Nearly every system we get asked to fix was designed correctly for the load it had at the time. Worth saying plainly, because flexible design gets discussed as though the alternative were carelessness. Usually it was a system sized honestly against the demand in front of it, by people who had no particular reason to expect that demand to double.
So the real question is how much you are willing to pay today for capacity you will not use for a decade, and where that money buys the most room.
Where the money actually buys room
Not everywhere equally. Some decisions are cheap now and enormously expensive to revisit; others can be deferred with almost no penalty. Knowing which is which is most of the job.
Duct bank and conduit are the clearest example. Pulling a larger conduit than you need, or laying a spare, costs very little while the trench is open and is close to unaffordable once it is closed. The same goes for substation footprint, spare breaker positions, transformer pads, and physical space in a switchgear lineup. Square footage is cheap while you are building, and getting it back later means taking an outage to do it.
Conductor and transformers sit differently. Those you can often upsize later during a planned replacement, so paying for headroom you may never use is a weaker argument.
Redundancy is not the same as flexibility
These two get run together constantly. A second utility feed protects you when the first one fails and does nothing whatever about a load that has outgrown the service. A facility can be fully redundant and completely inflexible at the same time, and plenty of them are.
Flexibility comes from a system that can take a new source or a new load without being redesigned around it: room in the protection scheme, a one-line that anticipated a tie, metering and controls that can see a new resource instead of being confused by one.
Getting the growth number honest
All of this rests on a projection somebody has to make, and the projections we see are usually either a straight line drawn through the last three years or a number supplied by whoever wanted the project approved.
A better version starts with what you actually know: committed expansions, equipment being electrified, tenants under lease, production lines going in. Then a separate speculative layer on top, labeled as such, so when the board asks which parts of the plan rest on assumptions you can point straight at them.
