AX Field Notes

All casesNo. 080

Philippine Airlines Rebuilds Its Contact Centre Around AI

One agent screen for every channel, then AI chat and a voice agent in place of the IVR

Case no.
080
Primary source
Trade press, Twilio and Sierra customer stories, iTNews interview
Source published
2026-02-25

1. Why this case

Philippine Airlines (PAL) is the flag carrier of the Philippines and is described in its Twilio case study as Asia’s oldest continuously operating commercial airline. It flies to about 31 domestic and 39 international destinations and carried more than 16 million passengers in 2025. Its contact centre handles close to two million inbound calls a year.

The case is useful because it shows the order in which an airline contact centre was rebuilt, and where AI came in. First the platform: contact centre infrastructure moved from on-premise systems to the cloud, and voice, chat, and online servicing were brought onto Twilio Flex so that agents see a passenger’s history in one place. Then a generative AI chat online. Then, in April 2026, a Sierra voice agent, deployed with PAL’s outsourcing partner ibex, that took over three Philippine hotlines and answers callers from the start on them, in place of the phone menu (IVR). PAL says passengers can always choose a live agent instead.

The case also shows how one company describes AI’s share of the work with three different numbers. In February 2026, AI supported about 10% of customer service tasks, about 45% of customer concerns were resolved without transfer to a live agent, and the target for April 2026 was AI handling about 80% of the tasks a live agent can perform. These measure different things.

All figures in this case come from PAL statements published in vendor customer stories (Twilio, Sierra) and in trade press based on them. None is independently audited.


2. Problem

Volume that spikes when flights go wrong. CMOtech Asia describes the pressure: “Airlines typically face sharp swings in contact volumes during delays, cancellations, and extreme weather.” PAL’s passengers reach the airline “across multiple touchpoints — from flight disruptions and rebookings to refunds, loyalty inquiries, and day-of-travel support,” in Twilio’s words.

Slow handoffs and separate channels. According to CMOtech, PAL “aimed to ease pressure created by slow handoffs between departments and separate channels for voice, chat, and online servicing.” The stated focus was “to reduce fragmentation across customer service channels during peak travel periods and disruption events.”

Legacy systems during a recovery. PAL entered Chapter 11 restructuring in the United States in 2021 and emerged later that year. Twilio’s case study frames the challenge as the need “to modernize its legacy systems and provide personalized customer experiences for millions of passengers while rebuilding from one of its most significant crises.” Over the five years before the May 2026 iTNews Asia interview, PAL moved its contact centre from on-premise infrastructure to cloud-based systems. Mark Anthony Munsayac, Vice President for Customer Experience, gave iTNews his test for when a system has to go: “If the cost of maintaining a system keeps increasing, if it delays issue resolution and affects passenger experience, then it becomes a reminder to reassess whether we should already replace the technology.”

Callers who switch languages mid-sentence. On the phone, PAL’s callers “move fluidly between English, Tagalog, and Taglish, often within a single sentence,” according to Sierra’s customer story. Sierra describes Munsayac’s goal as “customer support that felt faster and more personal at a scale human staffing alone couldn’t reach.”

Confirmed facts: about 31 domestic and 39 international destinations; 16M+ passengers in 2025; close to two million inbound calls a year; contact volume swings during delays, cancellations, and extreme weather; slow handoffs between departments; separate voice, chat, and online servicing channels; on-premise to cloud move over five years; Chapter 11 in 2021; callers mix English, Tagalog, and Taglish.


3. Solution

3.1 One platform for every channel, with the passenger’s history on the agent’s screen

“Twilio Flex now sits at the centre of its contact centre setup,” CMOtech reports. The platform is integrated with PAL’s customer systems and brings interactions from different channels together, which, in Twilio’s words, “provides agents with greater visibility into each passenger’s journey, enabling faster resolution and more context-aware support.”

PAL did not switch systems in one step. iTNews reports that the airline used “phased migration strategies, parallel system operations and rollback mechanisms,” and that every modernisation initiative included rollback planning before deployment. Munsayac: “Customer experience is a non-negotiable constraint. We are willing to delay timelines if needed rather than compromise passenger experience.”

3.2 Generative AI online, then a widening list of tasks

By February 2026, PAL had “introduced a generative AI chat experience for online assistance.” At that point, CMOtech reports, “AI currently supports about 10% of customer service tasks, including flight status checks and basic requests.”

The scope grew from there. iTNews reported in May 2026 that PAL “initially deployed generative AI for FAQ support but has gradually expanded capabilities to include booking retrieval, refund case creation, baggage information and personalised passenger assistance.” In February he had named the next layer: “expanding automation across both chat and voice, including options for involuntary and voluntary rebookings, as well as new bookings.” The aim behind it was steadier service, not only a lower average: “making this performance consistent throughout the day — not just on average — especially during flight disruptions when call volumes can spike.”

The target was set in February: “We’re aiming to reach what we call a ‘super AI agent’ state by April 2026, where around 80% of the tasks a live agent can perform can be handled by AI.”

3.3 A voice agent that replaced the IVR

In February 2026 PAL was “testing voicebots for selected call types.” According to Sierra, PAL “wanted to retire its IVR entirely rather than bolt AI onto one node of it, so the agent had to take the call from hello and finish the job: retrieving bookings, checking flight status, answering the long tail of baggage and fare questions, and more. Only out-of-scope requests reach a human, routed to the right queue with the conversation summary attached.”

PAL “evaluated three voice AI vendors in a head-to-head production bake-off, scored on customer experience, technical capability, and containment.” Before launch, the chosen agent recorded “98%+ accuracy across more than 900 FAQ test scenarios and a perfect pass rate on functional UAT.”

“Working with ibex, its long-standing BPO partner, PAL launched a Sierra voice agent on three Philippine hotlines in April 2026.” The agent:

  • understands English, Tagalog, and Taglish, with a custom Filipino voice persona tuned for PAL
  • connects to PAL’s booking and flight-status APIs
  • answers from more than 1,400 knowledge sources
  • hands off to live agents “with full conversation context”

PAL then routed all US traffic to the agent, which Sierra says added roughly 70,000 calls a month. Within three months the agent covered four hotlines instead of three. Rebooking and name-correction journeys are listed as next.

3.4 People: always reachable, and now coaching the agents

PAL kept the human route visible. Munsayac: “We’re not hiding our live agents - passengers can always choose to connect with one. Our goal is for customers to use digital and automated options because they deliver a better experience, not because they’re forced to.”

The people who run the AI are PAL’s own staff. According to Sierra, “A team of PAL customer experience specialists, the same people who had managed the airline’s chat agent for a year, now coach the voice agent directly.” Production conversation data is reviewed continuously by PAL, ibex, and Sierra.

3.5 From answering to reaching out

The same platform is being used to connect communications across departments. Munsayac: “We’re finalizing the design for an end-to-end communication platform to provide a more personalized experience for our passengers. It’s not just the customer experience team involved - we’re collaborating closely with marketing, engineering, and loyalty teams.”

Twilio gives one example of a message triggered by passenger data: “if a frequent flyer typically prepays for baggage but has not done so a few days before their flight, PAL can automatically send a reminder through the passenger’s preferred channel.” Munsayac: “Our focus is on servicing first. Revenue follows when we offer the right ancillary services to the right passenger, at the right moment, based on what we know about them.” In May he named AI-assisted quality monitoring, automated disruption notifications, and proactive support during missed connections as near-term gains. A customer data platform to unify loyalty, communications, and service operations is planned.

Confirmed facts: Twilio Flex integrated with customer systems, cross-channel view for agents; phased migration with parallel runs and rollback plans; generative AI chat live online (Feb 2026); AI supports ~10% of customer service tasks incl. flight status and basic requests (Feb 2026); scope expanded from FAQ to booking retrieval, refund case creation, baggage information (May 2026); rebooking and new-booking automation being added (Feb 2026); target ~80% of live-agent tasks by April 2026; voicebots in testing (Feb 2026); Sierra voice agent with ibex on three hotlines (April 2026), answering calls from the start where PAL “wanted to retire its IVR entirely”, booking and flight-status APIs, 1,400+ knowledge sources, handoff with summary; US traffic added (~70,000 calls/month), four hotlines within three months; live agents always selectable; CX specialists who ran the chat agent coach the voice agent; end-to-end communication platform in design; customer data platform planned.


4. Impact

MetricFigureBasis
Average contact centre wait timeUnder one minute“Since implementing the platform”; internal performance tracking (Feb 2026)
Customer satisfactionAround 95%“In recent measurements” (Feb 2026)
Deflection rateAbout 45%Concerns resolved without transfer to a live agent (Feb 2026)
AI share of customer service tasksAbout 10%Tasks AI supports, incl. flight status and basic requests (Feb 2026)
Customer service costAbout 30% monthly savingsCustomer service operations (Feb 2026)
AI targetAbout 80% of tasks a live agent can performTarget for April 2026
Voice agent customer surveys4.7 out of 5Since April 2026 launch, “on par with its best traditional human agent queues” (Sierra, Oct 2026)
Voice booking retrieval success30% → 75%After simplifying a verification step (Sierra)
Mis-routed calls, loyalty transfer flow15.5% → 7.2%After rewriting one question (Sierra)

PAL ties the first results to the platform and to self-service together. Munsayac: “Shorter handle times and expanded self-service options have helped drive operational efficiency and cost savings.” On satisfaction: “Together, these improvements have contributed to a meaningful increase in customer satisfaction, which in recent measurements has reached around 95%, placing us among the stronger performers in the industry.”

The two voice-agent fixes came from call data. Sierra writes that the team “found nearly a third of callers dropping at a single verification step, simplified the flow, and lifted retrieval success from 30% to 75%.” In the loyalty transfer flow, “callers were confusing two membership tiers,” and rewriting one question cut mis-routed calls “from 15.5% to 7.2%.” Sierra: “Neither fix came off a roadmap. Both surfaced in production conversation data.”


5. Insight

The platform came before the AI. The sub-minute wait was reported “since implementing the platform,” at a time when AI supported about a tenth of service tasks. Putting every channel and the passenger’s history on one agent screen gave PAL what Twilio calls “a strong foundation to roll out additional automation and AI-driven services.”

Three numbers, three bases. About 10% is the share of service tasks AI supports. About 45% is the share of concerns that end without transfer to a live agent. About 80% is a target share of the tasks a live agent can perform. Each counts something different, so they are not points on one scale.

Lookups first, changes later. PAL’s generative AI started with FAQs, then added booking retrieval, refund case creation, and baggage information. The voice agent launched on bookings, flight status, and baggage and fare questions, with rebooking and name correction next. Read-only tasks went first, and tasks that change a booking came later. iTNews separately reports that every modernisation initiative at PAL included rollback planning before deployment.

Replacing the menu, not adding a branch. Sierra says PAL wanted to retire its IVR entirely rather than put AI on one node of it, so on the hotlines it runs, the agent takes each call from the start and sends only out-of-scope requests to people. That choice makes the AI the front door for the whole hotline. Before launch, PAL ran a three-vendor bake-off scored partly on containment and tested the chosen agent on more than 900 FAQ scenarios.

The improvement loop runs on production data, and PAL’s people run it. The two voice-agent gains Sierra reports after launch came from reading where real callers dropped or were misrouted. The specialists who had managed the chat agent for a year now coach the voice agent. In this case, AI work in the contact centre includes a standing job: reviewing conversations and changing the agent’s flows and knowledge.

Limits. All figures are company-reported through vendors (Twilio, Sierra) or trade press based on the Twilio case study. Wait time and satisfaction have no published baseline, and the 95% satisfaction figure (all service) and the 4.7/5 survey score (voice agent) are different measures. The 30% saving is for customer service operations as a whole, not for AI alone. Whether the 80% target was reached by April 2026 is not reported. No headcount, agent numbers, or live-agent role changes are published beyond the specialist team that coaches the AI.

One-line takeaway: PAL put every channel and the passenger’s history on one platform first, then let AI take the front door of chat and the hotlines one task type at a time, with a person always one request away and its own specialists coaching the AI from real conversations.


6. Implementation hypotheses (not in the source)

6.1 A handoff record between the AI front door and the agent desk

Basis: Flex “consolidates passenger interactions across channels”; the voice agent sends out-of-scope requests to a human “routed to the right queue with the conversation summary attached”; deflection is counted as concerns “resolved without being transferred to a live agent.”

The likely shape is an AI layer in front of the Flex queues for both chat and voice. When the AI cannot finish a request, or the passenger asks for a person, it would create a transfer into Flex carrying a small record: detected intent, language, booking reference if retrieved, whether the caller passed verification, and a short summary. Flex routing would read the intent to pick the queue (loyalty, refunds, disruption), which is why a confusing question in the loyalty flow showed up as a measurable mis-route rate. The deflection rate would then be computed as conversations that end with no transfer record. The promise that a passenger can “always choose” a person is most simply kept as an intent that transfers immediately, without asking the passenger to try the bot again first.

6.2 Read-only tools first, transactions behind rollback switches

Basis: Chat scope grew from FAQ to “booking retrieval, refund case creation, baggage information”; rebookings and new bookings were being added; the voice agent connects to “booking and flight-status APIs” with rebooking and name correction next; every initiative had “rollback planning before deployment.”

Each task type the AI handles is probably a tool call with its own permission level. Flight status and booking retrieval only read data. Refund case creation writes, but it likely opens a case for staff to process rather than issuing money, which keeps a person on the decision. Rebooking and name correction change the ticket, so they would come last. Involuntary rebooking, where the airline caused the change, is probably simpler to automate because no fare difference is charged. Voluntary changes involve fare rules and payment and would more likely hand off at the payment step. Each new task type would be switched on per channel or per hotline, with the rollback being a routing change that sends that intent back to human queues. The “80% of tasks a live agent can perform” target implies a catalog of agent task types against which each switch is counted.

6.3 A review loop from call funnels to flow edits

Basis: Sierra reports a verification step where “nearly a third of callers” dropped and a loyalty question that caused mis-routes, both found in “production conversation data, which PAL, ibex, and Sierra review continuously”; PAL specialists “coach the voice agent directly”; the pre-launch test set had “more than 900 FAQ test scenarios.”

To find a single step where a third of callers drop, the team would need step-level counts for each journey: calls started, verified, booking found, resolved, transferred. Mis-routes would be counted when a receiving queue sends a transfer elsewhere. The coaching team’s work would then be editing flow wording, verification rules, and the 1,400+ knowledge sources, and rerunning the FAQ scenario set before each change goes live so that a fix in one flow does not break answers in another.


7. Sources

  1. CMOtech Asia — Philippine Airlines cuts waits with AI contact centre (Sofiah Nichole Salivio)
    https://cmotech.asia/story/philippine-airlines-cuts-waits-with-ai-contact-centre
    2026-02-25 (accessed 2026-11-17)
  2. Twilio — Philippine Airlines earns passenger trust by transforming customer support with Twilio Flex
    https://customers.twilio.com/en-us/philippine-airlines
    Publication date: not shown on page (accessed 2026-11-17)
  3. iTNews Asia — Lessons from Philippine Airlines’ digital transformation journey (Abbinaya Kuzhanthaivel)
    https://www.itnews.asia/news/lessons-from-philippine-airlines-digital-transformation-journey-626023
    2026-05-20 (accessed 2026-11-17)
  4. Sierra — Philippine Airlines’ AI voice agent answers in English, Tagalog, and Taglish — the way its callers actually talk
    https://sierra.ai/customers/philippine-airlines
    2026-10-07 (accessed 2026-11-17)