Load Shedding, Typhoons and Uptime: Will Your Offshore Assistant Actually Be Online?

There is a version of this conversation that happens on almost every first call with a founder who has never hired offshore. The rates make sense, the skills check out, the timezone works. Then they say some version of: "But doesn't South Africa have those rolling blackouts?" Or, if the shortlist is Philippine: "What happens when a typhoon hits?"

It is a fair question and it deserves a straight answer rather than reassurance. It also deserves a current one, which is harder than it sounds, because the honest answer for South Africa has changed completely within the last eighteen months and most of the material a founder finds when they search is describing a situation that no longer exists. The advice you get is frozen at the worst moment of a crisis that has since ended.

So this guide does three things. It sets out what actually happened in each market, using the operators' own published numbers. It separates the risks that are genuinely gone from the ones that remain and will remain. And then it does the part that matters most, which is showing you that uptime is not really a property of a country at all. It is a property of one particular desk, in one particular house, and you can specify it, verify it before you make an offer, and write it into an agreement for a rounding error of what the hire is worth.

The short version

1. National load shedding in South Africa has stopped. Eskom passed a full year with no load shedding on 15 May 2026, the first time that has happened since September 2018.
2. What remains in South Africa is load reduction, which is a different thing. It is local rather than national, often unannounced, and concentrated in specific areas with overloaded distribution infrastructure. Eskom's stated aim is to eliminate it by 2027.
3. The Philippines fails in a completely different shape. There is no national rolling schedule. There are occasional grid alerts when generating reserve runs out, and there is a tropical cyclone season that produces a small number of genuine, safety-driven days off.
4. Neither country is the real variable. Two assistants in the same city, on the same street, can have wildly different reachability depending on what is plugged into what. The spread within a country is larger than the spread between countries.
5. The fix is a consumer purchase and a written rule, not an infrastructure project. A battery that carries a laptop and a router, a second internet path that fails independently of the first, and a named fallback location covers essentially all of it.
6. Ask before the offer, not after the first outage. Continuity is a screening question. Asking it late turns a solvable equipment gap into a disappointment about a person.

What actually happened to load shedding

For most of the last decade, South Africa's problem was generation. An ageing coal fleet with too many units out on breakdown could not meet demand, so the utility shed load in scheduled stages to stop the grid from collapsing. At the worst of it the country was shedding on the large majority of days in the year, and the cost to small businesses was not the blackout itself but the impossibility of planning around it.

That recovery is now several years old and the numbers are public. Since the generation recovery plan began, Eskom's Energy Availability Factor, which is the share of installed capacity actually available to run, improved from 54.56% to 65.16%. Unplanned outages moved the other way, from 32.34% down to 22.88%. The utility reports meeting 100% of national demand and saving R26.9 billion in diesel costs over three years, diesel being the expensive open-cycle turbine fuel it burned to avoid shedding in the bad years.

The headline event is simple enough to state. On 15 May 2026 Eskom marked 365 consecutive days without a single second of load shedding, a milestone last reached in September 2018. The last stages were shed in March of the previous year: a stage 3 block over one weekend and an overnight stage the following week after five generating units went offline around the evening peak. Since then, nothing.

The government position is now explicit. The Department of Electricity has stated that the national electricity system is fully stabilised and that South Africa is no longer in an energy crisis, and the President has said the country has put load shedding behind it. Eskom's winter outlook again predicted no load shedding for the peak-demand months. The reported margin behind that is roughly 4,400MW more available generation than in the equivalent period a year earlier.

Two honest caveats belong next to that, because a guide that oversells this is no more useful than one that is out of date.

The first is that a system running at an availability factor in the mid sixties is a system that works, not a system with abundant margin. The transmission company's medium-term adequacy modelling still contains scenarios in which unplanned outages run above the base case and the reserve thins out. "Solved" is the right word for the crisis. "Impossible" is not the right word for a future outage.

The second is that generation was never the only way the lights go out. Cable theft, substation fires, storm damage, municipal distribution faults and the ordinary maintenance of a distribution network all continue, in South Africa exactly as they do in Texas or Sydney. What changed is that outages went back to being local, occasional and unscheduled events rather than a national timetable that everyone planned their week around.

Load reduction is not load shedding, and it is the thing that might actually hit somebody

This distinction matters more than any other single fact in this guide, because it is the one that survived the recovery and it is the one most employers have never heard of.

Load shedding was national. When it was in force, it affected everyone on a published schedule, in stages, with an app that told you your two hours. Load reduction is local. It is applied to specific feeders in specific areas where demand on the physical network has grown past what the equipment can carry, usually because of illegal connections, meter tampering and overloading rather than because the country is short of power. It is often applied without notice, and it typically hits in the early morning and early evening when household demand peaks.

Three consequences follow for an employer.

It is geographically specific, so it is a question about a person and not a country. Two candidates in the same metro can be on completely different sides of this. The right question is never "does South Africa have power cuts", it is "has your street had an outage in the last three months, and how long was it".

Its timing is unhelpful but not fatal for most roles. Early morning and early evening cuts land squarely on the start of a European working day and on the tail of a US afternoon. That is exactly the kind of predictable, bounded event a small battery solves, which is why the spec section below matters more than the news section above.

It is on a stated path to ending. Eskom has said it intends to eliminate load reduction by 2027, with the mechanism being smart meter conversion and network upgrades in affected areas. That programme has run into real friction, with more than 122,000 planned conversions delayed by community resistance and safety incidents, so treat the target as a direction rather than a date on which the risk disappears.

The Philippines fails differently, and the difference is instructive

Founders often assume the Philippine risk is a milder version of the South African one. It is not. It is a different shape, with a different frequency, a different duration and a different social contract around it.

Grid alerts

The Philippine grid does not operate a national rolling schedule. What it does is publish alert levels when the reserve margin thins. A yellow alert means operating reserve has fallen below the required level. A red alert means the reserve is gone and supply is insufficient to meet demand plus the required buffer, at which point distribution utilities may rotate outages, known locally as brownouts.

The most instructive recent episode ran over three days in May. The Luzon and Visayas grids went to red alert after two 500 kilovolt transmission lines tripped, disconnecting Ilijan units 1 and 2 at 1,200MW and three Excellent Energy Resources units totalling 1,262.1MW in one event. That sat on top of a background of 17 plants on forced outage and 14 more running derated, for about 4,242.5MW unavailable to the grid in total. Alerts held for a third straight day before being lifted that weekend.

Read that carefully, because the lesson is not the one the headline implies. A red alert is a warning about reserve, not an announcement that the lights are off. Many red alerts pass with no customer losing supply at all. When rotating outages do happen they are typically a few hours, in some service areas and not others, during a peak window. The practical planning assumption for an employer is a handful of afternoon interruptions a year for an assistant in an affected area, not a recurring daily schedule.

Tropical cyclones

This is the larger and more predictable factor, and it is the one worth building a policy around rather than improvising. PAGASA, the national weather service, counts an average of about twenty tropical cyclones entering the Philippine area of responsibility each year, of which eight or nine make landfall. The season runs June through November, and roughly seventy percent of them form in the July to October peak.

Most of those produce nothing more than rain where your assistant lives. A few produce a genuine, unavoidable, multi-day interruption: no power, no fixed line, damaged infrastructure, and a person whose priority is correctly their family and their roof rather than your inbox.

The rules around this are worth knowing because they are more formal than most foreign employers expect. PAGASA raises tropical cyclone wind signals from 1 upward, and Executive Order No. 66 governs automatic cancellation of classes and government work at given signal levels. For private employers, Department of Labor and Employment Labor Advisory No. 17 sets the framework: an employer may suspend work as a management prerogative in the interest of employee safety; employees who do report and work at least six hours are entitled to full regular pay; employees who do not work are subject to the no work, no pay principle unless a company policy, practice or agreement is more generous; and, importantly, an employee who declines to work because of imminent danger from the weather cannot be subject to any administrative sanction for it.

That last clause is the one to internalise. A Philippine assistant who tells you they cannot work through a signal 3 storm is not making an excuse and is not underperforming. They are doing the thing their own labour framework explicitly protects, and any manager who treats it as a reliability failure will lose a good person for no reason. Budget one to three genuinely lost days a year in the season, treat them as weather rather than performance, and decide in advance whether you pay for them. The generous answer costs very little and buys a great deal.

Uptime is a property of the desk, not the country

Here is the part that changes outcomes. Every failure described above resolves, at the individual level, into three layers that you can specify. Get all three and an assistant works through nearly everything short of a direct storm. Get none and they will be offline for a two-hour municipal fault that a colleague three streets away did not even notice.

Layer one: power at the desk

The single most common mistake founders make here is thinking about backup power as a house-sized problem. It is not. You are not powering a geyser, an oven or an air conditioner. You are powering a work stack, and a work stack is small.

Run the arithmetic. A modern laptop under real working load draws roughly 45 to 65 watts including charging. A fibre terminal and router together draw about 10 to 15 watts. An external monitor adds 20 to 30. A desk phone or headset charger is negligible. Call it around 100 watts with a monitor and around 60 without.

Against that, a 500 watt-hour portable power station carries the full stack for roughly five hours, or the laptop and router alone for closer to eight. A 1,000 watt-hour unit covers a normal working day outright. An inverter and battery installed on a household circuit gets you to the same place by a different route and is common in South African homes. None of these is exotic equipment. All of them are ordinary consumer purchases with a several-year life, which is why treating the equipment as a one-time onboarding item rather than an ongoing problem is the correct posture.

The detail that gets missed: the router has to be on the battery too. A charged laptop with a dead fibre terminal is a person who cannot work. This is the single most common gap in an otherwise well-equipped home office, and it costs almost nothing to close.

Layer two: two internet paths that fail independently

Redundancy only counts if the two paths do not share a failure. Fibre plus a mobile hotspot is genuine redundancy when the outage is a cut line or a router fault. It is much weaker than it looks during a wide-area power event, because mobile base stations run on their own batteries and a long enough outage takes the tower down with everything else.

So the requirement is more specific than "have a backup". It is: a primary fixed line, plus a mobile fallback on a different carrier from whichever one the assistant's household already uses, plus the knowledge of which of the two networks in their area actually holds up. Any assistant who has lived through a few outages knows this answer immediately. It is a very good interview question precisely because it cannot be bluffed.

Layer three: a named place to go

Some outages outlast the battery. For those, the answer is a location, decided in advance, not improvised at the moment of failure. A co-working space, a mall with reliable power and wifi, a library, a relative's house on a different feeder. The test is that it is reachable in under thirty minutes and that it has been named and agreed before it is needed, along with who pays for the day pass.

The reason to settle this in advance is behavioural rather than logistical. An assistant who has a plan uses it. An assistant who has no plan sits at home hoping the power comes back, and you find out four hours later.

The continuity specification

This is the whole thing on one page. Take it into a hiring conversation as it stands.

Notice that five of the seven rows are verified by a question rather than a document. That is deliberate. Continuity is one of the few areas where a short conversation genuinely tells you the answer, because people who have solved it describe their setup in specifics and people who have not describe it in adjectives.

Verify before the offer, not after

The failure pattern worth avoiding is not an outage. It is discovering the equipment gap in week three, after the relationship has already been framed as one about reliability rather than one about hardware. By then a solvable purchase has become a disappointment about a person, and it is very hard to walk that back.

Four things to do while you are still choosing:

1. Run one call at the real hour. Not a convenient mid-morning slot for them. The actual hour of the actual shift, on the actual connection. This single test surfaces more than any questionnaire.
2. Ask for the speed test at the desk. A screenshot with a timestamp. You are not looking for a headline number so much as a sane upload figure and a low enough latency for calls. Anything comfortably above 25 megabits down and 5 up handles video calls, shared documents and a helpdesk at the same time.
3. Ask about the last outage specifically. "When did your power last go out, how long was it, and what did you do?" The answer is either immediate and detailed or it is vague. Both are informative.
4. Ask what is plugged in. Not whether they have backup. What is connected to it. This is the question that finds the unplugged router.

If you are running a structured process, this fits naturally into the screening stage alongside the role questions. The interview question generator builds the role-specific set, and continuity items sit best right at the end of the first call, once you already want the person.

What to write into the agreement

Six clauses, none of them long, all of them worth more than the hour it takes to add them:

1. Working hours and the overlap you need. State the hours in the assistant's own timezone as well as yours, so nobody is doing mental arithmetic during an incident. The timezone overlap calculator will produce the actual window if you have not fixed it yet.
2. The notification rule. A message within fifteen minutes of losing power or connectivity, stating the expected duration and whether they are moving to the fallback location. This is the clause that does the most work.
3. The fallback obligation and who pays. Name the location, name the trigger, name who covers the day pass or the data. Ambiguity here reliably produces inaction.
4. Equipment ownership. If you fund the battery or the second connection, say whether it is an allowance, a reimbursement or company property, and what happens to it at the end of the engagement.
5. How lost hours are treated. Rescheduled, made up later, or simply not billed. A contractor invoicing hours needs a different answer from a full-time managed hire, but both need an answer in writing before it comes up.
6. Weather and safety carve-out. An explicit statement that a genuine safety-driven interruption is not a performance matter. In the Philippines this aligns you with the framework your assistant already works under, and it costs you almost nothing to be the employer who wrote it down.

The contract template generator produces a base you can add these to, and the contractor or employee guide covers the classification question that determines which version of clause five you need.

Planned absence is the bigger number, and almost nobody budgets for it

Here is a thing worth sitting with. If you total up the hours an assistant in either market is likely to lose to power or connectivity in a year, with a decent setup, you land somewhere in the low tens of hours. If you total up public holidays, that number is several times larger, entirely predictable, and published years in advance.

South Africa observes twelve public holidays, with a statutory rule that moves a holiday falling on a Sunday to the following Monday. The Philippines splits its calendar into regular holidays and special non-working days, set by annual proclamation. Neither set lines up with a US or UK calendar, and the mismatch runs in both directions: days your assistant is off while you are working, and days you are closed while they are available.

That is a scheduling problem with a complete solution, which is more than can be said for the weather. The offshore holiday planner lays both calendars against yours and marks every gap, and it is a better use of an hour than worrying about a grid that now meets 100% of national demand.

Common questions

Is load shedding really over in South Africa?

National load shedding has not been implemented for more than a year, with the 365-day mark passed on 15 May 2026 and the streak continuing since. The government has stated the country is out of the energy crisis. What has not ended is localised load reduction in specific areas with overloaded distribution networks, which Eskom aims to eliminate by 2027, nor the ordinary faults, cable theft and storm damage that interrupt supply in every country.

Should I hire in the Philippines instead to avoid this?

Not for this reason. The Philippines has its own interruptions, including grid alerts when generating reserve runs thin and a tropical cyclone season that produces genuine multi-day outages. The two markets are not meaningfully separated by uptime. They are separated by timezone, accent and language profile, cost, and the depth of particular skill pools, which is what the Philippines versus South Africa comparison is for.

How many hours a year should I actually expect to lose?

With a battery on the desk, the router on the battery, and a working mobile fallback, most interruptions become invisible to you and the realistic figure is small. Without that equipment, a single two-hour municipal fault becomes a lost half-day because the assistant has nothing to fall back on. The equipment, not the country, is what moves this number.

Who pays for the backup power and the second connection?

There is no single right answer, but there is a wrong one, which is leaving it undecided. Many employers treat it as a one-off onboarding allowance, on the reasoning that it is cheap relative to the role and it buys a year of not thinking about the problem. Others build it into the rate. Either works. Silence does not.

Should I use monitoring software to check my assistant is actually working during an outage?

No. Monitoring software is a poor substitute for a clear scope and it answers a question you did not ask. If you want to know whether the work is happening, look at the work. The performance scorecard generator builds the measures for that, and the reasoning behind avoiding surveillance tooling is covered in the security and access guide.

What about a typhoon that takes someone offline for days?

It happens, a few times a year across a whole team rather than for any one person. Plan for it structurally: no single person as the only holder of a critical daily task, written procedures so someone else can pick up a queue, and an explicit policy that safety-driven absence is not a performance issue. That is the same continuity thinking you would apply to a local employee having surgery, and it is why documented process is worth more than any piece of hardware.

What to do this week

1. Delete the outdated objection. If a mental model formed during the worst of the crisis is what is blocking a South African hire, it is describing a country that no longer exists in that form. Check the current position rather than the position you remember.
2. Add four continuity questions to your first call. Last outage and duration. What is plugged into the backup. Which mobile carrier and why. Where they would go if it lasted all day.
3. Decide the equipment position before you interview. Allowance, reimbursement, or built into the rate. Write the number down so you can answer it in the moment without improvising.
4. Write the notification rule into the agreement. Fifteen minutes, expected duration, moving or not moving. One sentence.
5. Solve the predictable gaps first. Map the holiday calendars and fix the overlap window before you spend another minute on the weather.

The founders who get this wrong are rarely the ones who took a risk. They are the ones who read something written at the height of a crisis, concluded that an entire country could not be relied on, and quietly narrowed their hiring to a market that costs three times as much and has its own hurricanes, wildfires and ice storms.

If you want the equipment question settled before a name is ever put in front of you, that is part of what a recruiter-led process is for. You can request candidates and see profiles where the working setup has already been checked, or book a meeting and walk through the specific hours you need covered and what happens on a bad day. If you are still at the earlier stage of the decision, how it works covers the process end to end, and the South African and Philippine market pages cover what each pool is actually strong at.

The grid was a real problem. It was also somebody else's problem to fix, and they largely fixed it. Yours is smaller and entirely within reach: one battery, two connections, one named backup location, and a rule about when to send the message.

Sources

• Eskom delivers 365 days without loadshedding, Eskom, on the 15 May 2026 milestone, Energy Availability Factor and unplanned outage figures, and diesel savings.
• Eskom's power system remains stable, recording year on year gains, Eskom, on availability improvements and the load reduction programme.
• Winter outlook for power grid again predicts no loadshedding, Eskom media statement.
• Eskom declares end to load shedding: is the energy crisis truly over?, IOL, on the Department of Electricity statement, load reduction in high-risk areas, the 2027 target and delayed meter conversions.
• South Africa marks exactly 365 days without loadshedding, Eyewitness News.
• NGCP places Luzon, Visayas grids under red, yellow alerts, Rappler, on the May grid event, tripped transmission lines and unavailable capacity.
• Brownouts return as Luzon, Visayas grids hit red alert; here's what the terms mean, Philstar, on the definitions of yellow and red alert.
• Tropical Cyclone Information, PAGASA, Department of Science and Technology, on annual cyclone counts, landfall frequency, season and wind signals.
• Executive Order No. 66, on cancellation or suspension of classes and work in government offices due to typhoons and other weather disturbances, Philippines.
• Rules on work suspension, payment of wages during weather disturbances, Department of Labor and Employment, Philippines, on Labor Advisory No. 17 and the treatment of private sector work suspension.