GitHub's 2024 Octoverse put the Philippines at #18 in the world, past Australia at #19. The rank isn't the interesting part. The slope is: 29% growth year over year on a base above 1.7 million accounts, the fastest in GitHub's Asia-Pacific table and ahead of India's 28% [TL22]. A country most sourcing shortlists treated as a call-centre destination is now adding developers faster than the country people mean when they say "offshore".
This is an evidence review of one delivery destination, written by a firm that delivers from it. Every figure below carries a marker back to a source we fetched and read, and that includes the ones that make our own case harder to argue — the corrected talent-pool number, the attrition rate, the IP ranking, the twelve-hour gap to New York. Where the evidence runs out, this piece says so rather than filling the hole with something that sounds right. There is a lot of Philippine sourcing content that doesn't do that, and a technical buyer who checks one number and finds it inflated is right to discard the rest.
The growth figure is real and the way it usually gets quoted is wrong. Both things are worth your time.
GitHub's own 2024 data has the Philippines at #18 globally, overtaking Australia, with 29% year-over-year growth on a base above 1.7 million — the fastest rate in its Asia-Pacific table, ahead of India at 28% [TL22]. That's a platform vendor reporting on its own users, which makes it unusually hard to spin. Two limits ride along with it: the Philippines doesn't appear in the 2025 Octoverse edition at all [TL24], so this is a snapshot rather than a trend line you can extend, and GitHub counts accounts — students, hobbyists, dormant profiles, duplicates — not employed developers. The platform reports more than 180 million of them worldwide [P2]. Nobody's labour pool looks like that.
For scale, the most credible global count of working developers is SlashData's 48.4 million, built from a survey run 29 times with over 10,000 developers per wave plus activity data, and explicitly not from pageviews, downloads or IP addresses [P1]. Platform accounts and developers are different units, and the gap between 180 million and 48.4 million is roughly how different.
Which brings up the number you'll meet everywhere else, and the correction we'd rather make ourselves.
The figure quoted in almost every Philippine sourcing pitch is 1.9 million. It describes the whole IT-BPM sector, which ended 2025 with export revenues above $40 billion — and that sector is overwhelmingly contact-centre and business-process work, roughly 1.3 to 1.4 million people, against about 0.17 million full-time equivalents in IT services — both 2022 estimates from the industry roadmap [R26]. Philippine formal computer-programming, consultancy and related employment is 90,573 workers [P7]. Allow for contractors, in-house developers sitting inside banks and telcos, and the informal end of the market, and the defensible bracket for working software developers is roughly 90,000 to 200,000.
That's an order of magnitude below the headline. It is also, on any reasonable reading, deep enough to staff from — 90,000 developers is a larger pool than most buyers will ever need, and the growth curve says it is getting deeper. What it changes is how you shop. A pool that size means the Philippine vendor market rewards reading vendor by vendor rather than assuming uniform depth, and any supplier still quoting you 1.9 million developers has either not checked or is counting on you not to.
The strongest argument for Philippine delivery is one almost nobody makes, because the whole category is written for American buyers.
Philippine Standard Time is UTC+8 with no daylight saving [TZ4], corroborated by a time-zone service returning a standard offset of 28,800 seconds and no DST interval for Asia/Manila [TZ5]. One sourcing note, since we checked: that rests on the IANA time-zone database rather than a Philippine authority. The Philippine Standard Time Act of 2013 never states a UTC offset [TZ2], and PAGASA, the agency the Act charges with keeping the time, doesn't publish one in retrievable text [TZ3]. The offset isn't in doubt. Its provenance is a database, not a statute, and you should know that before you quote it in a contract.
Set a 9-to-6 working day at both ends and intersect. Everything below is arithmetic on verified offsets [TZ10].
| Manila and… | Working-day overlap | In local terms |
|---|---|---|
| Singapore (UTC+8) | 9 hours | identical clocks |
| Perth (UTC+8) | 9 hours, year-round | identical clocks, no DST either side |
| Hong Kong (UTC+8) | 9 hours | identical clocks |
| Tokyo / Seoul (UTC+9) | 8 hours | Manila 09:00–17:00 / Tokyo 10:00–18:00 |
| Brisbane (UTC+10) | 7 hours, year-round | Manila 09:00–16:00 / Brisbane 11:00–18:00 |
| Sydney, AEST (UTC+10) | 7 hours | Manila 09:00–16:00 / Sydney 11:00–18:00 |
| Sydney, AEDT (UTC+11) | 6 hours | Manila 09:00–15:00 / Sydney 12:00–18:00 |
| Mumbai (UTC+5:30) | 6 hours 30 minutes | Manila 11:30–18:00 / Mumbai 09:00–15:30 |
| Auckland, NZST (UTC+12) | 5 hours | Manila 09:00–14:00 / Auckland 13:00–18:00 |
| Auckland, NZDT (UTC+13) | 4 hours | Manila 09:00–13:00 / Auckland 14:00–18:00 |
A Singapore, Perth or Hong Kong buyer working with Manila gets a full working day of overlap on identical clocks with no daylight saving on either side to reset twice a year. Tokyo and Seoul give up an hour. Brisbane gives up two and never moves. Sydney's number swings with Australian daylight saving — seven hours in AEST, six in AEDT [TZ7] — and the New South Wales government publishes the switch dates, 4 October 2026 and 4 April 2027 [TZ8], so it's a diary item rather than a surprise.
Now the limit, and it's a real one. Nobody publishes how much Asia-Pacific software work actually goes to the Philippines. IBPAP, the national industry body, publishes no client-geography breakdown [TZ11]; the central bank's balance-of-payments statistics were unreachable and the Philippine Software Industry Association publishes no statistics at all [TZ14]. One adjacent government figure runs the other way: the Philippines doesn't appear among Australia's top-10 services-import partners in the Australian Bureau of Statistics' latest release [TZ13]. So what's on offer here is time-zone geometry, which is checkable to the minute. It is not trade volume, and we're not going to imply otherwise.
For a US buyer none of the above applies, and the honest opening is the arithmetic: Manila runs 12 to 13 hours from US Eastern depending on daylight saving, which is close to perfect antipodal separation and zero natural overlap of standard business hours [T4]. That's the constraint. What the research says about it is more interesting than the constraint itself.
The paper that gets quoted at offshore teams is Herbsleb and Mockus. Its headline is that distributed work items took about two and a half times as long as comparable colocated ones on the active work interval [A4]. That number is real, and stopping there is the mistake.
Control for the other factors, including how many people a change had to involve, and distributed items are no longer significantly slower. The authors' own explanation is that splitting work across sites slows it down "primarily because it requires the involvement of more people than comparable work accomplished all at one site" [A6]. The delay data points the same way: cross-site delays occurred at roughly the frequency of local ones, but lasted 2.4 days against 0.9 [A7]. Not more blockages. Longer ones, with more people and more waiting packed into each. Date it properly — Lucent change requests from July 1997 to July 1999 [A5], before Git, before CI, before Slack. Old, and still the mechanism you can actually act on, because coordination headcount is a design choice and the map isn't.
That reframing is what makes an overnight cycle worth discussing, and it's worth being precise about which version of it survives contact with the literature. Continuous follow-the-sun — work handed off every evening and returned finished every morning — does not. Carmel's foundational paper concedes in its own abstract that the practice "has enjoyed few documented industry successes because it is acknowledged to be extremely difficult to implement" [FS1], and the most recent systematic mapping study, 57 papers spanning 1990 to 2017, still names the field's open need as evaluation research for testing feasibility and effectiveness in practice [FS4]. Three decades of attempts, no established result. We're not claiming to be the exception.
The narrow version is one-directional and does hold up: you specify, we build overnight, you review in your morning. One handoff per cycle rather than a nightly round trip. The evidence for that shape comes from asynchronous code review, and it's strong. Google's review process is fully async, with a median end-to-end latency under four hours across 9 million reviewed changes [AS18] and 70% of changes committed within 24 hours of going out for initial review [AS19]. The paper is explicit that asynchrony isn't the cause: the median change modifies 24 lines and the median reviewer count is one [AS21]. Across 291 open-source projects and 166,884 pull requests, 30% merge in under an hour and 80% within 3.7 days, with turnaround governed by project process rather than by the change itself [AS24, AS25]. The same Google study carries the finding that cuts against us — interviewees named geographical distance between author and reviewers as a cause of delay and misunderstanding [AS23]. Batch size is the lever. Distance shows up in the residual.
Which means the prescription is a design one, and Herbsleb and Grinter wrote it in 1999: assign work across sites "according to the greatest possible architectural separation in a design that is as modular as possible", and split only well-understood work where plans, processes and interfaces are established and likely to be stable, because instability multiplies the need for communication [EM23].
Two numbers you'll be shown, both of which fail on contact. The claim that pull request turnaround rises 50% for every six hours of time-zone difference is attributed to GitHub's 2023 State of the Octoverse. We fetched that report and queried it: it contains no finding on pull request turnaround and no time-zone correlation of any kind [F1]. A second claim, that teams with four or more hours of overlap resolve integration issues 60% faster, is attributed to Communications of the ACM, 2022; we could not locate any such article. Both trace back to a single outsourcing vendor's marketing page. Neither belongs in a decision, including when it points the way you already wanted to go — and note that Herbsleb and Mockus never measured time zones either. Their variable is site [EM8].
Language shows up in code review comments, in a specification someone misread at 2am, and in the gap between what a ticket said and what got built. Two independent test owners measure it, and they agree on the part that matters to a buyer choosing between delivery regions.
On EF's 2025 English Proficiency Index the Philippines ranks #28 globally with a score of 569 [E5]. Set that against the markets a cost-driven buyer is usually weighing it against: Brazil at #75 with 482 [E9], Colombia at #76 with 480 [E10], Mexico at #103 with 440 [E11]. Those gaps aren't marginal.
The second measure comes from the organisation that owns the test rather than a proficiency index. IELTS Academic results for 2024–25 put the Philippine mean overall band at 6.81, ahead of India at 6.58, Mexico at 6.54 and Colombia at 6.53 [EN5]. Underneath both sits a structural reason rather than an accident of sampling: English is a statutory medium of instruction under the Enhanced Basic Education Act of 2013, which requires Filipino and English to be gradually introduced as languages of instruction until they become the primary languages at secondary level [EN13].
It does not lead on every instrument — Argentina edges it on EF at #26 with 575 [E4], and the ranking reverses outright on IELTS General Training, a migration-track test with a different candidate mix [EN6]. That's the honest shape of it, and it's why the marketing version of this section is worth distrusting wherever you read it. Consistently strong across measures is the claim that survives a reader clicking through to the source. Best is not, and anybody selling you "best" hasn't opened the second dataset.
The part of Philippine delivery that never shows up in a sourcing scorecard is the one that matters the first time a contract is argued over rather than signed.
Philippine law is a blend of civil law, Anglo-American common law, Muslim law and indigenous law — described that way by the Director of Library Services for the Supreme Court of the Philippines [PH6]. That isn't a decorative fact. Philippine law libraries carry United States court reports and West's national reporter system as standard holdings, alongside American Jurisprudence and Corpus Juris Secundum, with some subscribing to Westlaw or LexisNexis [PH7]. Philippine Supreme Court decisions were once appealable to the US Supreme Court [PH8]. For a US counterparty, the concepts, the citation habits and the reference shelf are the ones your own counsel already works from.
In practice it means your counsel's drafting instincts transfer. The vocabulary of consideration, warranties, assignment and remedies reads the same way on both ends of the contract, which removes a category of misunderstanding that a purely civil-law jurisdiction will hand you for free.
One number goes the other way and should shape your paperwork rather than your shortlist. On the US Chamber's 2026 International IP Index the Philippines ranks 36th of 55 at 40.64% [K8, K9] — 9th of 15 within Asia and below the Asia average [PH3]. Legal familiarity is not the same thing as enforcement strength. Write the IP assignment explicitly, name the governing law, specify where disputes are heard, and don't rely on any default.
Before any figure: no government, academic or independent-analyst source publishes a methodology-backed bill rate — what a client actually pays per hour — for any region. The only obtainable source is an outsourcing brokerage's survey of its own partners, which never states whether its numbers are client rates or developer pay [R27]. Every "average Philippine developer rate" you've been shown is either a salary wearing a price tag or a vendor's own quote presented as market data.
So here's the one comparison that is genuinely like for like, because both sides are a monthly outlay in the same currency.
The US median annual wage for software developers is $135,980 [R1]. That's the salary, not the cost. The recurring fully-loaded figure — base plus payroll taxes, benefits, equipment, software and workspace — runs 1.25 to 1.4 times base salary, with benefits alone accounting for roughly a third of total compensation [C1]. Applied to the median, that is about $14,200 to $15,900 a month, for every month the seat stays occupied. Dev On Demand is $3,495 a month, which is roughly a quarter of that.
Three conditions keep it honest. The $3,495 is our price, not a regional market rate for Philippine engineering — nobody publishes one of those [R27]. The 1.25–1.4× multiplier is a general employer figure rather than a developer-specific one [C1]. And it counts only the recurring load: recruiting spend and the ramp months a new hire is paid through before full output arrives are both excluded, which makes this the conservative version of the comparison rather than the flattering one.
You'll notice there's no Philippine peso salary anywhere in this section, and no exchange rate. That's deliberate. A currency conversion needs a stated rate on a stated date to mean anything, and a silent one always flatters somebody. Comparing what a US seat costs an employer against what a subscription costs a buyer needs no conversion at all.
If coordination headcount is what costs you time, a delivery model either reduces it or it's decoration. Four mechanisms have evidence behind them, and none are proprietary.
Small batches, first. DORA's guidance is specific: any batch of code taking longer than a week to complete and check is too big, work sliced into hours-sized batches can typically be tested and deployed in under an hour, and developers should check multiple small releasable changes into trunk at least once a day [EM10]. That's the Google review finding restated as a working practice. Second, team autonomy — teams completing work without fine-grained coordination with people outside the team, and making large-scale design changes without another team's approval [EM11]. Every external approver is another person on the path from ticket to production.
Third, stable teams rather than rotating staff. Across more than 1,000 developer pairs handing over code ownership, offshoring succession roughly halved the productivity ratio between the person taking over and the person handing off, with a 95% confidence interval of [0.42, 0.67] [EM14]. A vendor who swaps people every quarter pays that cost repeatedly and bills you for it. Fourth, written specification, which is where the overnight cycle lives or dies. Across 228 companies in 10 countries, incomplete or hidden requirements were the most-reported requirements problem, at 109 companies or 48% [EM16], and roughly 40 to 50% of project effort goes on avoidable rework, with hastily specified requirements named as a major source [EM18].
Our own model rests on those four, in the site's own words. Dev On Demand "runs async by design" and "timezone alignment isn't required". The rhythm is a "3-day task cycle, daily async updates"; you "Submit a task, matched same-day", with a "5-day first ship". Our engineering is "distributed, headquartered in Manila".
Two hours of the gap are negotiable, and the effect differs by coast. We shift the Philippine working day by up to two hours in either direction. On the same 9-to-6 day the table above assumes, two hours earlier widens the live window with US Pacific from one hour to three against Pacific Standard Time, and from nothing to two against Pacific Daylight. Two hours later cannot create US Eastern overlap, because the gap is too wide for two hours to close; what it does instead is halve the handoff, landing work one to two hours ahead of your morning rather than three to four. Both are arithmetic from the standard offsets [T4], not a promise about how fast anyone replies.
Two things belong on your diligence list before you sign anything here, and both are questions to ask rather than reasons to walk.
Attrition. Aon's 2025 study of more than 700 businesses across six Southeast Asian markets, fielded July to September 2025 [X2], projects Philippine attrition at 20.0% — the highest in the region [X1]. That's the market, not any one employer, which is exactly why it's a useful question: ask every vendor on your shortlist for their own figure and how they measure it. Annualised or trailing twelve months? Voluntary only, or all departures? Does it include people moved between clients? A provider who can't answer without going away to check is telling you something. So is one whose number is suspiciously far below 20%.
EU data transfers. The Philippines holds no EU adequacy decision, so an EU buyer transfers personal data on Standard Contractual Clauses [K2, K4]. SCCs are a lawful basis and most EU buyers use them routinely, but they come with a transfer impact assessment and residual risk, and your DPO — not your engineering team — decides whether that's acceptable. Settle it before you shortlist rather than in week three of a contract review.
Neither of those is a defeat and neither is a footnote. They're the two places where a Philippine engagement most often goes wrong quietly, which makes them the two questions worth asking loudly.
