Complaint Handling AI Agent

What this agent does

Runs the clock, not the decision to close

A complaint lands, and no general rule in the US or the EU sets a day to answer by.

A clock applies by status, not by being a company, and the regimes rarely agree.

Reason

A complaint under Reg E allows sixty days to notify and ten business days to investigate.

A clock under Reg Z runs thirty days to acknowledge, two complete billing cycles to resolve.

A complaint against an airline runs thirty days to acknowledge and sixty to respond.

Decide

A clock under PSD2 Art. 101 runs fifteen business days, thirty-five where exceptional.

A complaint at a UK authorised firm owes a prompt acknowledgement, then eight weeks.

Out

A complaint is closed by a named resolution owner, never here, and the clock trail is kept.

Execute write actions only inside the approval boundaries agreed during implementation.

→ Product statement

Classifying, starting the clock, tracking what is owed and drafting the response belong to the agent. Resolving the complaint belongs to a named resolution owner, who signs what goes out.

Example workflow

One complaint, arrival to response

AgentHuman

1Complaint receivedEmail, web form, a phone note, a regulator referral or a thread escalated from the desk

2Regime and clock readThe regimes the company sits under, the status attaching each and the event a clock runs from

3Clocks started and trackedThe classification, the clocks running, what is owed and confidence

4Controls appliedRegime checks, clock-start checks, second-regime checks and confidence threshold

No human action required

Stages 1 to 4 run unaided, and nothing is sent at any of them — the agent is tracking, and the owner lane opens at the confidence gate.

5DecisionBranches at the confidence threshold

High confidence

Goes to the named resolution owner.

Low confidence

Adds a compliance read first.

Resolution owner review

The complaint is held with its clocks, what is owed and the drafted response.

Approve · Amend the draft · Send to compliance

Approved — by the named owner▼

6Case and complaint records updatedOnly where write access and records policy allow it

7Outcome evaluatedClock accuracy, classification corrections, deadlines met and what review found

Amendments

Each owner amendment is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required8 items

Deciding that a complaint is resolved.

Deciding which regime a complaint falls under.

Sending a final response to a complainant.

Offering redress, a refund or a goodwill gesture.

Automation boundaryAgent acts unaided

✓Match each complaint to the regimes configured.

✓Start the clock the regime attaches, at the event that triggers it.

✓Track what is owed by when, and show what is at risk.

✓Draft the response and hold it back for a named resolution owner.

No complaint is resolved and no response is sent here; a named resolution owner does both.

Judging whether a deadline was in fact missed.

Reporting a complaint outcome to a regulator.

Agreeing an extension with the complainant.

Changes to clock, regime or approval rules.

Example output

One complaint, annotated

This serves a desk that may have to show, long after the file closed, which clock a complaint ran on and when it started; below is one complaint exactly as the agent leaves it.

Complaint output · single caseIllustrative example

Complaint

What was raised

Clock applied

Regime of record

Confidence

Held for

Inbound, card dispute

A charge the customer says was never theirs

Ten business days, running

Reg E, notice 3 August 2026

Held unsent

The named resolution owner, by name

As receivedTaken from the complaint as it arrived — nothing on this side is written by the agent.

What the record holds Complaint as received Clock start and basis Drafted response

Why nothing left hereResolving a complaint is a call a named resolution owner makes.

ActionApproveAmend the draftSend to compliance

What the score decidesBelow the configured threshold a complaint picks up a compliance read before the owner.

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 complaintFrom the moment it arrives

03Classification

Read the regime, then hold

The customer-success escalation summarisation agent compresses an angry thread for an executive; this one tracks what is owed on the complaint and by when.

01Approved path

Nobody owes you a deadline

The first deadline reaching an ordinary trader is scheduled, not in force: from September 2028, twenty working days to tell an ADR entity whether you will take part — not to take part.

02Human review

What was checked, and not found

No general duty binds an ordinary business, in the US or the EU, to acknowledge, investigate or resolve a complaint by any date, and the EU carved out procedures before complaint-handling systems operated by the trader. No register, no retention, no report, no signature, no filing.

04Build an evidence trail

The complaint, the clock that applies to it and the person who answered stay together.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

Complaint intake channelsEmail · web form · phone notes
Regulator and portal referrals

Case managementZendesk · Salesforce Service
Case, queue and owner fields

Records and retentionSharePoint · Box · Google Drive
Complaint files and the clock trail

Agent

Complaint handling

Reads the complaint
Starts the clock
Holds for the owner

Regulator and ADR channelsCFPB portal · ADR entities
Referred complaints and the replies out

Observability & evaluationOpenTelemetry · Langfuse
Supported monitoring/evaluation sources

Integration availability depends on the client's existing systems and API access.

Agent controls

Six layers between the model and the response

Six ladders against one wall, the last the longest. What is reached is set out in the map below.

L6 · Outermost — last line of defenceInward → L1 · closest to the model

L6Rollback / safe modeFall back to logging the complaint alone when evaluation or production signals degrade.Roll back

L5Version monitoringTrack model, prompt, regime-rule and clock-configuration changes.Track

L4TraceabilityRecord the complaint, the regime read, the clock start and every action taken.Record

L3Owner approvalHold the response for a named resolution owner; the hold governs sending, not whether the answer is right.Gate

L2Regime guardrailsTest each response against the configured regime rules; an acknowledgement omitting what a rule requires returns it.Restrict

L1Confidence thresholdsRoute a weak classification to a compliance read before the owner sees it.Require review

Model coreComplaint classified — the regime read, the clocks running, what is owed and confidence

L1 – L2Test whether a response may go

L3Leaves the sending to a named owner

L4 – L5Keep the complaint and the clock behind it

L6Escalates to a person when signals degrade

How Nestack evaluates it

Evaluate the whole clock — not only the response that goes out.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — the response the customer reads

Depth of coverage ▼

E1Final-output evaluationWas the complaint run on the clock its regime actually sets?

E2Step-level evaluationDid the agent read the right regime, the right status and the live rules?

E3Tool evaluationDid it read and write the correct case and the correct clock?

E4Confidence calibrationDo weak classifications actually attract more owner corrections?

E5Slice evaluationHow does performance change across specific complaint classes?

E6Business outcomeHow many complaints needed a correction before the owner approved?

Floor — the deadline the company answers for

Failure modes

Where each failure originates in the agent

Seven failure modes, each placed where the clock first goes wrong.

Agent lifecycleDirection of processing →

01 · Retrieval1 mode

PG-03

Wrong regime status read

The status that attaches the clock is misread.

Stage gathersThe complaint, the status and the clock owed

02 · Reasoning2 modes

PG-04

Clock started at the wrong moment

The clock is run from the wrong event.

PG-06

Holding reply treated as final

An interim answer is filed as the response.

Stage proposesThe regime read, the clocks running and confidence

03 · Tool / write2 modes

PG-02

Thin classification passed on

A weak read moves on without the compliance look.

PG-05

Deadline bound to the wrong case

The clock is filed against another complaint.

Stage writesOnly where write access and approval policy allow it

04 · Output1 mode

PG-01

Complaint run on the wrong clock

The deadline tracked was not the one owed.

Stage returnsThe response the customer reads and the record

05 · Change / Version1 mode

PG-07

Silent reclassification

A complaint is reclassified after its clock ran.

Stage tracksModel, prompt, regime rules and clock dates

Sev-1 · a final response sent unapproved Sev-2 · a deadline missed on the wrong clock Sev-3 · signals degrade, complaint held back

See the complete Customer Support failure-mode catalogue →

Affected slices

Two-regime complaints absorb the corrections

A class-level clock-accuracy figure can read clean while complaints sitting under two regimes carry most of the corrections. Nestack reports the correction rate by class, not only in total.

Slice performance — reported separately, not only in aggregateIllustrative example

Slice	Failure rate	Lift	Status
Two-regime complaints	10.8%	3.7×	Review
Payment and card disputes	7.7%	2.6×	Review
Air travel complaints	4.8%	1.6×	Watch
General service complaints	2.4%	0.8×	Normal

Bar: correction-rate lift vs. general-service baseline · scale 0–4.0× · tick marks the 2.0× review threshold 2 of 4 slices over threshold

Evidence-linked improvement

What a missed clock costs

The loop shuts when the complaint run on the wrong clock is a regression case. That suite is what the next response sent is measured against.

Improvement cycle · five stagesSwitchback — the path turns at Improve and returns at Learn

01Detect

Correction rate rises in one complaint class.

02Diagnose

The complaint answered in nine days, in the one product line where the answer was owed in three, is worked backwards until a single cause is left standing.

03Improve

Every change goes out numbered, with the complaints that caused it attached.

04Verify

Nothing releases while one touched complaint case is still red.

05Learn

The case stays on, and the clock rules are rewritten alongside it.

Learn → DetectThe return edge. The next complaint 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, clock tracking, evaluation, integration, then production validation and handover.

Workstream Week 1Week 2Week 3Week 4Week 5Week 6

01Complaint-workflow and automation-boundary mapping.

02Case, inbox and regulator-portal feeds.

03Regime-to-status and deadline-obligation rule mapping.

04Complaint intake.

05Regime classification and clock start.

06Confidence scoring and review routing.

07Resolution owner review workflow.

08Case-and-inbox integration.

09Clock and classification cases.

10Guardrails and response controls.

11Complaint-trail instrumentation.

12Deployment, documentation and Agent Care handover.

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 complaint class, one queue ProductionProduction support systems AdvancedMultiple regimes / entities

Introduced at Pilot

Clock tracking to your regimes✓✓✓

Resolution owner approval✓✓✓

Complaint-population baseline✓✓✓

Introduced at Production

Reporting by complaint class—✓✓

Owner review workflow in your systems—✓✓

Approved write-back—✓✓

Case-and-inbox integration—✓✓

Introduced at Advanced

Multi-regime clock rules——✓

Multi-stage owner approvals——✓

High complaint volume——✓

Multi-regime deadline controls——✓

Build price From $5,000 From $8,000 Custom quote

Final build priceConfirmed after discovery based on integrations, workflow complexity, complaint 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

01The regimes your company actually sits under → Regime mapping and clock bindingWeek 1

02Representative complaints from each class → Complaint ingestion, clock logic and the classification baselineWeek 2

03Your complaint queues and who owns each → Regime mapping, clock rules and the automation boundaryWeek 1

04Access to relevant APIs, feeds or exports → Case, inbox and portal assessment, then integration setupWeek 2

05Complaints you would not want re-opened → Clock cases and the evaluation roundWeek 4

06What no complaint file may resolve → Confidence scoring, review routing, guardrails and release controlsWeek 3

07A named resolution owner who answers the complaint → Resolution owner review workflow, then pilot and production validationWeeks 5–6

Nothing else is required Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

Widths follow the work rather than the grid, which is why one band runs beneath another in week five.

Phase W1W2W3W4W5W6

Discovery W1

Build W2 – W3

Evaluate W4 – W5

Pilot & Launch W5 – W6

Week focus W1Regime discovery, clock mapping and the automation boundary W2Case and inbox integration and the complaint baseline W3Clock tracking, confidence logic and approval controls W4Evaluation suite, clock cases and failure-mode testing W5Portal integration, pilot complaints and targeted corrections W6One complaint quarter run under the resolution owner, then Agent Care handover

Reading the bandA bar sits on the weeks its own work is named in, and week five carries two by design.

At the end of W6Once the response record validates, Agent Care assumes the agent.

DurationSix-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Customer Support AI agent

Build a complaint-handling agent around the clocks your company actually sits under.

Show us one quarter of complaints and the regimes your company sits under. Not one deadline binds an ordinary business. Several may bind yours, they do not agree with one another, and a single complaint can be running on two of them at once.

Discuss this agent →

Nestack Agents · Complaint handlingAGT-SUP-05 · Agent Care available after launch