Open the conversation on a disclosure, keep the exchange in the state it opened in, and record the turn where a named person took over, for the legal function that owns the surface.
A conversation opens, and Art. 50(1) has been enforceable since 2 August 2026, undeferred.
02
A surface is deployed, and Art. 50(1) falls on the provider of the system, not on its deployer.
Reason
03
A vendor bot is renamed for the brand, and Art. 3(3) can turn that deployer into a provider.
04
A first turn passes, and Art. 50(1) wants the person told at the start of that interaction.
05
A notice is thought unnecessary, and the obvious exemption in Art. 50(1) is judged objectively.
Decide
06
A person takes the conversation over, and what the customer heard at the opening stops being true.
07
A topic sits outside brand policy, and the agent refuses it rather than answering around the edge.
Out
08
A transcript is kept, and nothing in Chapter IV obliges a deployer to keep one.
09
Execute write actions only inside the approval boundaries agreed during implementation.
→Product statement
The agent discloses on the first turn and holds the exchange to that opening. A named person takes over, and legal decides whether the company is a provider.
Example workflow
One conversation, first turn to handover
AgentHuman
1Conversation evidence receivedChat transcripts, surface configurations, handover events or vendor system records
2Conversation context assembledThe surface, the system behind it, the paragraph it answers to and the day the exchange opened
3Disclosure evidence draftedThe conversation, its opening notice, the handover events and coverage
4Controls appliedNotice-presence checks, timing checks, handover checks and completeness confidence
No human action required
Stages 1 to 4 run unaided, and nothing is settled at any of them — the agent is recording, and the counsel lane opens at the completeness gate.
5DecisionSplits at the completeness gate
Evidence sufficient
Goes to the named brand owner to approve.
Anything thin
Takes a brand counsel read first.
Counsel review
The conversation is held with its opening notice, its handover and the surface it ran on.
Approve · Append evidence · Send to counsel
Approved — by a named brand owner▼
6Surface and conversation records updatedOnly where write access and retention policy allow it
7Outcome evaluatedNotice timing, handover capture, reviewer corrections and what counsel found
Corrections
Each counsel correction is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required8 items
Deciding which role Art. 3(3) puts the brand in.
Approving the wording a surface opens with.
Judging a disclosure obvious under Art. 50(1).
Deciding that a reserved topic may be answered.
Automation boundaryAgent acts unaided
✓State that the customer is dealing with a machine on the first turn.
✓Carry the opening disclosure through every later turn.
✓Record the turn at which a named person assumed.
✓Refuse the topics the brand policy reserves for a person to answer.
The role and the surface are settled by a named person, inside the agreed boundaries.
Telling a customer that no disclosure was owed.
Answering a national market surveillance authority.
Setting the policy a live surface is held to.
Changes to surfaces, scripts or handover rules.
Example output
One conversation, annotated
Our sibling content agent answers for Art. 50(2) marking on generated assets; this page picks up Art. 50(1) at the moment a person starts talking to the brand.
Conversation record · single exchangeIllustrative example
Conversation
Recorded as
Surface
Evidence of record
Confidence
Held for
Branded chat surface, Art. 50(1)
Disclosed on the first turn, handover logged
Brand-named surface
Handover event, 3 August 2026
Held unsettled
The brand legal owner, by name
As receivedTaken from the transcript and the surface configuration — it reaches as far as those sources do.
What the record holdsOpening noticeSurface configurationHandover event
Why no role call hereWhether the brand is the provider is an Art. 3(3) call, not a model output.
ActionApproveAppend evidenceSend to counsel
What the score decidesBelow the configured threshold the conversation picks up a counsel read before approval.
Value
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value landsValue 01 – 04
Every conversationFrom the surface it opened on
03Evidence
Where the evidence is used
The shipped advertising pages are agency-side, serving a client who booked the work; this one answers to the brand legal function that decides whether the company is a provider.
01Approved path
Whose system is it
White-labelling a vendor bot as Ask Brand puts the system into service under the brand name, and Art. 3(3) then reads the brand as its provider.
02Human review
What was checked, and not found
Checked and not confirmed: the entry-into-force and Official Journal dates of the AI Act, which we did not verify in primary text; any harmonised standard or implementing act fixing a transparency format, none of which is in force; any published count of signatories to the Code of Practice on Transparency of AI-generated Content; and the operative provision behind the four-month transitional, which we could not locate and which binds providers rather than deployers.
04Build an evidence trail
The conversation, the disclosure it opened with and the person who took over stay together.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Conversational surfacesWeb chat · in-app assistant Surface configuration records
Vendor and platform systemsBot platform · contracts System and naming records
Contact centre and handoverAgent desktop · queue events Handover and transfer records
Agent
Conversational brand disclosure
Reads the conversations Tests the disclosure Holds for the brand owner
Case and retention systemsCounsel tooling · ticketing Disclosure and escalation records
Integration availability depends on the client's existing systems and API access.
Agent controls
Six thresholds between the model and the brand owner
Six thresholds crossed in turn, the last the narrowest. Whoever gets through is drawn in the map below.
L6 · Outermost — last line of defenceInward → L1 · closest to the model
L6Rollback / safe modeNarrow the agent to disclosure recording when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt and disclosure rules; the Digital Omnibus, Regulation (EU) 2026/1744, deferred Chapter III high-risk duties only.Track
L4TraceabilityRecord each conversation, the notice it opened with, the handover it carried and every read of it.Record
L3Brand owner releaseHold the conversation for a named brand owner; the hold governs approval, not whether Art. 50(1) landed on the brand at all.Gate
L2Disclosure guardrailsTest each surface against the disclosure policy you configure; the Code of Practice on Transparency, final 31 July 2026, is voluntary, and not signing it does not constitute non-compliance.Restrict
L1Confidence thresholdsRoute a thin conversation to a counsel read first; Art. 50(1) binds providers, and a brand-named surface can make the brand one.Require review
Model coreEvidence assembled — the conversation, its notice, the handover and coverage
L1 – L2Test whether a record may stand
L3Leaves the approval with a named person
L4 – L5Keep the conversation and the disclosure behind it
L6Hands to a person when signals degrade
How Nestack evaluates it
Evaluate the whole assembly — not only the conversation record that comes out.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the transcript a regulator reads
Depth of coverage ▼
E1Final-output evaluationDid the record show what the customer was actually told?
E2Step-level evaluationDid the agent read the right surface, the right session and the live configuration?
E3Tool evaluationDid it read and write the correct conversation and the correct handover?
E4Confidence calibrationDo low-confidence conversations actually attract more counsel corrections?
E5Slice evaluationHow does performance change across specific surfaces?
E6Business outcomeHow many conversations needed a correction before approval?
Floor — the account the brand must give
Failure modes
Where each failure originates in the agent
Seven failure modes, each placed at the stage where it first shows.
Agent lifecycleDirection of processing →
01 · Retrieval1 mode
LL-03
Stale surface read
The configuration read is not the one now serving customers.
Stage gathersThe surface, the session, the notice and the handover
02 · Reasoning2 modes
LL-04
Disclosure asserted, untested
A notice is recorded shown without a check.
LL-06
Provider role read as settled
Art. 3(3) is treated as already answered.
Stage proposesThe conversations, their notices and coverage
03 · Tool / write2 modes
LL-02
Thin record passed forward
A record moves on without the counsel read.
LL-05
Bound to the wrong session
A handover is filed against the wrong conversation.
Stage writesOnly where write access and approval policy allow it
04 · Output1 mode
LL-01
Approved, disclosure unrecorded
The record shows approval but not what the customer was told.
Stage returnsThe record a brand owner approves and counsel reads
05 · Change / Version1 mode
LL-07
Silent disclosure regression
A surface change drops the notice, not the record.
Stage tracksModel, prompt, disclosure rules and record fields
Sev-1 · a conversation opened undisclosedSev-2 · a handover reaches no recordSev-3 · surface degrades, agent hands over
A surface-level disclosure figure can read clean while the brand-named chat carries most of the rework. Nestack reports the correction rate by surface, not only in total.
Slice performance — reported separately, not only in aggregateIllustrative example
Slice
Failure rate
Lift
Lift vs. threshold
Status
Brand-named web chat
8.6%
3.6×
Review
In-app assistant
6.1%
2.6×
Review
Messaging channels
3.8%
1.6×
Watch
Vendor-hosted help widget
2.0%
0.8×
Normal
Bar: correction-rate lift vs. vendor-widget baseline · scale 0–4.0× · tick marks the 2.0× review threshold2 of 4 slices over threshold
Evidence-linked improvement
What an undisclosed handover costs
A cycle closes when the undisclosed handover is a regression case. That suite is what the next conversation logged is measured against.
Improvement cycle · five stagesSwitchback — the path turns at Improve and returns at Learn
01Detect
Correction rate rises on the brand-named chat.
02Diagnose
The moment a person took the conversation over and nobody told the customer is read back until one cause remains.
03Improve
The change ships numbered, and the conversations that forced it ride with it.
04Verify
Nothing releases while one touched conversation case is still red.
05Learn
It is retained for good, and the disclosure rules are amended in that same commit.
Learn → DetectThe return edge. The next exchange 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, disclosure assembly, evaluation, integration, then production validation and handover.
WorkstreamWeek 1Week 2Week 3Week 4Week 5Week 6
01Disclosure-state and automation-boundary work.
02Surface, vendor and handover sources.
03Surface-to-paragraph and role-determination mapping.
04Conversation and transcript ingestion.
05Surface, notice and record binding.
06Completeness scoring and review routing.
07Brand owner approval workflow.
08Contact-centre and CRM integration.
09Disclosure and handover cases.
10Guardrails and escalation controls.
11Conversation-trail instrumentation.
12Deployment, documentation and Agent Care handover.
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallelFinal 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 tierPilotOne surface, one quarterProductionProduction approval workflowAdvancedMultiple surfaces / markets
Introduced at Pilot
Disclosure assembly to your surfaces✓✓✓
Brand owner approval control✓✓✓
Deployed-surface baseline✓✓✓
Introduced at Production
Reporting by surface—✓✓
Approval workflow in your systems—✓✓
Approved write-back—✓✓
Messaging-channel integration—✓✓
Introduced at Advanced
Multi-market surfaces——✓
Cross-surface record packs——✓
High conversation volume——✓
Multi-surface disclosure controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build priceConfirmed after discovery based on integrations, workflow complexity, conversation volume, approval controls and deployment requirements.
Separate from buildBuild 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 bringWe build with it
01Your surfaces and the systems behind them→Surface inventory mapping and notice captureWeek 1
02Representative transcripts and handover events→Record binding, disclosure logic and the coverage baselineWeek 2
03Your disclosure policy and the wording it names→Surface mapping, role questions and the automation boundaryWeek 1
04Access to relevant APIs, feeds or exports→Surface, vendor and handover-source assessment, then integration setupWeek 2
05Conversations you would not want replayed→Handover cases and the evaluation suiteWeek 4
06What no chat transcript may establish→Completeness scoring, review routing, guardrails and approval controlsWeek 3
07A named brand owner to approve the record→Approval workflow, then pilot and production validationWeeks 5–6
Nothing else is requiredDeployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
These widths measure the work and not the drawing, which is why a single week carries two phases.
PhaseW1W2W3W4W5W6
DiscoveryW1
BuildW2 – W3
EvaluateW4 – W5
Pilot & LaunchW5 – W6
Week focusW1Conversation workflow discovery, surface mapping and the automation boundaryW2Source integration and the disclosure-coverage baselineW3Record assembly, disclosure logic and approval controlsW4Evaluation suite, handover cases and failure-mode testingW5Channel integration, pilot surfaces and targeted correctionsW6One deployment year run under the brand lead, then Agent Care handover
Reading the bandA bar runs across the weeks its own work is named in. Two phases share the fifth because the work does.
At the end of W6Once the conversation record validates, Agent Care assumes the agent.
DurationSix-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Conversational AI agent
Build a disclosure agent around the surface your customers take for a person.
Show us one branded surface and the turn it opens on. Art. 50(1) names the provider, not the deployer, and a bot put into service under your own name can make you that provider under Art. 3(3). Chapter IV asks no deployer for a log, so yours is voluntary evidence against Art. 99.
