Sooner or later a vendor will show you a number like this one. A risk review that took two to ten days now takes five minutes to one day, with 97 to 99 per cent less manual effort.
That claim is real, and it comes from a real company. It is also the standard shape of a vendor case study. Learn to read it, because you will be handed one within the month.
What was reported
PLDT is a large telecoms group in the Philippines. In September 2026 it described three internal AI assistants, all built on UiPath software. The announcement came from UiPath and was published by the technology news site TNGlobal.
The first, called KAI, is a knowledge assistant used for proposal writing and customer support research. PLDT estimates it saves 25,000 to 30,000 hours a year and answers 99 per cent faster.
The second, Ellie, is a workflow assistant for enterprise teams. PLDT estimates 18,000 to 25,000 hours a year saved, 25 to 35 per cent less manual effort and 80 per cent faster customer responses.
The third, ERICA, scores and assesses the annual risk review. That is where the headline comes from: two to ten days down to five minutes to one day.
What the report says about itself
TNGlobal deserves credit here. It states that the figures are “company estimates rather than independently verified performance data”. That is unusual for coverage of a vendor announcement.
The announcement gives no measurement period, no audit and no baseline. We do not know how the old process was timed, how many reviews were counted, or whether the quick version covers the same work as the slow one. A 99 per cent figure with nothing behind it is a claim, not a result.
None of this means PLDT is wrong. It means you cannot yet tell.
Three questions for any figure like this
Over what period? A saving measured over one good week is not an annual saving. Ask how long the measurement ran and when it ended.
Against what baseline? “Two to ten days” is a wide range, and the best case of the old method may be a slow case of the new one. Ask what the old process was timed on, and who timed it.
Checked by whom? An estimate made by the team that built the system differs from a figure the finance team signed off. Ask who verified it and how.
These questions cost nothing and take a minute. A vendor with a solid result will answer them quickly. A vendor without one will move to the next slide.
Why the range matters most
Look again at the headline: “two to ten days” before, “five minutes to one day” after. Both ends are ranges, and they overlap in an awkward way. A review that took two days under the old method and one day under the new one is a 50 per cent saving, not a 99 per cent one.
The 97 to 99 per cent figure only holds if the typical case sits near the best end of both ranges. That may be exactly what happened. Without the count of reviews and the timing method, you cannot tell whether it is typical.
This is the habit to build. When a number arrives as a range, ask where in the range most of the cases fell. The average case is the one that decides whether the project pays for itself.
The part worth copying
There is a second lesson in the story, and it holds even if every figure is rounded up.
Look at what PLDT chose to work on. Proposal writing, customer support research and the annual risk review are not technical systems. They are administrative work that every department has, and all three are slow, repetitive and already complained about.
Nobody needed to be persuaded that those tasks were a problem. Choosing work like that gives you an easy comparison, a willing team and a result you can check.
What to do this month
Pick a task people already complain about. Ask your team which job they dread because it is slow and repetitive. That is your first candidate.
Time it now. Write down how long it takes today, on how many cases, over what period. You are building your own baseline before anyone shows you theirs.
Time it again after. Use the same task, the same count and the same person doing the timing. Keep the two sets of numbers side by side.
Ask the three questions of every case study you are shown. Period, baseline, checker. Write the answers down, including when there are none.
If PLDT’s numbers hold up, they are impressive. If they do not, the asking costs you a minute. Either way you will know more than the slide told you.
Source: TNGlobal, 10 September 2026, reporting a UiPath announcement. Figures are PLDT’s own estimates and have not been independently verified.