A pool can be plentiful and still unwinnable if your offer sits two tiers below market. Comp is where feasibility quietly dies — so you read it honestly, in total-comp terms, before anyone makes a promise.
ARead total comp, not base
The single most common comp mistake is comparing base salaries. In semiconductor and AI, the package is the number that matters — and equity is often the biggest piece.
Base
The floor, and the least interesting part. Easy to benchmark, but on its own it badly understates what top talent actually earns.
Bonus
Target + actual differ. A 15% target paid at 0.7x is a very different number. Always ask which you're looking at.
Equity
Often the largest component for senior IC and AI roles. Public RSUs and private options are not comparable — value and risk differ enormously.
Sign-on
The hidden lever. Used to buy out forfeited equity or close a gap. Frequently the difference-maker in a competitive offer.
The rule
Benchmark in total target compensation — base + target bonus + annualized equity. Comparing anything less makes a competitive offer look generous and a weak one look fine.
BWhere to pull comp data
No single source is gospel. Triangulate — crowd-sourced for reality, survey for rigor, postings for what's being offered right now.
Best for: a macro floor by occupation and metro. Lags the market but anchors the low end.
CNormalize before you compare
Raw numbers from three sources aren't comparable until you put them on the same footing. Four adjustments do it.
1
Level the levels
Map every source to one consistent ladder. A company's "Senior" may be another's "Staff." Anchor to a known scale before comparing anything.
ExampleNvidia Senior (IC4-ish), Apple ICT4, and a startup "Staff Engineer" can all be the same market level — or three different ones. Pin it first.
2
Adjust for geography
Comp is metro-specific. Bay Area sets the ceiling; Austin, San Diego, and Phoenix run lower. Normalize to your target location, not a national average.
ExampleThe same analog design level can swing 25-35% between the Bay and a lower-cost semi hub. A national figure misleads in both directions.
3
Make equity comparable
Annualize equity over the vest, and never equate public RSUs with private options. Note the assumptions — a private 409A is a guess, not a price.
ExampleA $400K RSU grant over 4 years = $100K/yr of liquid value. A private option grant of "equivalent" face value is worth far less in hand.
4
Pick a percentile, on purpose
Decide where you want to compete — median (P50) holds the line; P75 wins scarce talent. State it explicitly so the benchmark has a target, not just a range.
ExampleFor a tight RF/analog pool, benchmarking to P50 quietly guarantees you lose. Name P75 as the target and price against it.
DUsing AI to build the benchmark
AI triangulates sources, levels the ladders, and converts a messy pile of numbers into a clean total-comp range with a stated percentile.
The hard data. Source the actual numbers here; let AI interpret and assemble, not invent.
Prompt 1 · Total-comp range from sources
Build a total-comp benchmark for [role] at the [level] level
in [metro].
- Pull recent data points from Levels.fyi, Glassdoor, and any live
posting ranges you can find; cite each
- Express everything as total target comp: base + target bonus +
annualized equity
- Give me P25 / P50 / P75
- Flag where public-RSU vs. private-equity assumptions affect the number
Note your confidence and where the data is thin.
Prompt 2 · Level the ladders
I have comp data from different companies using different level names
for [role]: [paste the titles + numbers].
Map them all to one consistent market level (junior / mid / senior /
staff / principal), explain how you matched each, and flag any that are
ambiguous so I can verify before trusting the comparison.
Prompt 3 · Benchmark vs. the req
The req budgets [band] for [role] in [metro].
Market total comp at this level runs [range].
Tell me, in plain terms a hiring manager will get:
1. Where the band sits vs. market (which percentile)
2. What that means for who we can realistically attract
3. The smallest change that would make this competitive
(lift base, add sign-on, flex location?)
EGuardrails
Before you quote a number
!Never compare base to base. Total target comp or nothing. Base-only benchmarks have lost more competitive hires than bad sourcing ever has.
!Crowd-sourced skews high and senior. Self-reported data over-represents top earners and big tech. Anchor it against a survey source where stakes are high.
!Private equity is an assumption, not a price. A 409A valuation is a snapshot, not liquidity. Say so out loud when equity is in the mix.
!You are not giving comp advice. You're reporting the market. Final banding belongs to the comp team — present the benchmark, not a recommendation to pay X.