Before we talk about AI, let's talk about the commute
I've spent a significant amount of my career in Philippine contact centres. Not visiting them. Working in them, managing them, walking the floor at 2am during a night shift in Manila. And the thing that stays with me is not the technology or the metrics or the client reviews. It's the people.
Most of the agents I've worked with start their day on a trike. Then a jeepney. Then, if they're headed somewhere like BGC or Ortigas, another connection to get to the mall where most of Manila's contact centres now operate. The malls make sense operationally: the transport routes already exist, the infrastructure is reliable, there's food and security and somewhere to decompress between shifts. But the commute to get there can easily run two hours each way, through heat and traffic that would break most people before they'd even logged in.
They arrive, they sit down, and they take calls. Strict metrics. Adherence to the minute. Quality scorecards that leave almost no room for judgment. And they do it, shift after shift, with a warmth and professionalism that remains one of the most remarkable things in the contact centre industry globally.
The Philippine BPO sector employs over 1.7 million people and generates revenues the industry body IBPAP projects will reach USD 59 billion by 2028. It is one of the most significant industries in the country's economy. And it is sitting on something that barely anyone in the AI conversation is talking about.
Philippine contact centre agents are the most knowledgeable people in your AI strategy. They just have no formal way to tell you that.
What the floor actually knows
Every day, across tens of thousands of seats in Manila, Cebu, Clark, and Davao, Filipino agents are absorbing something that no weekly business review, no NPS dashboard, and no detractor analysis slide can replicate. They are absorbing real customer conversations, at volume, in real time, across every product, every process failure, every broken journey, every moment where the company's stated customer experience and the customer's actual experience diverge.
They know which products generate the most confusion at the point of purchase, because they hear it fifty times a shift. They know which refund processes break down consistently, because they are the people building the workarounds in real time. They know which automated flows are sending customers in circles before they arrive frustrated on a live call, because they hear about it in the first thirty seconds of every one of those calls.
They know more about the upstream causes of customer problems than the CX manager in Sydney reviewing the weekly slide, more than the product team who shipped the feature that's generating the complaints, and often more than the operations leadership who built the process that keeps generating the friction.
This is not a small thing. This is the most valuable intelligence available for AI use case identification. And most organisations are not capturing it deliberately, rewarding it commercially, or acting on it systematically.
The problem with binary processes
To understand why that intelligence goes untapped, you need to understand what the Philippine BPO model was built to do.
The industry scaled fast. The large BPOs grew by standardising everything: binary processes, prescriptive scripts, quality scorecards that measure adherence rather than outcomes, and attendance policies that would look out of place in most other professional environments. This was not malicious. It was the most efficient way to deliver consistency at scale across hundreds of thousands of agents and dozens of clients.
The problem is what it does to the people inside it.
Filipino agents are, by cultural instinct, creative, warm, and naturally inclined to find solutions. The contact centre environment, particularly in the large BPO model, systematically trains that out of them. Initiative is a compliance risk. Bespoke problem-solving deviates from the script. A quality scorecard that penalises any response not found in the approved knowledge base will, over time, produce agents who have stopped trusting their own judgment entirely.
There is a phrase I come back to constantly in this industry: tell me how I'm measured and I'll tell you how I'll behave. It is a self-fulfilling prophecy. If you measure agents on script adherence, you get script adherence. Not problem-solving. Not empathy. Not the contextual judgment that makes a difficult interaction land well. You get exactly what you measured for, and nothing more.
Quality scorecards built around script adherence don't measure customer outcomes. They measure agent compliance. Those are not the same thing. In fact, they are frequently in direct conflict. The agent who goes slightly off-script to actually solve a customer's problem will often score lower than the agent who followed the process and left the customer no better off.
There is also the question that still hangs over some interactions, particularly those servicing Western markets. The "which country am I speaking to" dynamic. The honest answer is that most customers have moved past this. What they care about is whether their problem gets solved, not where the person solving it is sitting. The issue was never really geography. It was resolution. And an agent who is empowered to actually solve a problem, rather than read from a script, resolves that concern faster than any amount of process compliance ever will.
Then AI arrives. And everything gets worse.
Introduce AI into this environment without thinking carefully about what it does to the people inside it, and you do not get transformation. You get a morale crisis.
Here is what typically happens. The AI is deployed to handle tier-one contacts: the simple, high-volume, repetitive interactions that are easy to automate and make the containment metrics look good. And it works, up to a point. The bot handles the balance inquiries, the order status checks, the basic FAQs.
What's left for the agents are the escalations. The complex cases. The customers who have already tried the bot and failed. The customers who are angry. The customers with edge cases the AI cannot handle. The interactions that require exactly the judgment, empathy, and contextual problem-solving that the binary process environment has spent years training out of the agent workforce.
The average difficulty of every single interaction goes up. The volume does not go down proportionally. The metrics do not change to reflect the different nature of what agents are now doing. And the agent who has spent their career being told to follow the script is now expected to exercise sophisticated judgment on the hardest conversations in the queue, often without adequate support.
AI left the hardest work for the humans. Then judged them on metrics designed for the easiest work.
Agent-assist AI compounds this. At its best, it surfaces relevant information, suggests responses, and reduces cognitive load. At its worst, it feels like surveillance: another system monitoring every word, flagging deviations, adding another layer of automated oversight to an already over-monitored environment. For agents who have been treated as execution resources rather than thinking professionals, the arrival of AI that watches what they type does not feel like support. It feels like the next stage of a process designed to replace them.
The morale impact is real. Attrition in Philippine contact centres is already a chronic challenge. Deploying AI in a way that increases workload difficulty, reduces autonomy further, and signals to the workforce that their days are numbered is a reliable way to make that worse.
What the industry is getting backwards
Most of the conversation about AI and Philippine BPOs focuses on the threat: how many roles will automation eliminate, what is the timeline, how do BPOs maintain revenue as tier-one volume decreases. These are real questions. They are also the wrong place to start.
The right place to start is this: what does the BPO actually know, and how do we use it?
I have sat in too many client reviews where a CX manager from Sydney or London, who has never spent a shift on a contact centre floor in their life, is asking a BPO account manager to justify a detractor score. The account manager produces a root cause analysis. The quality team has spent forty hours going back through call recordings to categorise complaints into themes that fit the client's reporting framework. The training team has been pulled into workshops about why scores went down this month.
All of that is retrospective. All of it is designed to satisfy a client narrative rather than solve a problem. And none of it captures the thing the BPO already knows from three months of taking those calls: the problem is upstream, it is in the product or the process or the policy, and it is generating the same friction over and over again regardless of how good the agents are.
Root cause analysis on detractors, as typically practised, answers the question: "Why did this specific group of customers score us poorly this period?" It almost never answers the more important question: "What is systematically wrong upstream that keeps generating the same problems?" The first question produces a PowerPoint slide. The second produces an AI use case.
The BPO is sitting on the answer to the second question. They know it from the calls. What they are rarely given is the permission, the commercial incentive, or the organisational channel to surface it in a way that gets acted on.
BPOs as AI advisors: what this actually looks like
The BPOs that will thrive in an AI-intensive contact centre market are not the ones that automate fastest. They are the ones that reposition themselves as strategic advisors on AI use cases, drawing on the frontline intelligence that no client organisation can generate internally.
This is not a radical idea. It is a natural extension of what the best BPO account managers are already doing informally. The shift is making it formal, commercial, and systematic.
In practice, it means several things.
First, capturing frontline intelligence deliberately. Not through post-hoc root cause analysis, but through structured processes that surface agent knowledge in real time: regular structured input from team leaders, pattern recognition across interaction data, and formal channels for agents to flag upstream problems they are seeing at volume. The agents already know. The system just has to ask.
Second, using that intelligence to lead AI use case identification. The BPO, not the client's CX manager reviewing a dashboard, is best placed to identify which interaction types are genuinely automatable, which require human judgment, and which represent problems that should be fixed upstream rather than handled better downstream. That advisory capability is worth paying for, and clients who understand this are getting better AI programmes than those who treat the BPO as an execution vendor.
Third, changing how agents are recruited, trained, and measured as AI takes over tier-one volume. The agents left handling escalations need different skills and different support than those who spent their careers on scripted tier-one. Metrics need to reflect the actual difficulty of what they are doing. Scorecards need to reward judgment and empathy alongside adherence. The binary process model that scaled the industry was fit for purpose in one context. It is not fit for purpose in the AI-augmented contact centre.
And fourth, treating the workforce transition with the respect the workforce deserves. These are people who commute two hours to do a difficult job with professionalism and warmth. They are not a cost line to be managed down as AI scales up. They are the domain experts whose knowledge is the most valuable input in the AI programme. Treating them that way, in how they are consulted, compensated, and involved, is both ethically right and commercially rational.
The organisations getting this right
A small number of BPOs are already making this shift. They are investing in agent upskilling for AI collaboration roles, building proprietary AI and CCaaS capabilities of their own, and repositioning their value proposition from "we deliver consistent execution" to "we bring CX intelligence you cannot generate anywhere else."
The industry body IBPAP projects that 80% of Philippine BPO firms will invest in worker upskilling for AI roles by 2030. That is an encouraging number. The risk is that upskilling becomes a cost-minimisation exercise rather than a genuine repositioning: teach agents to use AI tools, keep the binary process structure, and continue measuring them the same way. That does not produce a more capable workforce. It produces a more automated version of the same dynamic.
The clients who will benefit most from the intelligence the Philippine BPO sector holds are the ones who stop treating the BPO relationship as a procurement exercise and start treating it as a partnership. That means creating commercial incentives for BPOs to surface upstream problems. It means stopping the retrospective root cause analysis theatre and replacing it with forward-looking use case development. It means giving account managers the latitude to tell their clients uncomfortable truths about what the calls are actually revealing, rather than spending those hours building slides that confirm what the client already believes.
The intelligence gap between what the BPO knows and what the client acts on is where the best AI use cases are buried. Closing that gap is the highest-value thing either party can do right now.
The Philippines BPO industry is not under threat from AI. It is sitting on the most valuable raw material for AI strategy in the market. The industry that figures out how to surface, structure, and sell that intelligence will not just survive the AI transition. It will lead it.