I compare mine to similar characters - specifically within the faction. I do tend to low-ball a little bit, but I'm getting better at that.
I also look at characters that maybe do the same thing in a different faction - keeping in mind that factions are balanced differently (i.e. Bastila Shan would have to be way more expensive in any other faction, even aside from the fact that she shouldn't be in any other faction). Also, I tend to look more at the actual usefulness of the piece as opposed to just volume of abilities - like, if a piece has Verpine Shatter Rifle, Double Claw Attack, and Repair 20, that sounds like a lot of abilities, but they're all red herrings in relation to each other.

Lastly, if it's a commander (or has a Camaraderie/Synergy/Rapport), I look at who it's helping much more than how it helping them. For instance, for any given ability, I think it should be way cheaper to give it to Clone Troopers than to give it to Stormtroopers, because Stormtroopers can be played at least semi-competitively, but Clone Troopers can't really at all.

an excellent question.
I took past discussions you had on this topic to mind,
and asked the locals too, back in ~13/14, surf.
this factored into discussions of whether SWMinis is "P-NP set complete",
and what factors cause SWMinis to have combinatorial sets larger than Chess, Go and some other tabletop games...
I am told it involves Sigma-Rho Calc. (Sigma Additivity's ugly cousin) and is a nested multivariate problem...
which in turn, turns out to involve Penrose Game Theory,
and Conway-Gardner-Rushkhovski combinatorics, K Wilber ring groups and all in order to answer it.
(that was a good paper to grade hehe).

a brief tangent to the theory;
Ideally, costing would be a function of playtesting the meta-combo interactions.
that would help more accurately define the cost of some things ahead of time,
rather than the Undercosting phenomena of the VSets and late WotC sets.
(Hutts has done some dot-plots around power creep and costing, which can be near a standard deviation different for VSets compared to early WotC).
And as you suggested Surf in the past, the multivariate is self-limiting from, what is the most you can legibly cram onto a statcard of a given standard? - size 6pt font seems to set the limit.
this in turn seems to limit the number of SA's, CE's and Force Powers, based on nCr, which can fit on the card...
good old babushka combinatorics hehe.
that is the 'hard theoretical limit'.

---

now, to the method itself.
we used to use a 'trial-and-error'/'pin the tail on the donkey' spitballing random guess method.
This was quick, but, it was often very different between the prototype cost and final cost,
as well as, it didn't compare apples with apples, leading to obsoleting of customs, or inconsistent weighting of the synergy of a special ability.

so
we came up with an Excel Spreadsheet,
which weights all the abilities on your prospective stat-card.
there's then, a subtotal with "any other modifier/synergy modifiers'.
before the final cost is tab'd in Column C3.

this project is eventually supposed to be AGI interoperable with the "automagic statcard converter"
so as we can scrape all statcards from the SWMinis Database, and make a statcard for them,
using GIMP Script-Fu and some AGI tricks...
well, we'd also like to be able to import the costing list from all custom abilities in the same database...
it gets tricky from there hehe.
It will be a multivariate, and it's hoped it'll reduce the time-per-statcard to less than 3mins total.

---

abilities can be +cost or -cost.
stuff like Speed 0/emplacement and under are -to net cost.

We know from past discussions here, and at Gamers, and WotC,
that, there is a base template Zero-point.

For 4points,
you get 10HP, 10DEF, +4ATK, 10DMG. No SA's, No ForcePowers or ForcePoints, No CEs.
that is the 'key' or base-line.

you then make it a nonlinear for attack. no miniature has a natural DEF greater than 28. or ATK greater than 20.
and nested If for the HP increment (since it is per 10).
Damage is +5 for 20DMG, +15 for 30DMG, +25 for 40DMG.

SA's incur a points penalty beyond 5 (ideally this would be |Avg or Median of SAs rounded to nearest whole integer from the SuperSet of all StarWarsMiniatures at time of making)
Force Powers is +5 for Force 1. +5 per force to 4 force, then non-linear thereafter. +10 for MotF 2, +25 for MotF 3 etc.
If you have greater than 4 force powers, there's a penalty...

CE's 1 CE is +5 points, with synergy factors highly weighing on that.
synergy would be objectively determined ideally from playtesting combinatorics and AGI, which is where the Electronic Journal of Combinatorics and StarWarsMiniatures Compositions (analogous to Chess Compositions) would come into it...

---

this produces the cost. it tends to err on the side of 'overcosting',
if we compare "TentativePrototypeCost" with "FinalizedCost" or "cross-comparativewithsimilarpieces",
we see it can be as much as 30-40% overcosted...
so we then playtest or look for abusive meta-combos, and subtract cost until its alright.
the logic on this is multivariate again,
and a number of factors balance -
are we oblomov/obsoleting an existing WotC or other mini?
are we shifting the meta? mathematically, what exactly is "meta-combinatorics", and what does the mathematical equivalent expression of "shifting the meta" mean? is that a non-standard analysis thing, like the Pingpong ross-littlewood?
a multivariate-ad-infinitum?

this is balanced by
"compareWithLikeStatCardInstances" -
so, if your custom is 99% identical to another extant custom, it will suggest you use the same cost.

thanks for reading.
if only Bloomilk had 'spoiler partitions', so as I could truncate this,
to prevent a look of a wall of text hehe.

@Koba - since I'm a History teacher and barely got a C- in Algebra, I'll take your word on it

@ Hutts, while I think Baine was too simple about it, I agree with you that its complex enough without following Koba's formula ....

Frankly I inherited an Excel program around 2007 from the ancestral WOTC customs forum and tweeked it occasionally, with an overhaul by donnyrides last year. It's been +/- 2; close enough since and has served me well over the last decade