Puro tayo AI tools pero lugi pa rin? Strategist Jerry Ilao says Filipino workers may be leaving productivity gains on the table

BY EDISON JOSEPH GONZALES

If employees are already using artificial intelligence (AI) several times a week, the next productivity question may be less about getting them to use it and more about teaching them what they are missing.

At a company workshop led by Filipino AI business strategist Jerry Ilao, founding president and co-founder of the Philippine AI Business Association (Paiba), participants reported turning hours-long reporting and analysis tasks into work that took minutes using commonly available AI assistants.

In one workshop, 87 percent of participants said they were already using AI multiple times a week – the highest rate Ilao had seen across his workshops. Even so, Ilao said many participants discovered significant room to improve how they used AI assistants.

An inventory aging analysis that previously took four hours was redone in three minutes. A sales and collection report went from five hours to five minutes, while a network sales report was completed in 10 minutes instead of six hours.

The examples were participant-reported, the AI-assisted times included review and corrections, and results vary by task.

“None of these people built an agent or automated anything,” Ilao said. “They took a report they already knew well, gave the same data to ChatGPT, Claude, Gemini, or Copilot, and compared the result with the version they built by hand. The benchmark is what made it meaningful.”

Ilao’s 4A Blueprint, a practical AI adoption framework for Philippine businesses, maps AI use across four levels: Assistants, where people use AI to do their own work faster; Automation, where AI runs a set workflow on its own; Agents, where AI takes on a defined job within limits people set; and AI-First, where the business is built around AI.

Ilao’s principle is to meet the business where it is and start there.

Ilao said companies should not feel compelled to jump straight to AI agents. He advises companies to start at Level 1, which he describes as the cheapest and easiest level to implement.

“At Level 1, companies can start by training employees to use AI effectively, giving them access to approved tools, and putting a clear AI policy in place,” Ilao said. “That is one of the fastest and lowest-cost ways to start improving productivity. Once the company is ready, it can move to automation and AI agents. But it does not have to start there.”

As companies move from assistants to automation and agents, implementation complexity, governance requirements and operational risk generally increase.

In the framework, a higher level is not automatically better. A business may be exactly where it should be at Level 2 if automation already delivers the needed business value without the added complexity of AI agents.

“Level 1 gets underestimated because it sounds basic,” Ilao said. “But the judgment your people build there is what every level above runs on.”

Across selected workshop sessions, participants’ average self-reported confidence in using AI at work moved from 3.22 to 4.26 out of 5, based on matched before-and-after responses.

Eighty-nine percent rated their AI-assisted output better than their usual work, and 94 percent said they would use the approach again.

These are self-reports, not measured productivity gains.

Ilao recommends four steps: pick one analysis the team has already completed manually; give AI the same data and question; include review time when measuring the result; and compare the AI-assisted output with the original.

Employees should use company-approved tools and must not upload confidential, sensitive or personal information unless their organization’s policies and data protections allow it.

If the approach works, the company can then turn the prompt into a shared assistant on approved company accounts, with clear rules on what employees must not upload.

The best prompts, Ilao adds, should be shared within the company.

“A good prompt is not disposable,” he said. “Treat it as intellectual property, like an SOP. Later, the same logic can become a reusable AI skill or form part of an automated workflow or agent.”