushes you to delete.
- Typical delivery
- 4–6 weeks
- Deployment
- Your live studies
- Research lead
- Consent-scoped
- After launch
- Agent Care available
Nestack Agents · User-research analysis
AGT-PRD-03 · Agent Care available after launch