Split the annual savings figure into two lines before you trust any payback number: the kilowatt-hours the house actually uses on site, valued at the full retail tariff, and the kilowatt-hours pushed back out through the municipal meter, valued at whatever feed-in credit applies. A used unit is worth the whole retail rate. An exported unit is worth a fraction of it, and on plenty of South African municipal accounts it is worth nothing at all. Blend the two into one cheerful "annual savings" line and the payback you calculate stops describing the house you actually live in.
Stand with me at the meter box on the boundary wall at ten past six on a July evening. The sun has dropped behind the neighbour's roof, the inverter fan has gone quiet, and the little red pulse light on the meter is ticking away like a metronome because the kettle is on and the geyser element has just kicked in. Everything the panels made between ten and three went into the geyser, the pool pump and the tumble dryer. Whatever was left over went out through this box into the street. The meter doesn't care which is which. It counts both. The account values them completely differently.
That took me an embarrassing amount of time to see properly, and I only saw it because I stopped reading the inverter app's headline number and started reading the municipal account next to it.
The overlap is the only part that behaves like the brochure
Run the inverter app side by side with your own habits for a week and the shape of it becomes obvious. Generation is a smooth bell: nothing at six in the morning, a fat hump from about ten to three, nothing by six. Consumption is jagged. A spike at half past five in the morning when the kettle and the geyser go. A low, flat base during the day, mostly the fridge and the standby draw. Then a wall between six and nine at night when everyone is home, the stove is on and the lights are burning.
Where those two shapes overlap, you are saving money at the full retail rate. Every one of those units is a unit your municipality did not sell you. Outside the overlap, in the midday hump where generation runs ahead of the house, those units leave through the meter and get credited at the feed-in rate, which is usually lower, and sometimes is nothing because feed-in isn't permitted on your tariff at all.
So the honest formula is not savings equals generation times a rate. It is:
annual savings = (self-consumed kWh x retail rate) + (exported kWh x feed-in credit)
That's it. Two lines. The reason quotes don't show it that way is that the blended single-line version always looks better.
What the blend hides
Take an illustrative system generating 6 000 kWh over a year, where the household manages to use 2 500 of that on site and exports the other 3 500. Use placeholders, not my numbers: say R3.00 per kWh at retail and R0.75 per kWh as the feed-in credit, purely to make the arithmetic visible.
Self-consumed: 2 500 x 3.00 = R7 500. Exported: 3 500 x 0.75 = R2 625. Annual saving: R10 125.
Now change nothing except the timing. Same panels, same roof, same account, but the geyser and pool pump run in daylight and a battery carries the evening peak, so 5 500 kWh is self-consumed and only 500 is exported. Self-consumed: R16 500. Exported: R375. Annual saving: R16 875.
Identical hardware. Two-thirds more savings. The difference is not the equipment, it's the overlap.
That is where the blended number does its damage. Divide total savings by total generation and you get an average value per kilowatt-hour, something like R1.69 in my first case. People quote that figure as though it's a property of the system, a spec you can shop for. It isn't. It's a property of your household's timing, and it moves every time somebody in the house changes shift or buys a heat pump.
The part that surprises almost everyone
Bigger arrays tend to make this worse, not better, and nobody puts that on the front of a brochure.
The first few kilowatts of panels get eaten by the daytime base load and whatever you can shift into daylight. Those earn the full retail rate. But once the array is already covering the midday load, the next panel's output lands squarely in the hump, where there's nothing left to use it, and it exports at the feed-in credit instead. Marginal panels earn the marginal rate, and the marginal rate is the bad one.
You paid the same price per panel for a kilowatt-hour that earns a third of what the one next to it earns. That's the sentence I wish somebody had said to me before I started counting roof space instead of counting appliances.
So the array isn't sized against the annual bill. It's sized against the midday load and the battery's charge rate. If your bill is large mainly because of evening consumption, extra panels are an expensive way to attack it. The battery is what attacks it, and the battery has its own arithmetic.
Where a battery fits into the rate, not the headline
Without a battery, self-consumption is capped by whatever the house happens to be drawing while the sun is up. Fridges, standby draw, a timed geyser, a pool pump on a schedule. Everything after that is exported, and everything after sunset is bought back at retail. That cap is what stretches simple payback on a batteryless system, and it's the reason two houses on the same street with the same quote can have wildly different outcomes.
With a battery, the midday hump gets moved into the evening. Each stored unit that discharges into the house avoids the full retail rate rather than earning the feed-in credit, which is a large jump in value per unit. Not the full rate, mind you, because round-trip losses take a bite, and whatever the inverter burns staying awake takes another. A stored unit is worth something less than retail. But it's still worth far more than a unit you export, which is the whole argument.
And the battery costs money, so don't fold it into the panel payback and call it one number. Run the panels with no battery first, then run the increment. The question isn't whether the battery helps the rate. It does. The question is whether the units it rescues from export are worth what you paid to rescue them.
The local rules set one of the two rates
There's a limit to how far any general formula can carry you, and this is it. Municipalities differ. Some credit exported energy through a bi-directional meter on an approved scheme, some credit at a lower rate than retail, some restrict it to particular tariff categories, and some don't permit it at all and expect reverse power flow to be prevented. That means the feed-in credit in your formula is not a number you can look up in a national table. It's whatever your own municipality's small-scale embedded generation rules actually allow, and until that is settled, the export half of your calculation is a guess.
I don't assume an export meter will be approved just because the inverter supports it. If feed-in isn't available to you, put zero in that second bracket. Then look at what's left: the entire value of the system sits in the overlap, and sizing and scheduling decisions have to be made on that basis.
What I'd actually do with a quote
Ask for the split. Not the annual savings line, the split. How many kilowatt-hours did you assume are self-consumed, how many exported, what retail rate, what credit rate. If the person who built the proposal can't answer that, the payback figure on the last page is decoration.
Then do the check yourself, in this order. Pull twelve months of municipal accounts, not one. Note what you actually pay per unit all-in. List the loads that can be moved into daylight, and be honest about which ones you'll genuinely move. Look at the inverter graph for a normal weekday and see how much the house draws between ten and three without anyone trying. That flat daytime line is the real ceiling on self-consumption without a battery.
Then run the two-line formula with your own numbers, and run it again with a pessimistic self-consumption figure. If the payback still holds up when you assume most of the generation is exported at a poor credit, you're looking at a system sized for your house. If it only works when nearly everything is self-consumed, you're looking at a system sized for a spreadsheet.
