If I Started My Solar Installation Again, I’d Design It Very Differently

My solar and battery system works well. I want to say that clearly at the top, because what follows is a list of things I would change, and it would be easy to read it as a system full of regrets. It is not. It generates, it stores, my battery optimiser squeezes real money out of the Agile price curve, and on a good day the house runs itself off sunshine and cheap overnight electricity. If you offered me the exact same installation again I would take it happily.

And yet. Having lived with it, watched a year of data flow past, and started bumping into its edges every time I want to add something, I have a clear and slightly frustrating picture of how I would design it if I could start over. Almost none of my changes are about the parts you can see. They are about the parts you cannot: the space I did not leave, the cable I did not run, the headroom I did not buy. Every one of them would have cost almost nothing on installation day. Several of them will now cost me thousands to fix, or simply cannot be fixed without tearing things out.

This is the article about that gap — the enormous difference between the cost of designing for the future and the cost of retrofitting it. It is the most useful thing I have to say about domestic solar, and it comes straight out of a career spent watching the same lesson play out in enterprise infrastructure, where the projects that aged well were never the cheapest ones. They were the ones that left room.

The principle, before the list

There is one idea underneath everything here, and it is worth stating on its own because every specific example is just an instance of it.

The expensive thing is never the component. The expensive thing is the access to install the component — the open wall, the empty conduit, the spare capacity, the moment when the van and the electrician and the exposed cable run all exist at once.

On installation day, that access is free. The walls are open. The conduit is being run anyway. The electrician is already there, already terminating cables, already certifying. Adding a little more — a bigger conduit, a spare cable, an inverter one size up, a data run to where the battery lives — is marginal. It is minutes and a few pounds against a job that is happening regardless.

The day after installation, that access is gone. The walls are closed. The conduit is full. The electrician has moved on. Now every addition is its own project, with its own call-out, its own disruption, its own certification, and frequently the removal of something you already paid for. The component still costs the same. The access has gone from free to expensive, and the access is the real price.

Enterprise IT taught me this in a hundred forms. You do not run a single cable to a new rack; you run a bundle and leave most of it dark, because the second visit costs more than the extra cable ever will. You do not buy a switch with exactly enough ports; you buy headroom, because the day you need port forty-nine you do not want to be replacing the whole switch. Domestic solar is the same discipline wearing different clothes, and I simply did not apply it hard enough the first time.

Let me go through the specifics, roughly in order of how much I wish I had done them.

Oversize the conduits — this is the cheapest regret to avoid

If I could send one instruction back in time, it would be this: run bigger conduit than you need, and run more of it than you need.

Conduit — the protective tubing that cables run through — is astonishingly cheap. The labour to install it is almost entirely in the routing: deciding where it goes, drilling through walls, fixing it in place, making it neat. Once you are doing all of that, the difference in cost between a conduit sized exactly for today’s cables and one with generous spare capacity is trivial. A few pounds of material. Nothing in labour, because the route is the route regardless of the diameter.

My conduits are sized for what was installed. They are, in the language I would use professionally, running at close to a hundred percent utilisation. Which means the day I want to add a cable — for a second battery, for an EV charger’s load-management signal, for a data run to something new — there is no room, and “add a cable” becomes “run a new conduit,” which becomes “route a new path through the building,” which is exactly the expensive access problem I just described.

The rule I would give anyone: conduit should never be full on installation day. Leave it half empty. Leave draw strings in the spare capacity so pulling a future cable is a two-minute job rather than a fishing expedition. You are paying for the route once; buy enough route to use twice.

Plan the cable routes as if you will use them again

Related, but distinct: think about where the cables go, not just how big the conduit is.

When my system went in, the cable routes were planned to solve the immediate problem — get power from the array to the inverter, from the inverter to the battery, from the battery to the consumer unit — by the shortest sensible path. That is a perfectly reasonable way to install a system. It is a poor way to design one that will grow.

The routes I would plan now would treat the inverter and battery location as a hub that future things will need to reach, and I would run pathways to the places I might plausibly expand: to where a second battery might go, to the garage or driveway where an EV charger lives, to the loft where more panels might terminate, to wherever the network gear sits. Not the cables themselves, necessarily — just the pathways, the conduit and the draw strings, so that reaching those places later is a pull rather than a build.

This is capacity planning, and it is exactly the same mental move as planning structured cabling in a building. You do not cable to today’s desk layout; you cable to a grid that survives the next three reorganisations, because the reorganisations are certain and the cable is cheap. My solar cabling was designed for the layout of the system, not for the layout of every system it might become.

Buy inverter headroom — the ceiling I hit first

Of all the limits I have bumped into, the inverter is the one that bites soonest and hardest, and it ties directly to the G98 and G99 rules I wrote about earlier in this series.

My inverter is sized for my system. Sensible, proportionate, and — this is the trap — sized with essentially no headroom above what was installed. Which means that the natural next step, adding more battery capacity or more generation, runs straight into the inverter’s ceiling. To grow past it, I do not add a component. I replace the inverter, which is one of the most expensive single items in the whole system, and possibly re-open the grid-connection question at the same time.

Contrast that with the alternative that was available on day one: a slightly larger inverter, perhaps configured with an export limit to keep the grid connection simple, sized to leave room for the battery and generation I might add later. The marginal cost of the larger inverter at purchase would have been a fraction of the cost of replacing the smaller one later. I bought exactly enough, and “exactly enough” turned out to mean “nothing left for the future.”

In enterprise terms this is the sin of sizing a platform to the current workload with no allowance for growth, and then discovering that scaling means a forklift upgrade rather than an incremental one. I have warned customers off that exact mistake. I made a domestic version of it in my own house.

Position the battery for the battery you’ll have, not the one you bought

Battery positioning looks like it is only about today’s battery. It is not. It is about whether tomorrow’s second battery has somewhere to live.

Home batteries are heavy, they have thermal and safety requirements about where they can be mounted, and they need to be near the inverter and the electrical connection. When mine was installed, it was placed in a spot that suited it perfectly — and left no obvious, prepared space beside it for a second unit. Adding capacity later means either finding a new location and running new cable to it, or reworking the space around the existing one.

If I designed it again, I would treat the battery location as a bay, not a spot. I would ask: if I double the storage, where does the second unit go, and is that space reserved, accessible, correctly rated, and already reachable by conduit? Leaving a prepared, empty space next to the battery costs nothing but a bit of wall. Not leaving it means the second battery’s installation includes a location-hunting exercise and a new cable run that the first one’s placement should have anticipated.

Design the future-proofing that costs almost nothing

Beyond the big-ticket items, there is a whole category of decisions that cost essentially zero at installation and quietly determine what is easy later. I would build every one of these in without hesitation next time.

Network connectivity to the equipment. This is the one I most underestimated, coming from an IT background where I should have known better. My inverter and battery live in a spot with awkward network access, so getting reliable data off them — for monitoring, for Home Assistant integration, for the optimiser that makes the whole thing worthwhile — was harder than it should have been. A single Ethernet run to the equipment location, pulled while the conduit was open, would have made everything downstream trivial. Wireless is a fallback, not a plan, for something you want to depend on. Run the cable.

Monitoring hooks from day one. I treat observability as non-negotiable in every system I build, and I wrote at length in the battery optimiser piece about how the Grafana dashboard was what let me trust the automation. But the physical ability to get the data has to be designed in. That means network to the equipment, an inverter and battery chosen partly for how open their data interfaces are, and a location where the monitoring gear can sit. You cannot bolt good observability onto a system that was installed with no way to see inside it.

Home Assistant integration as a first-class requirement. If you know you want to integrate with Home Assistant — and if you are the kind of person reading this, you probably do — then say so before the equipment is chosen, because it should influence which inverter and battery you buy. Some expose clean local APIs. Some are locked cloud gardens that fight you at every turn. The difference between “integrates in an afternoon” and “reverse-engineering a cloud API for weeks” is a purchasing decision made months earlier. I got reasonably lucky. I would rather have been deliberate.

Smart tariff readiness. The system should be designed knowing it will run on a half-hourly tariff, because that changes what “good” looks like. An inverter and battery that can be commanded to charge and discharge on a schedule — programmatically, reliably, at the granularity the tariff works at — are worth far more to a smart-tariff household than ones that only offer crude built-in timers. This is a capability you specify up front, not a setting you discover you are missing.

Design for the capabilities you might grow into

Then there is the category of things I do not have yet, but designed myself out of being able to add easily. These are the “allow for growth” decisions, and they are the ones enterprise architecture is most obsessive about.

UPS and backup capability. Most grid-tied systems, mine included, do nothing useful when the grid goes down — the inverter shuts off for anti-islanding safety, and you sit in the dark next to a full battery, which is a special kind of insult. True backup or islanding capability, where the system can keep selected circuits alive during an outage, is largely a design and wiring decision made at installation: which circuits are backed up, how the changeover works, what the inverter is capable of. Retrofitting it means rewiring at the consumer unit. I would design the backup path in from the start even if I did not commission it immediately, because the wiring is the expensive part and it is cheapest while the walls are open.

Generator input. I do not have a generator and may never want one. But allowing for an input — a way to feed the system from an alternative source during a long outage — is a wiring provision, and provisions are cheap. This is the “leave a port” instinct: you do not have to use it, but the cost of the option at build time is near zero, and the cost of adding it later is a rewire.

EV charging coordination. My EV charger and my battery are, for now, two systems that occasionally fight over the same cheap half-hours — something I flagged as a roadmap item in the optimiser. Designing for coordination means the charger can be load-managed, can see the rest of the system, and shares the signalling paths that let one intelligence schedule the whole house. That is a cabling and equipment-selection decision. Two systems that cannot talk were, at some level, a design choice I made by not making it.

Additional panels and roof expansion. If there is any chance of adding panels later — a garage roof, an extension, an outbuilding — the inverter headroom and the cable pathways to those roof faces should anticipate it. Adding a string to a system designed to accept one is straightforward. Adding it to a system running at its ceiling means, again, the inverter conversation and possibly the grid-connection conversation all over again.

The enterprise IT parallel, made explicit

I keep gesturing at my day job, so let me make the parallel concrete, because it is not decoration — it is genuinely where this discipline comes from, and I will develop it fully in the enterprise IT article that closes this series.

In enterprise infrastructure, the cost of change is dominated by whether you designed for it. A datacentre laid out with spare power, spare cooling, spare cable containment and spare rack space absorbs growth as a series of cheap incremental additions. A datacentre built to exactly today’s requirement absorbs growth as a series of expensive disruptive rebuilds — and the second kind always, always costs more over its life, even though it looked cheaper on the day it was signed off.

The specific habits map almost one-to-one:

Every one of those enterprise habits exists because an entire industry learned, expensively and repeatedly, that retrofitting growth costs far more than provisioning for it. The lesson is not solar-specific. It is a general truth about physical systems, and domestic renewables are physical systems that happen to be installed by a trade that — as I argued in the first article — has not always inherited the capacity-planning culture that enterprise infrastructure had beaten into it.

Common mistakes, including mine

Pulling the threads together, here are the mistakes I see most often — several of which I made myself.

• Sizing every component to exactly today’s need. The single most common and most expensive pattern. Exactly enough is a ceiling, and ceilings hurt.
• Treating the installation as an endpoint, not a starting point. A system designed as a finished object rather than a platform that will grow forecloses growth by default.
• Ignoring the data path. Installing energy equipment with no thought to how you will monitor and integrate it, then discovering that the interesting half of the value — optimisation, automation, visibility — is hard to reach.
• Full conduits. The cheapest regret to avoid and one of the most common to have. If it is full on day one, it is a wall around your future.
• Positioning for the component, not the capacity. Placing the battery, inverter and cabling for exactly what was bought, with no reserved space or pathway for the obvious next addition.
• Leaving the grid-connection headroom on the table. Sizing right up to the G98 line with no plan for what happens when you want more, so that the first expansion re-opens the whole connection question.

Best practice, if you are installing now

If you are at the design stage, here is what I would actually do, in priority order:

Design the system as a platform, not a purchase. Ask, for every decision, “what does this foreclose?” A choice that closes doors should be made consciously, and only when the saving is worth the lost option.

Provision generously for the cheap things. Oversize conduit. Run spare pathways. Pull an Ethernet cable to the equipment. Leave a prepared bay for a second battery. Wire the backup path. These are the near-free provisions that make the future cheap, and there will never be a cheaper moment to make them than while the installer is already there with the walls open.

Buy headroom on the expensive things where you can. A slightly larger inverter, an export-limited larger system, an equipment choice with open data interfaces — these are where a small premium at purchase buys a large saving at expansion. Spend it where the retrofit cost is highest.

Specify integration and monitoring before you specify equipment. If you want Home Assistant, smart-tariff control, and real observability, those requirements should shape which inverter and battery you buy. Decide them first, then choose hardware that serves them.

Write down the reasons. Every non-obvious decision — why this inverter, why this export limit, why the battery is here — should be recorded, because future-you, or the next installer, inherits a system they can only safely change if they can understand it. This is the same discipline that runs through everything I build, from the homelab to my professional work.

The lessons I actually took

Three, distilled from a year of living with the result.

Future-proofing is almost free at build time and almost never affordable afterwards. This is the whole article in a sentence. The asymmetry is enormous and it runs entirely in one direction. Provision generously while the access is free, because the access is the real cost and it disappears the day the installer leaves.

A working system and a well-designed system are different things. Mine works beautifully and was designed adequately. The gap between “works” and “well-designed” is invisible until you try to change something, at which point it becomes the only thing you can see. Designing for change is what separates a system you can live with from a system you can grow with.

The discipline is transferable, and I already had it. The most annoying part is that I knew all of this. I have spent years telling customers to provision for growth, leave headroom, design for change. I simply did not apply it hard enough to my own house, because it did not feel like infrastructure — it felt like a home improvement. That was the error. It is infrastructure, and treating it as anything less is how you end up with a system that works today and fights you tomorrow.

Summary

• My solar and battery system works well — the changes I would make are almost all about what I did not provision for, not about what was installed badly.
• The governing principle: the expensive thing is never the component, it is the access to install it. On build day, access is free — open walls, running conduit, an electrician already present. Afterwards, access is expensive, and access is the real price.
• The cheapest regrets to avoid: oversize the conduit and never leave it full, run spare pathways to where the system might grow, and pull a network cable to the equipment so monitoring and integration are trivial.
• The most expensive ceiling I hit first was inverter headroom. Sizing to exactly today’s load means the first expansion replaces the inverter rather than building on it — tied directly to the G98/G99 limits.
• Position the battery as a reserved bay, not a spot. Design in the backup, generator and EV-coordination paths even if unused, because the wiring is the expensive part and it is cheapest while the walls are open.
• Specify integration and monitoring before hardware: whether an inverter exposes clean local data is a purchasing decision that decides whether Home Assistant and optimisation are an afternoon or a fight.
• The whole discipline is enterprise capacity planning applied to a house. Retrofitting growth always costs more than provisioning for it — a lesson infrastructure learned expensively and domestic solar has not yet fully inherited.

Next in the series: the questions every homeowner should ask before buying solar — a practical checklist that turns all of this hindsight into questions you can ask before you sign.