Main site

Travel & Hospitality AI agent · Loyalty

Loyalty-Programme Support AI Agent

Answer member questions by quoting the balance and the rules held in the system of record, hold every award for the programme team, and refuse anything the terms do not already say.

Typical delivery
4–6 weeks
Deployment
Your stack
Team releases
Quoted only
After launch
Agent Care

What this agent does

Quotes the record, not the outcome

In

  1. 01

    Ingest the member record, the balance and the programme terms from supported loyalty systems.

  2. 02

    Normalise tier, earning and award language, so each answer carries the record it was read from.

Reason

  1. 03

    Quote the balance as the system held it at that moment, and say it is a record and not a promise.

  2. 04

    Cite the award rule from the published terms, in the words of the terms, without restating it.

  3. 05

    Refuse a forward-looking answer about availability, value or a chart, rather than softening it.

Decide

  1. 06

    Flag partner earning, transfers and co-brand credit, where another programme's terms and regime apply.

  2. 07

    Hold every award, adjustment and goodwill decision for the programme team who authorise it.

Out

  1. 08

    Retain the question, the record quoted, the flags, the amendments and the words actually sent.

  2. 09

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

Product statement

The agent quotes the record and the terms; the programme team decides any award or adjustment, and the operator remains bound by whatever is said.

Example workflow

One question, record to answer

  1. 1Member question receivedApp, web chat, contact centre, email or a partner's own channel
  2. 2Record assembledBalance, tier, posted activity, pending adjustments and the terms in force, each with its source
  3. 3Answer draftedQuoted balance, rule cited, confidence
  4. 4Controls appliedPromise checks, forward-looking refusals, partner-terms flags and confidence threshold

No human action required

Stages 1 to 4 run unaided, and nothing is committed at any of them — the agent is quoting, and the team's lane opens at the confidence gate.

5DecisionBranches at the confidence threshold
High confidence

Goes to the programme team to send.

Low confidence

Adds a programme-owner read first.

Team sends it

The answer is held with the record quoted, its flagged lines and the confidence.

Send · Amend · Escalate to the programme owner

Sent — the member has the answer

  1. 6Loyalty systems updatedOnly where write access and approval policy allow it
  2. 7Outcome evaluatedAnswer amendments, flagged-line outcomes, repeat contacts and post-send corrections

Amendments

Every team amendment is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Telling a member what their points would be worth.
    • Saying an award seat or an award night is available.
    • Promising that a stated balance will be honoured.
    • Characterising a devaluation or a change of terms.

  • Automation boundary

    Agent acts unaided

    • ✓Quote the balance the system of record held at that moment.
    • ✓Cite the published rule that answers the question,.
    • ✓Name whose terms govern, since an airline scheme is not.
    • ✓Refuse what cannot be quoted, and route the member to a person.

    Any write happens inside the boundaries agreed at implementation, never ahead of approval.

    • Granting an award, a waiver or a goodwill credit.
    • Reinstating expired points or a closed account.
    • Moving points between accounts or to a partner.
  • Changes to answer, escalation or retention rules.

Example output

One member answer, annotated

Everything the agent says is attached to the record it was quoted from.

Loyalty answer output · single member

Illustrative example
Member
Mid-tier member
Quoted line
The record shows this now; it is not a promise that it will stand
Balance at read
84,000 pts
Source of record
Loyalty system of record
Confidence
91%
Attribution
Programme name and terms

As received

Taken from the loyalty system of record and the terms in force — nothing on this side is written by the agent.

Records quoted

  • Balance at read time
  • Posted activity
  • Published programme terms

Why this wording

It reports what the record held, and stops short of what the terms do not promise.

What the score decides

Below the configured threshold the answer picks up an owner read before the team sends it.

Action

SendAmendEscalate to the programme owner

Value

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 question

From the member record

03Quoting

Quote from the record

Draw on the balance, the posted activity and the programme terms in force, and nothing beyond them.

01Approved path

The promise is the one you make

Routine balance and rule questions reach the team already quoted and already sourced.

02Human review

Send review to what was refused

Every refusal and every partner flag is put in front of the programme team, so the read starts where commitment risk sits.

04

Build an evidence trail

The answer, the balance record behind it and the person who authorised any change stay on the account.

Integrations

Typical integrations

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

  • Loyalty platforms

    • Comarch
    • Kognitiv
    • Salesforce Loyalty
    • Antavo

  • Reservations and PSS

    • Amadeus
    • Sabre
    • Opera Cloud
    • SynXis

  • Service desk

    • Zendesk
    • Salesforce
    • Genesys
    • Intercom

Agent

Loyalty-programme support

  • Reads the record
  • Quotes the terms
  • Holds for the team

  • Partners and co-brand

    • Card issuers
    • partners
    • Transfer and points APIs

  • Observability & evaluation

    • OpenTelemetry
    • 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 member

Each control encloses the last. Whatever slips past all of them is named below.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeCut the agent back to balance retrieval when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt, answer-library and terms-configuration changes.Track
L4TraceabilityRecord the question, the record quoted, the flags, the amendments and the send time.Record
L3Team releaseHold answers for the programme team; it governs sending, not whether a sent answer is right.Gate
L2Policy guardrailsTest drafts against configured promise, refusal and partner-terms rules; a failure returns the draft.Restrict
L1Confidence thresholdsRoute low-confidence answers to a programme-owner read before the team sees them.Require review
Model coreAnswer drafted — the quoted balance, the rule cited, flagged lines and confidence
L1 – L2
Test whether an answer may stand
L3
Puts the sending in the team's hands
L4 – L5
Keep the answer and the balance record behind it
L6
Cuts back to balance retrieval when signals degrade

How Nestack evaluates it

Evaluate the answering workflow — not only the sentence at the end.

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

Surface — the answer the member reads

  1. E1Final-output evaluationDid every statement match the record or the published terms?
  2. E2Step-level evaluationDid the agent read the right account, the terms in force and the regime governing them?
  3. E3Tool evaluationDid it read and write the correct account and the correct field?
  4. E4Confidence calibrationDo low-confidence answers actually attract more team amendments?
  5. E5Slice evaluationHow does performance change across specific member groups?
  6. E6Business outcomeHow many answers needed an amendment or a correction after sending?

Floor — the commitment the operator answers for

Failure modes

Where each failure originates in the agent

Seven ways an answer goes wrong, placed by stage.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • CF-03

      Stale balance read

      A balance is read after the account has already moved past it.

    Stage gathers

    Accounts, balances, activity and the terms in force

  2. 02 · Reasoning

    2 modes

    • CF-04

      Promise beyond the terms

      An answer undertakes what the programme's terms disclaim.

    • CF-06

      Rule restated, not quoted

      A published rule is paraphrased into a different rule.

    Stage proposes

    The answer the team sends and the member keeps

  3. 03 · Tool / write

    2 modes

    • CF-02

      Award described as held

      Availability seen on a screen is reported as secured.

    • CF-05

      Wrong programme's terms

      A partner's rules are answered under the host's terms.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • CF-01

      Value put on points

      An answer attaches a cash worth to a points balance.

    Stage returns

    The answer the team sends and the member acts on

  5. 05 · Change / Version

    1 mode

    • CF-07

      Silent promise regression

      A model or rule change widens what the agent will undertake.

    Stage tracks

    Model, prompt, answer library and terms config

  • Sev-1 · a commitment leaves the agent
  • Sev-2 · a wrong balance reaches a member
  • Sev-3 · record degrades, answer routes to review
See the complete Travel & Hospitality failure-mode catalogue

Affected slices

Overall quality can hide one bad cohort

The members under-counted in an aggregate figure are the ones who asked about an award, a partner credit or an expiry The cohorts that carry it are named, not averaged away..

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Award availability asks5.3%3.3×Review
Partner and transfer credit4.0%2.5×Review
Expiry and clawback cases3.2%2.0×Watch
Plain balance questions1.3%0.8×Normal

Bar: team-amendment-rate lift vs. plain-balance baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

The loop closes in the suite, not the review

The cycle does not end in a post-mortem. It ends in a case the next release has to clear, and that suite is what the next member answered is measured against.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Team-amendment rate rises in a question slice.

  2. 02

    Diagnose

    Not the model first — the account, the terms quoted and the answer sent are read together until the cause narrows to one.

  3. 03

    Improve

    Each change carries a version and the accounts that forced it.

  4. 04

    Verify

    Release waits on the affected cases going green again.

  5. 05

    Learn

    The case is kept for good, and the answer library is rewritten.

  6. Learn → Detect

    The return edge. The suite the next answer meets is 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, answering workflow, evaluation, integration, then production validation and handover.

Workstream

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

  1. 01Member-question workflow discovery and boundary setting.Week 1
  2. 02Loyalty and service-desk source assessment.Week 1, Week 2
  3. 03Programme terms, tier and partner-terms rule mapping.Week 1, Week 2
  4. 04Member-record ingestion and normalisation.Week 2
  5. 05Quoting logic and record binding.Week 2, Week 3
  6. 06Confidence scoring and owner routing.Week 3
  7. 07Programme-team release workflow.Week 3, Week 4
  8. 08Loyalty-platform and desk integration.Week 5
  9. 09Promise and award-rule cases.Week 4, Week 5
  10. 10Guardrails and authorisation controls.Week 4
  11. 11Account-trail instrumentation.Week 5, Week 6
  12. 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 tierPilotOne programme, one teamProductionProduction loyalty systemsAdvancedMultiple programmes / brands
Introduced at Pilot
Quoting from your record and terms✓✓✓
Programme-team release✓✓✓
Answer-accuracy baseline✓✓✓
Introduced at Production
Reporting by member tier—✓✓
Approval workflow in your systems—✓✓
Approved write-back—✓✓
Loyalty-platform integration—✓✓
Introduced at Advanced
Multi-programme terms rules——✓
Multi-stage programme approvals——✓
High contact volume——✓
Multi-programme answer controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on integrations, workflow complexity, transaction 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.

  1. 01Your programme terms and tier definitions

    Member-record ingestion and terms mapping

    Week 1

  2. 02Member questions you have already answered

    Quoting baseline, answer scoping and record binding

    Week 2

  3. 03Your refusal rules and who authorises awards

    Programme terms, tier and partner-terms rule mapping

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    Loyalty, reservation and desk assessment, then integration setup

    Week 2

  5. 05Promises you would not want kept

    Award cases and failure-mode testing

    Week 4

  6. 06What no answer may commit the programme to

    Confidence scoring, refusal routing, guardrails and authorisation controls

    Week 3

  7. 07Named programme owners to release awards

    Authorisation workflow, 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

The bands sit on the weeks the work occupies, which is why the fifth carries two kinds at once.

Phase

W1

W2

W3

W4

W5

W6

  1. DiscoveryW1
  2. BuildW2 – W3
  3. EvaluateW4 – W5
  4. Pilot & LaunchW5 – W6

Week focus

W1Member-question discovery, terms mapping and the automation boundary

W2Source integration and the quoting baseline

W3Answering workflow, confidence logic and authorisation controls

W4Promise cases, refusal guardrails and failure-mode testing

W5Loyalty-platform integration, pilot members and targeted corrections

W6One award cycle answered under the programme team, then handover

Reading the band
A bar covers the weeks its work is named in, and nothing else. The week 5 overlap is real, not padding.
At the end of W6
The cycle closes the validation and Agent Care takes over monitoring.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Travel & Hospitality AI agent

Build a loyalty support agent that quotes the record and promises nothing.

No regulator has written a rulebook for the programme itself, and your terms let you change the chart whenever you choose. So the one sentence anyone can hold you to is the sentence your agent wrote — which is why the first question we ask is whose signature stands behind an answer that reaches past the terms. Show us the terms, the system of record and who authorises an award, and the refusals get built around them.

Typical delivery
4–6 weeks
Deployment
Your programme
Team releases
Quoted only
After launch
Agent Care available

Nestack Agents · Loyalty support

AGT-TH-05 · Agent Care available after launch

Build a loyalty support agent

Four questions so the call starts from your actual setup — then pick a time that suits you.

About you
Your members

No obligation. We’ll map the workflow, name the automation boundary and tell you if an agent is the wrong answer.