Collections & Negotiation AI Agent (Premium Arrears)

What this agent does

Chases the arrears, settles nothing

Take the arrears as billing has them — policy, instalment plan, amount past due, days late and the notices already sent.

Read the contact record first: consents, channel and language, opt-outs, disputes and every attempt already made.

Reason

Check what the policy and the notice rules in that jurisdiction actually permit before a word about lapse is written.

Match the account to a payment plan the file already supports — the instalments paid, when, and how often they slipped.

Scan what the customer said for hardship, illness, bereavement, insolvency, service or anything else that means stop.

Decide

Inside the script, the hours and the attempt cap, the contact goes out with the offer a named officer already signed.

A vulnerability signal, a dispute or a cease-contact request ends the automated path and books a named officer.

Out

Send approved wording on a consented channel, with the required disclosure, counted against the cap.

Retain the script version, the authority in force, the consents relied on, what was said and who cleared it.

→ Product statement

This page covers what the policyholder owes the insurer — premium and instalment arrears, lapse prevention, payment arrangements and reinstatement. The agent makes the contact itself, but only on a script, a cadence, an hours window and an attempt cap a named officer signed. Settlement terms, waivers, write-offs, cancellation and any placement with an agency are that officer's decision. Recovery between carriers — subrogation, deductible and salvage negotiation — is a different counterparty and is not in scope here.

Example workflow

One arrears case, end to end

AgentHuman

1Arrears case openedA past-due instalment or premium with the plan, the days late and the notices already served

2Permissions readConsents, channel and language preference, opt-outs, disputes, cease-contact requests and the attempt count so far

3What may be said checkedThe grace period, the notice already served and the statutory clock, so nothing is said about lapse that is not true yet

4Affordability and signalsThe plan the payment history actually supports, and a scan of the customer's own words for hardship, illness or insolvency

No human action required

Stages 1 to 4 run without a person in the loop — the permission read, the notice clock, the affordability check and the signal scan all finish before anything leaves. Nothing has been said to the customer yet.

5DecisionSplits on whether the contact is inside the rules and the case is clear of signals

Inside the rules, no signals

Sends the approved contact and offer.

Any signal, or outside the rules

Stops, and books a named officer.

Collections officer

Takes the case with the signal named, the balance as billing holds it and every prior contact shown. Terms, waivers and anything adverse are theirs alone.

Agree terms · Hold · Refer

Decided — handed back▼

6Contact sent and countedApproved wording on a consented channel, inside permitted hours, with the disclosure attached and the attempt logged

7Outcome evaluatedContact compliance, dispute and cease-contact handling, signal recall, complaints and escalations, by cohort and channel

Officer changes

Every term an officer sets differently from the offer is counted.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required8 items

Agreeing a settlement, waiver or write-off.

Cancelling, lapsing or non-renewing for non-payment.

Serving the notice that starts a cancellation clock.

Placing a balance with a collection agency.

Automation boundaryAgent acts unaided

✓Contact on the script, hours and attempt cap an officer signed.

✓Offer only the arrangements already on the approved plan matrix.

✓Read the notice clock before saying anything about lapse.

✓End the automated path on a vulnerability, dispute or stop request.

Write actions run only inside the approval boundaries agreed during implementation. The agent may send an approved contact and record an approved plan. It may not agree terms outside the matrix, and nothing adverse to the customer runs unattended, on any tier or in any configuration.

Reporting a balance to a credit bureau.

Deciding that a disputed balance is owed.

Discussing the debt with anyone but the customer.

Changing the script, the cadence or the authority.

Example output

One arrears contact, annotated

Everything the agent proposes is attached to the permission and the notice position it came from.

Collections output · single contactIllustrative example

Case

Notice position

Consent on file

Proposed action

Confidence

Signals found

Second instalment past due

In grace, unserved

SMS and email

Reminder plus plan offer

93%

None on this account

As receivedThe balance as billing holds it, the notice position under the policy and that jurisdiction, and the consents.

Checks cleared Inside permitted hours Under the attempt cap No dispute or opt-out

Why this offerFrom the plan matrix an officer signed, matched to the instalments already paid.

ActionAgree termsHoldRefer

What the score decidesWhether this contact may send unattended — never that a signal is absent.

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 account in arrearsFrom billing and the instalment ledger

03Script & cadence

Apply the rules that govern the contact

Permitted hours in the customer's own time zone, the attempt cap across every channel, the consents on file and the notice clock — read before anything is written, and stamped with the script version.

01Approved path

Let the routine reminder run

An account inside the script, under the cap and clear of signals gets its reminder and its approved plan offer without an officer opening it. The cadence is narrow by design, and it is written down.

02Human review

Hand the hard ones over early

A hardship disclosure, a dispute, a stop request or an arrangement the payment history will not carry ends the path before anything is said, and reaches a named officer with the signal quoted.

04Build an evidence trail

Retain the script and authority version, the consents relied on, the hours and attempt count, the wording sent and the officer's own decision — on both paths.

Integrations

Typical integrations

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

Billing & policy adminGuidewire · Duck Creek
Sapiens · Billing and instalment ledgers

Payments & plansCard and ACH rails · direct debit
Payment plans · reinstatement APIs

Contact & preferencesPreference centre · consent records
Do-not-contact lists · language on file

Agent

Collections & arrears contact

Reads permissions
Sends on script
Stops on signals

Telephony, messaging & casesDialler and IVR · SMS and email
Case management · Agency placement

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 customer

Each control wraps the one inside it. A contact clears every layer before it is sent, and nothing adverse to the customer clears them at all.

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

L6Contact recordScript version, consents, attempts, wording sent and who decided.Record

L5Authority gateTerms, waivers and anything adverse need a named officer.Gate

L4Signal stopHardship, vulnerability, dispute or insolvency ends the automated path.Stop

L3Script lockApproved wording only, with the disclosures the rules attach to it.Restrict

L2Cadence & hoursAttempt caps across all channels, and hours in the customer's time zone.Cap

L1Permission checkConsent, channel, language, opt-outs and stop requests, read first.Check

Model coreProposed contact — the script it matched, the plan it may offer, and confidence

L1 – L2Decide whether this contact may go at all

L3Decides what may actually be said

L4Takes the customer out of the cadence

L5 – L6Put it with a person, and keep the trail

How Nestack evaluates it

Evaluate the contact — and who it reached.

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

Surface — the message the customer receives

Depth of coverage ▼

E1Contact-rule evaluationWas every contact inside the hours, the cap and the consent on file?

E2Content evaluationDid the wording state only consequences the notice position allows?

E3Signal recallDid it catch the hardship the customer disclosed, in their words?

E4Tool evaluationDid it read the right account, balance and preference record?

E5Slice evaluationHow does performance change across cohorts, channels and languages?

E6Business outcomeComplaint and escalation rate, and what happened to those who disclosed.

Floor — the complaints and the harm behind them

Failure modes

Where each failure originates in the agent

Seven failure modes plotted against the five stages of the agent lifecycle.

Agent lifecycleDirection of processing →

01 · Permissions2 modes

CN-01

Stop request not honoured

An opt-out on one channel; the cadence ran on another.

CN-02

Chased a balance not owed

Already paid, or disputed and the path never paused.

Stage readsConsents, opt-outs, disputes and every prior attempt

02 · Cadence & timing2 modes

CN-03

Contacted outside permitted hours

Time zone read from the policy address, not the number.

CN-04

Cap counted per channel

Each channel under its own limit, the total over all.

Stage checksThe permitted hours, the cap and the notice clock

03 · Signals1 mode

CN-05

Hardship disclosure missed

The customer named it plainly; the cadence carried on.

Stage scansWhat the customer actually said, in their own words

04 · Offer1 mode

CN-06

Plan the file cannot carry

An arrangement offered that fails on its own first date.

Stage proposesThe arrangement from the matrix an officer signed

05 · Contact1 mode

CN-07

Consequence overstated

Wording said cover ends before the notice makes it true.

Stage sendsThe wording the customer receives, and when it goes

Sev-1 · the contact should never have gone Sev-2 · the customer is chased for the wrong thing Sev-3 · the pattern breaches, not the contact

See the complete Insurance failure-mode catalogue →

Affected slices

The cadence lands hardest where people say least

A few cohorts carry most of the breaches — a disclosure nobody caught, a contact after an opt-out, wording that overstated what happens next. A cadence tuned on tenure and postcode stands in for things a carrier may not act on directly. Nestack reports performance by slice, not only in total.

Slice performance — reported separately, not only in aggregateIllustrative example

Slice	Failure rate	Lift	Status
Customers who disclosed hardship	7.3%	3.5×	Review
Non-English contact preferences	5.2%	2.5×	Review
Disputed and part-paid balances	3.8%	1.8×	Watch
Repeat routine first reminders	1.3%	0.6×	Normal

Bar: contact-breach lift vs. routine first-reminder baseline · scale 0–4.0× · tick marks the 2.0× review threshold 2 of 4 slices over threshold

Evidence-linked improvement

Every complaint lands on the cadence that caused it

When someone complains about being chased, the breach usually sits in the cadence, the script or a signal nobody caught — not in that contact.

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

01Detect

Contact breaches, complaints or missed signals move in one cohort.

02Diagnose

Traced to the permission read, the cadence, the signal scan or the wording.

03Improve

The script, cadence or plan matrix is changed, re-signed by the collections lead and version-linked.

04Verify

Re-run against held-out cases, including every complaint and every disclosure.

05Learn

The missed signal becomes a regression case and a phrase the scan must catch.

Learn → DetectThe return edge. Every cycle re-reads the cadence itself — a threshold change that added a channel or an attempt without a signature is reversed before the next cycle runs.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, contact rules and permissions, script and offer logic, evaluation, contact-channel integration, then production validation and handover.

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

01Workflow discovery and automation-boundary definition.

02Contact rules, permitted hours and caps.

03Billing, policy-admin and payment APIs.

04Consent, preference and opt-out integration.

05Notice clock, grace period and reinstatement.

06Payment-plan matrix and authority sign-off.

07Script and disclosure wording sign-off.

08Vulnerability and hardship signal detection.

09Contact-compliance and signal-recall evaluation.

10Officer review workflow and agency handoff.

11Telephony, messaging and complaint routing.

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 line, nothing sends ProductionOne line, live contact AdvancedMulti-line / multi-jurisdiction

Introduced at Pilot

Arrears case assembly and balance read✓✓✓

Consent, opt-out and stop-request checks✓✓✓

Draft mode — proposes, sends nothing✓✓✓

Vulnerability and hardship signal detection✓✓✓

Baseline evaluation✓✓✓

Introduced at Production

Permitted hours and attempt caps—✓✓

Approved scripts and required disclosures—✓✓

Live contact on consented channels—✓✓

Plan matrix inside a signed authority—✓✓

Introduced at Advanced

Multi-jurisdiction contact rules——✓

Agency handoff and enterprise controls——✓

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

Final build priceConfirmed after discovery based on the lines and jurisdictions in scope, the billing, payment and telephony systems involved, the channels and languages supported, contact volume, the authority and caps agreed, 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 lines, balances and jurisdictions in scope → Workflow discovery and automation-boundary definitionWeek 1

02Your contact rules — hours, caps, channels and disclosures → Contact rules, permitted hours and attempt capsWeek 1

03Access to billing, policy admin and the preference centre → Consent, preference and cease-contact integrationWeek 2

04Grace periods, notice templates and reinstatement rules → Notice clock, grace period and reinstatement logicWeek 2

05The plans you will offer, and who may waive anything → Payment-plan matrix and settlement-authority sign-offWeek 3

06Complaints you have had about collections contact → Evaluation suite, regression cases and failure-mode testingWeek 4

07The named officer who holds the authority and the stop control → Officer review workflow, agency handoff and the audit trailWeeks 5–6

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

Delivery timeline

Four phases across six weeks

Phases are drawn over the weeks they actually occupy. Nothing reaches a customer before week 6 — the cadence runs in draft until the stops are proven.

Phase W1W2W3W4W5W6

Discovery W1

Build W2 – W3

Evaluate W4 – W5

Pilot & Launch W5 – W6

Week focus W1Lines, jurisdictions, contact rules and who may waive anything W2Billing and preference-centre integration, and the notice clock W3Scripts, disclosures, the plan matrix and signal detection W4Evaluation suite, contact-compliance and signal-recall testing W5Telephony and messaging integration, in draft — nothing sends W6First live contacts inside the signed caps, then handover

Reading the bandBars span only the weeks their work is named in. Draft running in week 5 is real — the agent proposes contacts and sends none.

At the end of W6Live contacts have run inside the signed hours and caps, and every waiver in that period was a named officer's.

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

Next step · Insurance AI agent

Write the cadence down before you automate the chase.

Show us how an account reaches arrears, the contact rules you work under, and who may agree a plan or waive anything. We'll set the hours, the caps and the signals that stop it, and show you every phrase the agent may never use.

Discuss this agent →

Nestack Agents · Collections & arrears contactAGT-INS-05 · Agent Care available after launch