A below-band base salary is not a reason to walk. It is a reason to re-price the offer with evidence.

1. Benchmark the base, not the story.

Ask for the role level, location, and team. Compare that to current 2026 senior ML compensation bands. If base sits 8–15% under band, name the gap in dollars, not percentages. “This is $28k under the mid for Staff ML in this market” is a counter. “I was hoping for more” is not.

2. Separate equity vesting cliffs from total comp theater.

A four-year grant with a one-year cliff is not four years of value. Discount unvested equity, especially if the cliff has not started. If they cannot move base, ask whether they can front-load year-one vesting, shorten the cliff, or add a signing grant that vests on the same clock as the original.

3. Weight performance bonus last.

Bonus is the most reversible piece. Treat target bonus as expected value, not guaranteed cash. If they offer a higher bonus to cover a weak base, ask what percent of the team actually hit target last year. A 20% bonus that pays 60% of the time is not a 20% bump.

How to counter

• Restate the level and location they used.

• Put the band midpoint next to their base.

• Offer two paths: raise base into band, or keep base and add a cash signing plus equity that vests in year one.

• Give a date you will decide.

If you have an active offer in hand and want to run the exact math against current 2026 senior ML compensation bands, check out The Full AI/ML Hiring Toolkit here: https://whop.com/ml-offer-toolkit/the-full-ai-ml-hiring-toolkit-guide-questions-comps

Signing bonuses look simple. The clawback is where most AI/ML offers get expensive.

Use this as a 10-minute screen before you accept.

1. Write the cash you actually keep

Let:

• B = gross signing bonus

• t = combined marginal tax rate (federal + state + FICA/Medicare). If you do not know it, use 0.40 as a default for US senior IC offers.

• B_net = B × (1 − t)

Example: $50,000 bonus at t = 0.40 → $30,000 net.

2. Convert the clawback into a daily cost

Typical clause: repay a pro-rata (or worse, full) amount if you leave within T days (often 365 or 730).

• Pro-rata daily clawback: c = B / T (sometimes they claw gross, not net)

• Full clawback: c_full = B / days_you_actually_stayed

Net daily value while you stay: v = B_net / T

If they claw gross while you only received net, your true loss on early exit is:

loss = B − (B_net × days_stayed / T) (pro-rata, gross clawback)

That gap is the tax trap. Always ask: “Is repayment gross or net of taxes already withheld?”

3. Benchmarks for AI/ML engineers (2026 screen)

Use these as order-of-magnitude checks, not guarantees:

Red flags

• Window > 24 months

• Full (not pro-rata) repayment after month 3

• Clawback on gross with no tax true-up

• Bonus paid in month 6+ but clock starts on day 1

• Tied to performance or immigration status, not just tenure

Green flags

• Pro-rata by calendar day

• Net-of-tax or employer gross-up on clawback

• Paid in first 30 days

• Window ≤ 12 months

4. Compare bonus vs. raising base

A signing bonus is a loan with a job attached. Annualize it:

bonus_as_raise = B_net / (T / 365)

Example: $50k gross, t=0.40, T=365 → $30k for one year of staying. That is a ~$30k one-year raise, not a $50k raise.

If you can trade $15k of bonus for $10k more base, the base usually wins after year 1 (raises, 401k match, and next offer all stack on base).

Phrase it:

“I can take $X signing with a 12-month pro-rata clawback, or $Y more base with no clawback. On a 24-month stay, base is worth $Z more after tax. Can we move $X into base?”

5. Decision rule (write this on the offer)

Stay probability in the window: p

Expected net bonus: E = p × B_net − (1−p) × expected_clawback

If E is small versus the stress of being trapped, ask to shorten T or switch to pro-rata.

Worked example

• B = $40,000, t = 0.40, B_net = $24,000

• T = 365, pro-rata gross clawback

• You leave at day 180

• Clawback owed ≈ $40,000 × 185/365 ≈ $20,274

• Cash you kept after tax and clawback ≈ $24,000 − $20,274 = $3,726 for half a year of lock-in

That is a bad trade. Negotiate net clawback or a 90-day cliff then pro-rata.

6. Phrases that work in AI/ML loops

• “Please confirm clawback is pro-rata by day and calculated on net proceeds.”

• “Please start the clawback clock on bonus pay date, not start date.”

• “I will trade $A of bonus for $B of base so we both avoid a 12-month trap.”

• “If the role is high-churn (new org, new manager), I need a 6-month window.”

---

Full offer worksheets, equity vs cash models, and negotiation scripts are in the Complete Suite of the Full AI/ML Hiring Toolkit:

https://whop.com/checkout/ch_NnXDeoHrgzJa61C/

How to spot a lowball base on the first recruiter call (without sounding like a jerk)

Recruiters often float a range before they know your comps. Your job is not to “win” the call. It is to figure out whether the band is real, whether it is below market, and whether you should keep talking.

1. Get the band in writing-adjacent language

Ask for the approved base range for this level, not “what do you think we can do.”

Try:

• “What’s the approved base range for this level, not the stretch?”

• “Is that range for L5 / Staff / Senior, and is it cash only?”

If they only give total comp, split it: “Can you break that into base vs bonus vs equity at the current 409A / last round?”

2. Anchor against a band, not a feeling

Before you react, map their number to:

• level (IC vs staff vs manager)

• location / remote policy

• cash vs RSUs vs options

• refresh cadence (year 1 vs year 2+)

A “$180k base” for Senior ML in a high-cost hub with no refresh is a different product than $180k + strong RSU.

3. Three tells it is a lowball

• Range is wider than ~$20–30k on base (“$140–200k”). That usually means they want you to self-select low.

• They ask your current / last number first, then “match plus a bit.”

• Equity is quoted as a huge upside story while base is sticky and bonus is “target, not guaranteed.”

4. Counter without picking a fight

You are clarifying fit, not accusing them.

If the band is low:

“Thanks for sharing the approved range. For this level and location, I’m seeing cash closer to $X–$Y. If the role is locked at $Z base, I don’t want to waste your hiring manager’s time. Is there flexibility on level or cash, or should we pause?”

If they press for your number first:

“Happy to share a range after I understand level and the approved band. I’m looking at $X–$Y base for this scope. If that’s outside the box, better to know now.”

If they say exploding / “this is the best we can do” on call one:

“I can move fast once the cash band is in market. I’m not going to commit on a first-pass number. If you can send the band and level in writing, I’ll get you a clear yes/no.”

5. What to write down after the call

• quoted base min/max

• whether it was “approved” vs “I think”

• level they used

• what they asked you first (current pay vs expectations)

• any “we’ll make it up in equity” line

If two of those smell off, treat it as a lowball until a written band says otherwise.

---

If you want the full spreadsheets, scripts, and comps for AI/ML offers, grab The Full AI/ML Hiring Toolkit here: https://whop.com/ml-offer-toolkit/the-full-ai-ml-hiring-toolkit-guide-questions-comps

Use code careerboost for 40% off.

A candidate I coached last month got a $185k offer from a Series B, plus a $40k sign-on, and a competing $210k cash package from a Big Tech team. They went back to the startup with “I need $210k base or I’m out.” The startup walked.

That is the most common counter I see from AI/ML engineers: treating the other offer as a hostage note instead of a negotiation.

Mistake 1: leading with the competing number and nothing else. Comp committees already know market ranges. What they do not have is a reason to stretch for you. Frame the counter around the work you will ship in the first two quarters, the models you have already shipped, and the hiring bar they just put you through.

Mistake 2: ignoring total compensation shape. Startups often cannot match Big Tech cash. They can move on sign-on, refresh, leveling, start date, and remote policy. If you only fight on base, you leave the only levers they actually control unused.

Mistake 3: countering before you understand the band. Ask where you sit in the level and what the next level looks like. A $10k base bump at L4 can be weaker than a clean L5 packet with the same cash.

Mistake 4: giving a deadline you cannot keep. “I need an answer by Friday” only works if you will actually walk. Fake urgency burns trust and often kills the deal.

Mistake 5: never writing the counter down. Verbal “can you do better” is easy to ignore. A short written recap of the offer, the gap, and 2–3 specific asks is what actually gets a rec committee to reopen the packet.

If you have a live offer, drop the structure (not the company name if you prefer) and we can tear the counter apart in the thread.

For complete equity spreadsheets, 12 senior interview model answers, and full negotiation scripts, check out the Full AI/ML Hiring Toolkit ($12.99): https://whop.com/ml-offer-toolkit/the-full-ai-ml-hiring-toolkit-guide-questions-comps