Plan-Recommendation AI Agent

What this agent does

Builds the comparison, not the authorisation

Account, tariff and usage records arrive from the billing, CRM and product-catalogue systems in scope.

The consent basis on the account is read first, and no comparison is built where the basis is missing.

Reason

Published plan terms, the price card and the broadband label are normalised into one comparable shape.

Eligibility rules configured for the market decide which plans may appear in the comparison at all.

Each figure shown is bound to the published term it came from, and anything the terms leave open is marked.

Decide

Material restrictions — throttling thresholds, add-a-line and trade-in conditions — travel with the headline.

Location and roaming signals stay out unless express prior authorisation is recorded for that customer.

Out

The comparison, the consent basis and the adviser's decision are kept, so a later complaint is answered from it.

Write actions run only inside the boundaries agreed at implementation; changing a plan is not one of them.

→ Product statement

The agent proposes a comparison; the named adviser decides whether it is shown, and the carrier stays the seller of record.

Example workflow

One recommendation, record to release

AgentHuman

1Account record receivedBilling account, product catalogue, price card and published label

2Consent basis checkedThe approval tier recorded for this customer, and what it permits the comparison to use

3Comparison builtPlan options, restrictions and confidence

4Controls appliedPublished-term checks, restriction checks, market eligibility rules and confidence threshold

No human action required

Stages 1 to 4 run unaided, and no plan moves at any of them — the agent is comparing, and the adviser's lane opens at the confidence gate.

5DecisionBranches at the confidence threshold

High confidence

Goes to the named adviser to release.

Low confidence

Adds a compliance read first.

Adviser approval

The comparison is held with its published sources, its restrictions and the confidence.

Approve · Amend · Send to compliance review

Approved — released to the adviser▼

6Sales systems updatedOnly where write access and approval policy allow it

7Outcome evaluatedAmendment rate, restriction-disclosure outcomes, label surfacing and post-sale complaints

Amendments

Every adviser amendment is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required8 items

Changing a rate plan, or adding a billable add-on.

Verifying the customer's authorisation for that change.

Using usage data where the consent basis is missing.

Quoting a term that is not on the published price card.

Automation boundaryAgent acts unaided

✓Compare published plans inside the category the account already holds.

✓Surface the broadband label next to the plan it belongs to.

✓Check the consent basis on file before a comparison is built.

✓Carry each restriction with its headline, and hold for the adviser.

Any write stays inside the boundaries agreed at implementation, and never moves the plan itself.

Offering a product outside the customer's own category.

Using location or roaming signals to shape an offer.

Calling a device lease a purchase, or a phone free.

Changes to eligibility, consent or approval rules.

Example output

One comparison, annotated

Everything the agent shows is attached to the published term it was taken from.

Recommendation output · single accountIllustrative example

Account

Line shown

Quoted from

Consent basis

Confidence

Disclosure shown

Mobile · two lines

Unlimited data, with the speed threshold shown beside it

Published price card

Total-service, on file

88%

Broadband label link

As receivedTaken from the published price card and the label on file — nothing on this side is written by the agent.

Published terms used Price-card entry Broadband label Consent record

Why this comparisonIt compares inside the category the account already holds — the adviser's call.

ActionApproveAmendSend to compliance review

What the score decidesBelow the configured threshold the comparison takes a compliance read before.

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 comparisonFrom the published terms

03Comparison

Compare from the price card

Draw on the published plan terms, the label on file and the consent basis recorded for that customer.

01Approved path

Compare, and show the label

Routine plan comparisons arrive already built, with their restrictions attached.

02Human review

Send review to the risky offers

Flagged restrictions and low-confidence comparisons are marked, so the adviser's read starts where the risk sits.

04Build an evidence trail

The comparison, the consent it relied on and the adviser who released it stay together.

Integrations

Typical integrations

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

Billing and chargingAmdocs · Netcracker
CSG · Optiva · Ericsson

Product cataloguePublished price card
Broadband consumer labels

CRM and careSalesforce · Dynamics 365
Pega · ServiceNow

Agent

Plan recommendation

Reads the terms
Builds the comparison
Holds for approval

Order and salesOrder management · CPQ
Retail POS · self-service

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 offer

The layers nest. What each one misses is named in the map below.

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

L6Rollback / safe modePull the agent back to the published price card when evaluation or production signals degrade.Roll back

L5Version monitoringTrack model, prompt, comparison-rule and market-configuration changes.Track

L4TraceabilityRecord the terms used, the consent basis, the restrictions shown and the release.Record

L3Adviser approvalHold comparisons for the named adviser; it governs release, not whether the comparison is right.Gate

L2Policy guardrailsTest comparisons against the published terms and the recorded consent tier; a failure returns them.Restrict

L1Confidence thresholdsRoute low-confidence comparisons to a compliance read before the adviser sees them.Require review

Model coreComparison built — plan options, restrictions, label link and confidence

L1 – L2Test whether a comparison may stand

L3Puts the release in an adviser's hands

L4 – L5Show what the comparison was built from

L6Reverts to the published price card when signals degrade

How Nestack evaluates it

Evaluate the recommendation workflow — not only the plan that was shown.

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

Surface — the comparison the customer sees

Depth of coverage ▼

E1Final-output evaluationDid each figure shown match the published term behind it?

E2Step-level evaluationDid the agent use the right consent basis, price card and market rules?

E3Tool evaluationDid it read the correct account and the current price card?

E4Confidence calibrationDo low-confidence comparisons actually attract more adviser amendments?

E5Slice evaluationHow does performance change across specific plan families?

E6Business outcomeHow many comparisons needed an amendment, or a correction after the sale?

Floor — the outcome the carrier answers for

Failure modes

Where each failure originates in the agent

Seven failure modes, placed at the stage each one starts.

Agent lifecycleDirection of processing →

01 · Retrieval1 mode

ZR-03

Superseded price card read

A term is quoted from a price card the carrier has since replaced.

Stage gathersAccount, price card, label and consent record

02 · Reasoning2 modes

ZR-04

Consent tier assumed

A global setting stands in for the approval recorded on the account.

ZR-06

Restriction left off the claim

An unlimited headline is compared without the threshold behind it.

Stage proposesPlan options, restrictions and confidence

03 · Tool / write2 modes

ZR-02

Cross-category offer built

A comparison is assembled outside the category on the account.

ZR-05

Location signal used

Roaming data shapes an offer with no express authorisation on file.

Stage writesOnly where write access and approval policy allow it

04 · Output1 mode

ZR-01

Label not surfaced

A broadband plan is shown without its consumer label beside it.

Stage returnsThe comparison the adviser releases to a customer

05 · Change / Version1 mode

ZR-07

Silent comparison regression

A model or rule change widens what the comparison will assert.

Stage tracksModel, prompt, comparison rules and market config

Sev-1 · a plan changed without authorisation Sev-2 · a wrong term reaches the customer Sev-3 · price card stale, offer routes to review

See the complete Telecom failure-mode catalogue →

Affected slices

Overall accuracy can hide one plan family

Aggregate comparison accuracy can sit inside its band while bundled broadband and mobile carries most of the amendments. Nestack reports the amendment rate by slice, not only in total.

Slice performance — reported separately, not only in aggregateIllustrative example

Slice	Failure rate	Lift	Status
Bundled broadband and mobile	5.3%	3.3×	Review
Promotional and trade-in offers	3.8%	2.4×	Review
Legacy and withdrawn plans	2.4%	1.5×	Watch
Single-line mobile renewals	1.1%	0.7×	Normal

Bar: adviser-amendment-rate lift vs. single-line renewal baseline · scale 0–4.0× · tick marks the 2.0× review threshold 2 of 4 slices over threshold

Evidence-linked improvement

A cycle ends in a test, not a note

A cycle is done when the miss has become a test the next release has to survive. That suite is what the next comparison shown is measured against.

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

01Detect

Adviser-amendment rate rises in a plan family.

02Diagnose

Not the model first — the price card, the consent record and the rules are read until one cause holds.

03Improve

Version-stamp the change and attach the comparisons that exposed it.

04Verify

The affected cases run again, and a fail stops the release.

05Learn

It becomes a standing test, and the comparison rules change with it.

Learn → DetectThe return edge. The next detection 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, comparison workflow, evaluation, integration, then production validation and handover.

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

01Recommendation workflow discovery and boundary definition.

02Billing and catalogue source assessment.

03Consent-tier, eligibility and disclosure-rule mapping.

04Account and price-card ingestion.

05Comparison logic and term binding.

06Confidence scoring and flag routing.

07Adviser approval workflow.

08Billing and sales-system integration.

09Consent and disclosure cases.

10Guardrails and approval controls.

11Comparison-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 market, one plan family ProductionProduction billing systems AdvancedMultiple markets / brands

Introduced at Pilot

Comparison from published terms✓✓✓

Adviser approval✓✓✓

Comparison-accuracy baseline✓✓✓

Introduced at Production

Reporting by plan family—✓✓

Approval workflow in your systems—✓✓

Approved record write-back—✓✓

Billing-system integration—✓✓

Introduced at Advanced

Multi-market price and term rules——✓

Multi-stage sales approvals——✓

High comparison volume——✓

Multi-market plan controls——✓

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

Final build priceConfirmed after discovery based on integrations, workflow complexity, transaction 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 plan catalogue and published price card → Price-card ingestion and published-term bindingWeek 1

02Representative recommendations already made → Comparison baseline, term binding and restriction extractionWeek 2

03Your consent tiers and disclosure rules → Consent-tier, eligibility and disclosure-rule mappingWeek 1

04Access to relevant APIs, feeds or exports → Billing and catalogue assessment, then integration setupWeek 2

05Comparisons that should not have been shown → Disclosure cases and the evaluation suiteWeek 4

06What must reach an adviser before a plan changes → Confidence scoring, flag routing, guardrails and approval controlsWeek 3

07Named advisers to review comparisons → Adviser approval 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

Phases are drawn over the weeks they actually occupy, which is why week 5 doubles up rather than padding.

Phase W1W2W3W4W5W6

Discovery W1

Build W2 – W3

Evaluate W4 – W5

Pilot & Launch W5 – W6

Week focus W1Recommendation workflow discovery, consent mapping and the boundary W2Billing and catalogue integration, and the comparison baseline W3Comparison workflow, confidence logic and approval controls W4Evaluation suite, disclosure checks and failure-mode testing W5Sales-system integration, pilot comparisons and targeted corrections W6One sales cycle compared under the care team, then handover

Reading the bandEach bar spans only the weeks its work is named in. The fifth week genuinely carries two kinds of work.

At the end of W6The last checks clear on live comparisons and monitoring moves to Agent Care.

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

Next step · Telecom AI agent

Build a plan-recommendation agent around your consent basis.

Bring us your price card, your consent tiers and your disclosure rules. A named adviser signs the comparison before a customer sees it, and the boundary is drawn around that signature.

Discuss this agent →

Nestack Agents · Plan recommendationAGT-TL-02 · Agent Care available after launch