Customer
Airline, travel-seller and airport users run live operations on Amadeus. When an issue is passed on without ownership or updates, they wait, chase and repeat themselves.
Fact Impact: interpretationAssociate Director, Customer Care Center (Level 1) · Panel business case
From pass-through to ownership
Make the Customer Care Center the place where every Amadeus customer's issue is owned from first contact and resolved there whenever it safely can be, with AI removing routine effort so people can focus on judgement, empathy and complex problems.
6 unitsone global support organization since Aug 2026, with the CCC as the single entry pointFact
~200people across regions, round the clock and multi-language; the Philippines is the largest regionFact
1 → manystarts with one product; built to take on more products over timeFact
The storyline
- Problem: the front door mostly forwards work
- Diagnosis: seven root-cause hypotheses, tested against a baseline
- Future: L1 that owns, resolves, qualifies, communicates and prevents
- Enablement: AI where Amadeus says the levers are untapped, agent-assist first
- Execution: three horizons; protect and baseline first
- Evidence: ten KPIs with counter-metrics; a value model Amadeus fills in
No Amadeus operational figures (volume, AHT, FCR, SLA, CSAT, cost) were provided, and none are invented here. Baselines are established in the first 60 days.
01 · The problem
The single front door mostly forwards work today
"The Customer Care Center today operates largely as a pass-through layer… with a significant proportion of reported incidents routed to other support teams. The model remains predominantly manual."Amadeus candidate briefing Fact
Today: pass-through
- Cases routed onward instead of owned
- Triage, escalation and ownership rules differ across merged teams
- Manual: automated triage, guided troubleshooting, AI assist and self-service largely untapped
- The customer's experience depends on who picks up
Target: professional L1
- One owner per case, from first contact to closure
- A higher share resolved at first contact, correctly
- Routed once, to the right team, with a complete case package
- Routine and avoidable demand removed or self-served
The interviewer's words: "professionalize L1 as a customer service layer rather than a pass-through."
02 · Why it matters
Pass-through costs customers time, resolvers capacity, and people a future
Operations
Every case that L1 could have closed or qualified better uses capacity in the Customer Resolution Center, Product & Engineering Support and Claims, and adds handoffs and rework.
InterpretationBusiness
More products are coming. A manual pass-through model scales only by adding people. A professional L1 also routes chargeable Airlines Premium Services requests correctly.
Fact Leakage: assumptionPeople
A routing desk offers no mastery and no future. Amadeus wants a genuine progression path from the front line into specialist roles.
Fact03 · Current state
We know the shape of the problem, but not yet its size
What Amadeus has told us Fact
- Six-unit global organization since Aug 2026; the CCC is the single entry point
- Phone, portal, chat and automated channels; multi-language; round the clock
- ~200 people; four senior managers (three regional plus a service being folded into L1)
- Starts with one product, with more to follow
- A "significant proportion" of incidents is routed onward
- Four levers largely untapped; governance not standardized
- The role is the final escalation point for major customers
What we must measure Baseline by Day 60
- Volume, arrival pattern, channel and language mix
- FCR, routed rate, reopen rate
- SLA attainment, backlog age
- CSAT or effort (introduce it if absent)
- AHT, occupancy, cost per contact
- Share of demand that is routine, invalid or avoidable
- Tools, knowledge base, data and AI policy
Targets set before a baseline are guesses. The first deliverable of this plan is a baseline leadership can trust.
04 · Root causes
Pass-through is a symptom: seven hypotheses to test in 60 days
SymptomCases pass through L1 instead of being owned and resolved
H1 Knowledge
Can't find the approved answer fast enough; fragmented across legacy teams and languages
Test: tag sampled routed tickets "resolvable with existing knowledge?"H2 Authority
Knows the fix but isn't permitted to apply it
Test: resolution-action analysis; permissions reviewH3 Tooling
Manual triage; no guided troubleshooting
Test: routing-accuracy sample; reassignment countsH4 Governance
Different triage, escalation and ownership rules by team
Test: returned-to-L1 rate; ping-pong analysis; comms auditH5 Failure demand
We create our own contacts: status chasing, repeats, incident duplicates
Test: contact-reason sampling; 7-day repeat rateH6 Measurement
Scorecards reward handling speed, not resolution
Test: scorecard and incentive reviewH7 People
Uneven onboarding, product depth and progression
Test: performance by tenure and region; attritionHypotheses The Day-60 output is a ranked driver tree. For each top contact driver it gives the root cause, an owner and the lever: eliminate, automate, assist, empower or route better.
05 · Future state
Professional L1 owns, resolves, qualifies, communicates, and prevents
1
Own
One named owner per case from first contact to closure, even when a resolver team does the technical work.
2
Resolve
Close at first contact whenever it is safe, with the knowledge, authority and tools to do it.
3
Qualify
When a case must go on, route it once, to the right team, with a complete case package.
4
Communicate
The customer never chases: acknowledgement, a predictable update cadence and a clear closure.
5
Prevent
Every repeat or avoidable contact is a defect to remove, not volume to handle.
Common triage and escalation rules Proposed
One taxonomy
Product · category · severity/impact · customer tier, identical in every region
Definition of ready
Minimum information per case type before escalation; resolvers return cases only against it
No ping-pong
At most one reassignment without a team-lead review
Ownership rule
Whoever opens the case owns customer communication until closure or a formal transfer
06 · Operating model
A layered front door where the CCC keeps ownership all the way through
- CustomerAirline · travel seller · airport · corporate
- Digital & self-serviceApproved how-to content, case status, known-issue notices
- Virtual agent & intelligent triageIntent, data capture, duplicate and known-incident match, routine answers; always one step from a human
- AI-assisted human L1Owns · resolves · qualifies · communicates
- Senior L1 / product specialistsHarder L1 cases, coaching, knowledge authoring: the first rung of the progression path
- Resolver unitsCustomer Resolution Center · Product & Engineering Support · Claims · Airlines Premium Services
- Product / engineeringDefect fixes at source
Ownership spine
The CCC keeps customer ownership and communication even when the work moves down the stack. The brief says the CCC "stays with the customer." Fact
Knowledge loop
Support Enablement & Procedures owns knowledge Fact. The CCC flags gaps and drafts fixes, the unit approves and publishes, and article accuracy is measured. Proposed
Major-incident mode
A known-issue notice, tickets linked to the parent incident, templated proactive updates, and an L1 incident coordinator connected to the engineering incident lead. Surges are absorbed without flooding resolvers. Proposed
07 · AI-enabled L1
AI goes where Amadeus says the levers are untapped, agent-assist first
| Lever (Amadeus's words) | What AI does | Who decides | When |
|---|---|---|---|
| AI-assisted agent support | Retrieves approved knowledge with sources; summarizes case history; drafts replies and translations; flags missing information | Agent reviews and sends | Pilot · first 90 days |
| Automated triage | Suggests category, product, severity and route; spots duplicates and links to known incidents; pre-fills fields | Agent confirms; auto-routing only once accuracy is proven | Pilot Scale · months 4–12 |
| Guided troubleshooting | Step-by-step flows built from approved procedures, adapted to the case | Agent follows and uses judgement | Months 4–12 |
| Self-service & virtual agents | Routine how-to, status and known-issue answers; structured data capture before handoff | Customer can always reach a human; high-risk intents are never automated | Months 4–12 Year 2+ |
Proposed Why agent-assist first: it is the lowest risk and teaches us fastest, and it improves the knowledge that self-service will later depend on. Customer-facing automation follows the evidence.
08 · Human + AI operating model
Autonomy is earned by evidence, one intent at a time
Level 1
Assist
AI drafts, retrieves, summarizes. A human decides and sends. This is the starting point everywhere.
Level 2
Recommend
AI proposes a category, route or step; the agent approves in one click. Only once audited accuracy meets the agreed threshold.
Level 3
Automate
Only low-risk, reversible, well-defined tasks, with sample audits and a kill switch. The customer can always reach a human.
AI can
- Identify intent and language
- Retrieve approved knowledge, with sources
- Recommend troubleshooting steps
- Summarize case and customer history
- Draft replies and translations
- Flag missing information before escalation
- Link duplicates to known incidents
- Recommend escalation and the right team
Humans own
- Exceptions and ambiguous cases
- Anything with financial or operational consequence
- Customer recovery and complaints
- Sensitive interactions
- Major-customer and executive escalations
- Final escalation decisions
AI must never independently
- Change customer production environments or configuration
- Create, change or cancel bookings or passenger records
- Grant access or reset credentials outside verified flows
- Make commercial commitments (premium charges, credits, SLA concessions)
- Take Claims or financial decisions
- Declare or close a major incident
- Answer from outside approved knowledge
Proposed Controls: grounding in approved knowledge · weekly accuracy audit · comparison groups · kill switch · Security, Privacy and Legal sign-off before any pilot · monthly AI review to promote, hold or demote each intent.
09 · Roadmap
Three horizons: protect and baseline first, then standardize and scale
| Your brief's five themes Fact | First 90 days | Months 4–12 | Beyond year one |
|---|---|---|---|
| Reducing routine demand | Classify demand as routine / invalid / avoidable; fix top knowledge gaps | Self-service and virtual agent for proven routine intents | Proactive notices; defects removed at source |
| Lifting FCR | Baseline; routed-ticket analysis; P1 agent-assist pilot | Scale assist; guided troubleshooting; L1 resolution matrix | Most routine issues resolved instantly; specialists for the rest |
| Introducing automation | AI readiness with Security; pilots against comparison groups | Triage suggestions → Level 2; virtual agent (Level 3 intents only) | AI-enabled front door, governed intent by intent |
| Triage & escalation governance | Interim escalation minimum by Day 30; common rules v1 by Day 60 | One ownership model and QA framework across merged teams | Rules embedded in tooling; continuous calibration |
| Carrying the team through change | SM co-ownership; floor time; pulse; change freeze on non-essentials | Academy, certification, progression path live | Specialist, knowledge and AI-quality career paths |
Horizon 1 · First 90 days
Learn · Protect · Baseline · Make consistent
Days 1–30 · Listen & protect
- 1:1s with the four senior managers; leadership charter
- Daily service-health review to protect SLAs
- Voice of agent, customer and resolver units
- Data, tool and AI-policy inventory
- Interim communication standard + escalation minimum (named owner, info checklist)
Days 31–60 · Baseline & diagnose
- Published baseline: FCR, routed rate, SLA, quality, CSAT
- Demand classified as value / routine / invalid / avoidable
- Routed-ticket analysis tests H1–H7
- Common triage and escalation rules v1
- Quick wins: top knowledge gaps, templates
Days 61–90 · Design & start pilots
- P1 AI agent-assist (one team, with a comparison group; guided-knowledge fallback if AI isn't approved yet)
- P2 Intelligent triage + escalation quality gate
- Governance cadence live
- Self-service intents selected, not launched
- Day-90 readout: baseline, causes, targets
Horizon 2 · Months 4–12
Standardize · Equip · Deflect · Develop
- One ownership model, taxonomy and QA framework across merged teams
- Scale proven pilots; guided troubleshooting
- L1 resolution matrix: more authority where data shows it's safe
- Self-service and a virtual agent for proven routine intents, with human handoff
- Onboarding academy, product certification, progression path
- Product onboarding playbook ready by Q3; next product onboarded when Amadeus schedules it
- Follow-the-sun WFM; major-incident surge playbook
Horizon 3 · Beyond year one
The proactive, AI-enabled front door
- Customer: told before they have to call; routine questions answered instantly; complex issues reach a knowledgeable owner first time
- Cost to serve: products and volume absorbed without linear headcount growth
- People: specialist, knowledge and AI-quality roles; L1 as a talent pipeline
- Insight: contact drivers feed Product and Engineering, so defects are fixed at source
Days 90–120 · Pilot decision gate: for each pilot, scale, extend, adjust or stop, against success criteria agreed before launch and measured against a comparison group.
09b · People & change
Carrying ~200 people through the change is a workstream, not a footnote
Leadership team first
The four senior managers co-own workstreams (rules, a pilot, onboarding, incident readiness) as well as their regions. Shared scorecard, visible credit.
A cadence across time zones
A monthly all-hands repeated per region; a weekly note on what changed, what's next and what we heard; skip-levels in every region in the first 60 days.
Involvement
Change champions per shift and language; agents co-design the pilots and the escalation checklist; "you said, we did".
Capability & careers
Onboarding academy, product certification, coaching from QA and FCR data, and a path to senior L1, specialist, knowledge and AI-quality roles.
Pace & honesty
The organization is new, so fewer priorities, a freeze on non-essential change in the first 30 days, and pulse surveys. AI is to assist, not replace, and year-one capacity is reinvested.
Proposed Senior L1 and specialist roles are formed from existing experienced agents; no net headcount increase is assumed.
10 · How we evidence progress
Ten executive KPIs, each paired with a counter-metric
Illustrative dashboard layout. Every baseline is established by Day 60, and targets are set from it. No values are shown because none have been provided.
Knowledge · authority · triage · agent-assist
FCR ↑ · routed rate ↓ · first-time acceptance ↑
Rework ↓ · repeat contact ↓ · resolver load ↓
Capacity reinvested in ownership, quality, new products
CSAT ↑ · cost per resolved case ↓
Proposed AHT is a diagnostic at L1, not a target, during the transformation. As L1 resolves more, AHT may rise while effort per resolved case falls. I would accept that trade-off, and show it.
11 · Business value
A value framework Amadeus can populate, not a savings promise
Gross opportunity at full run rate–
Formulas and table view
- A. Demand avoided = V × a × eₐ
- B. Digital resolution = (V − A) × s × d × (1 − Cd)
- C. Shift-left = Vh × r × l × (Cr − Cl)
- D. Handling efficiency = Vh × h
- E. Rework removed = Vh × w × wr
Shift-left mostly releases capacity in resolver units, not the CCC budget. It is value to Amadeus's support organization and to the customer, and I would report it that way.
Proposed Year 1: reinvest, don't cut. Released capacity funds ownership, quality and the next product. Whether to take out any cost is a month-12 decision, made on evidence.
12 · Risks & guardrails
The biggest risks are knowledge quality, adoption and pace, and each has an early warning
Likelihood →
| Risk | Mitigation | Early warning |
|---|---|---|
| AI gives a wrong answer | Approved knowledge only, with citations; human sends; audited threshold; kill switch | Audited accuracy ↓, overrides ↑ |
| SLAs slip during change | Daily service review; scoped pilots; no changes in peak or incident windows; rollback | SLA and backlog in pilot vs. control |
| Merged teams don't adopt common rules | Co-design with senior managers and resolver units; adherence measured | Rule adherence by region; returned cases |
| Poor knowledge quality | Joint knowledge loop with Support Enablement & Procedures; top drivers first | "No answer found" rate |
| Over-automation | Risk test; autonomy levels; one step to a human | Repeat contact after bot |
| Change fatigue | Fewer, visible priorities; pace the change; listen | Pulse scores; attrition |
13 · Trade-offs
The trade-offs I would accept, and why
- Consistency before optimization. A common v1 rulebook in 60 days beats a perfect one in nine months.
- Standardize the core, localize the edge. One taxonomy, ownership model and quality standard, with flexibility for language, time zone and customer specifics.
- Agent-assist before customer-facing AI. Slower visible automation, but lower risk and faster learning.
- Resolution over speed. A higher AHT is acceptable if FCR rises and effort per resolved case falls.
- Real deflection over headline containment. A bot that causes a repeat contact is failure demand.
- Reinvest before we reduce. Year-one capacity funds ownership, quality and new products.
- Pilot, prove, then scale. Comparison groups and explicit stop criteria, including stopping things that don't work.
14 · Why me
I have professionalized front lines before, while keeping the service running
| What this transformation needs | Where I've done it |
|---|---|
| Build a professional L1 from a weak base | Built a global bank's retention care operation from zero: 100 agents live on Day 1, with SOPs, QA and SLA governance, scaled to 400. Still running 15+ years later. |
| Unify merged teams under one standard | Six concurrent service lines for a Fortune 500 tech client (PH, India, US) under one set of standards and governance |
| Introduce AI responsibly | Sponsored an AI support assistant across a 300-person tower; moved procedures into NotebookLM so engineers answered from approved documentation, not tribal knowledge |
| Instrument what isn't measured | Introduced CSAT to a B2B account that had none, against pushback; reached 4.6 on a 5-point scale |
| Improve cost to serve without breaking service | ~$4.0M per-FTE model → ~$2.1M managed service (AI tasks, follow-the-sun); most affected staff redeployed |
| Lead through incidents and escalations | 24/7 war rooms, RCAs to client executives; rebuilt a skeptical client director's trust through a daily, then weekly, "Go-Green" plan |
How I lead: business problem → customer impact → operational diagnosis → L1 transformation → technology and AI → measurement → continuous improvement.
15 · Recommendation
The objective is not AI in front of customers. It is ownership at the front door.
The goal is not to put AI in front of Amadeus customers because AI is available. It is to make the Customer Care Center the place where a customer's issue is owned from the first contact and resolved there whenever it safely can be. AI removes routine effort so our people can spend their time on judgement, empathy and complex problems.
Protect
Service levels, from Day 1
Baseline
A trusted baseline by Day 60
Standardize
Common rules within 90 days
Prove
AI in controlled pilots
Scale
Only what the evidence supports
Stephen Rey Salazar · Candidate, Associate Director, Customer Care Center (L1)