Doors closed ₨6.29M / ₨10M generated · ₨4.00M courier-verified Watch the proof →
THE FLAGSHIP AGENT

Your ads have an
employee now.

Ad Pilot is an AI performance marketer that lives inside your ad account. It signs a weekly contract, watches performance hourly, 09:00–18:00 PKT, tells you what to kill and what to scale — and then gets graded on whether it was right. You don't need a team of media buyers. You need one person who can read. The evidence cockpit runs everywhere; the pilot itself flies Pakistan first.

01

The working week of a machine.

Monday
The contract
You set the week: orders target, budget, CAC ceiling. Ad Pilot is accountable to the contract — never to vibes.
Hourly
The watch
Every hour, 09:00–18:00 PKT, it pulls live Meta performance and checks reality against contract pace. It catches its own misses with a catch-up pass.
When it matters
The verdict
Opus-grade reasoning issues one of six calls — with the evidence attached. Silence is a feature: no news, no noise.
Always
The human decides
Done, Dismiss, or Discuss. Nothing touches your ad account — Ad Pilot is read-only by design. It opens the exact ad for you; it never edits it.
72h later
The grading
Deterministic code — not the model — scores the call on 3-day before/after windows, and writes down what ignoring it cost.
02

Six verdicts. No essays.

A performance marketer's whole job, compressed into six decisions. Each comes with the evidence; none comes with a meeting.

✕KILL
Bleeding spend with no path to the contract's CAC ceiling. Kill it today, before it eats another day of budget.
✂CUT
Trim the budget, keep the ad alive for data. Starve it, don't shoot it.
—HOLD
Nothing worth saying. Holds are free — they don't spend the call budget. The machine learns to speak only when it matters.
↑SCALE
The evidence says push. Budget up while the window is open — with the numbers that say why.
↻REFRESH CREATIVE
The ad works; the creative is tiring. Frequency rising, CTR decaying — brief the studio before the account pays for it.
＋LAUNCH NEEDED
A gap in the account: nothing testing this angle. The machine asks for new creative instead of waiting for someone to notice.
03

The model proposes. Code disposes.

Every AI demo looks magic until it spends your money wrong. So Ad Pilot is built the other way around: the AI suggests, deterministic code decides what it's allowed to say.

⛨
Hard gates after the model
The model proposes verdicts; TypeScript gates run after it speaks. No matter how persuasive the AI is, it cannot exceed the contract, the budget rules, or the evidence bar.
◔
A call budget, not a firehose
Max 8 calls per run — and HOLDs don't count. Attention is the scarcest resource in your company; the machine is rationed accordingly.
◈
Steering memory
Your guidance, decisions, preferences and context are injected into every run. Correct it once — it gets less generic every week you work with it.
⟲
Backtests before trust
Run the rulebook against your account's history before it ever judges your present. You see its track record before you grant it authority.
patience_days3A fresh call sits quiet for 3 days before the Pilot judges its outcome. No knee-jerk verdicts.
learning_grace_days3A new ad set gets 3 days of grace while Meta's learning phase wobbles. The Pilot does not punish a toddler.
no_thrash_hours24A decision cannot be reversed within 24 hours of making it. The machine is not allowed to thrash your budgets.
already_actioned_hours24Once a call is acted on, the same call can't be re-issued for 24 hours. No nagging.
max_step_pct30%No budget move larger than ±30% in one step — no matter how confident the model sounds.
zero_purchase_multiple3× ceilingSpend reaches 3× the CAC ceiling with zero purchases → the kill gate fires. Hope is not a strategy.
morning_cutoff_hour13:00 PKTVerdicts land before 1pm Pakistan time — decisions in the morning, execution all day.
min_purchases_for_scale5SCALE requires at least 5 real purchases. The Pilot does not scale on two lucky orders.
scale_requires_good_days3SCALE also requires 3 good days — one hot afternoon doesn't move your money.
scale_cooldown_hours48After a scale, 48 hours of silence before the next one. Escalation has brakes.
max_active_calls8At most 8 open calls at once — and HOLDs don't count. Attention is rationed.
protect_proven_winnertrueA proven winner cannot be killed, full stop. The model's worst day cannot hurt your best ad set.
evidence_floor_spendautoBelow a minimum spend, there is no verdict at all. The Pilot would rather say nothing than guess.

Anyone can claim "guardrails". We publish ours — every number above is a typed default the founder can edit and the model cannot. And you can replay the whole rulebook against your own history before trusting a single call: that's what the backtest is for.

04

An AI with a track record.

Every other AI tool asks for faith. Ad Pilot keeps a ledger.

⚖
Graded by code, not by itself
72+ hours after every verdict, deterministic grading compares 3-day windows before and after. The model never marks its own homework.
₨
The cost of ignoring it
Dismissed a KILL? Graymat writes cost_of_ignoring_pkr into the ledger — the exact rupees that decision cost. The most honest number in your marketing.
Σ
Verified impact, not estimates
Followed verdicts land in the Impact Ledger as verified rupees generated or saved — counted the same conservative way as everything else in Graymat: against delivered cash.
CAUGHT / IN THE ACT
It pauses itself.

The live cockpit, captured 05 Aug and redrawn in the site's own ink — every number transcribed from the raw capture. Three consecutive SCALE days tripped the daily guardrail, so Ad Pilot paused its own authority and said so on the dashboard, next to the loop it still runs hourly, 09:00–18:00 PKT. Discipline before magic, on the record.

Meet the other agents →
Ad Cockpitapp.graymat.ai
PAKISTAN AD PILOT CAPTURED 05 AUG 2026
Ad Pilot BETA — calibration watch loop ON off pace Ask your Pilot ▶ Run now

A senior AI performance marketer watching the PK Meta accounts hourly against this week’s contract. Read-only: it recommends, you press the buttons in Ads Manager.

OFF PACE 1,850 / 10,000 orders (19%) at 41% of the week. Blended CAC Rs 361 vs Rs 400 ceiling. Spend 21% of cap. why?
7 things need a human today · 4 watched, no action needed
SCALE +25% (Rs 13,500 → ~Rs 16,900/day) high expires in 48h first called 05 Aug, 21:56 PKT

The honest fine print

TodayMeta ads, Pakistan first — said up here, not just down here. UAE and BD get the full evidence cockpit; the pilot follows.
Read-only, alwaysNo Marketing API writes anywhere in the codebase. Deeplinks take your human to the exact ad — the click stays human.
Ships darkOff behind a feature flag, like everything in Graymat. Nothing wakes up until the founder flips it on.
No silent failureIf Meta data fails, a banner says so. A missing spend feed never gets to make your CAC look good.

It doesn't work alone.

Five more employees run recovery, retention, upsell, relationships and the safety net.